Search

BACKGROUND: Previous studies have shown the correlation between lumbosacral sagittal plane parameters and natural absorption of lumbar disc herniation. However, the lumbosacral sagittal plane parameters included lumbar lordosis angle, lumbosacral joint angle, sacral inclination angle and many other parameters. The effects of each parameter on the natural absorption of the herniated disc were different. In addition, there are few studies on the reabsorption of a specific segment of intervertebral disc herniation at present, and most of the measured data are obtained from digital radiography or CT, while the correlation between lumbosacral sagittal plane parameters measured from MRI and reabsorption after L5/S1 intervertebral disc herniation is rarely reported. OBJECTIVE: To study the corresponding changes of lumbar sagittal plane parameters after L5/S1 intervertebral disc herniation reabsorption and to screen out the lumbosacral sagittal plane parameters with the most significant changes during intervertebral disc reabsorption. METHODS: Totally 57 patients with lumbar disc herniation who had complete MRI image data were selected and met the diagnostic criteria for lumbar disc herniation and only received non-surgical treatment for reabsorption of L5/S1 protrusion segments. MRI measured the protrusion area of the maximum protrusion plane in the coronal plane, lumbosacral sagittal plane parameters [lumbar curvature index, lumbar lordosis (α), L5/S1 disc angle (β), intervertebral height measurement, lumbosacral joint angle, sacral platform angle, sacral inclination angle, and lower lumbar lordosis angle]. Besides, lumbosacral sagittal plane parameters were ranked in the importance of variables by random forest model in R software, and then significant variables were fitted with multiple linear regression. The changes between parameters before and after treatment were analyzed and compared by paired sample t-test. RESULTS AND CONCLUSION: (1) A total of 57 patients with L5/S1 lumbar disc herniation were included in this study, and the symptoms and imaging features of the patients were significantly relieved to a large extent. (2) Before treatment, there were 4 cases of grade 1, 29 cases of grade 2 and 24 cases of grade 3 according to the Classification of Michigan State University. After treatment, there were 48 cases of grade 1 and 9 cases of grade 2. (3) The random forest model suggested that intervertebral height, lumbar curve index, sacral inclination angle, and lower lumbar lordosis angle changed significantly in L5/S1 disc herniation reabsorption, and the order of their change significance was lumbar curve index > intervertebral space height > sacral inclination angle > lower lumbar lordosis angle. (4) Lumbar curve index, lumbar lordosis and sacral platform angle increased, with statistical significance (P < 0.05). There were no significant differences in disc angle, intervertebral height, lower lumbar lordosis angle, sacral inclination angle or lumbosacral joint angle (P > 0.05). (5) Lumbar curvature index was the most significant parameter of the lumbosacral sagittal plane in herniated disc reabsorption. In addition, lumbar curve index, sacral inclination angle, and lower lumbar lordosis angle are commonly used clinically to describe the change of lumbar curvature, suggesting that L5/S1 disc herniation reabsorption is correlated with the change of lumbar curvature. It is indicated that in the treatment of lumbar disc herniation, a clinical cure can be achieved by improving or restoring the disordered lumbar curvature. [ABSTRACT FROM AUTHOR]

Objective BackgroundMedical staff satisfaction is an important indicator for evaluating the reform of medical communities, which is crucial for the healthy development of medical communities. Objective To investigate the current situation of medical staff satisfaction from 20 primary medical institutions in the closely 一 knit county medical community in Xinyi City, Jiangsu Province, so as to provide reference for improving the construction of county medical communities. Methods A Stratified random sampling method was used to conduct a face - to - face questionnaire survey of medical staff within the closely - knit county medical community in Xinyi City, Jiangsu Province. A random forest model was used to rank the importance of independent variables and to filter the variables, and the filtered variables were included in a logistic regression for a multifactor analysis of satisfaction. Results The overall satisfaction score of medical staff in Xinyi city county medical community is 3. 86 ± 0. 76. The random forest model showed that the error rate of out - of 一 pocket estimation is the lowest when the number of variables is 7, and it is considered in turn whether the reform of medical insurance payment mode is helpful to control medical expenses, length of service, working hours per week, awareness of medical insurance payment mode reform, professional and technical post level, awareness of medical community, and whether the reform of medical insurance payment mode affects diagnosis and treatment behavior. The logistic regression showed that the reform of the health insurance payment mode was very helpful in controlling the unreasonable increase of medical expenses (OR = 8. 153,95% CI: 3. 404 一 19. 531, P <0. 001), where the higher the knowledge of the reform of the health insurance payment mode of the medical community (OR 碍血沁「強=1-986, 95% CI: 1. 190-3.314, P 二 0. 009; ORii in「o讪厂 2. 665, 95% CI: 1. 409 - 5. 038, P = 0. 003), and the longer the weekly working hours (OR40 一劲/叶应=0. 575, 95% CI: 0. 364 一 0. 910, P =0. 018; OR-8()h/week = 0. 286, 95% CI: 0. 167 一 0. 489, P <0. 001), the more satisfied the medical staff were. Conclusion The overall satisfaction of medical staff in the medical communities in Xinyi city County is not high, and their cognition and evaluation of the reform of medical insurance payment methods and working hours affect their satisfaction. It is necessary to comprehensively consider the actual needs of medical staff in the medical community, improve the level of medical community construction in the county, and improve the job satisfaction of medical staff. [ABSTRACT FROM AUTHOR]

BACKGROUND: The incidence of hip fracture patients is increasing day by day. Because of their poor physical condition, these people often need to stay in hospital for a long time. However, the prolonged length of stay leads to a decrease in bed circulation rate and an increase in economic burden. At present, there are few prediction models for delayed discharge of hip fractures. This study aims to find the best model for delayed discharge of hip fractures and guide clinical decision-making. OBJECTIVE: To explore the risk factors of prolonged length of stay in patients with hip fractures, establish two different risk prediction models, obtain the best risk prediction tools, and provide a reference for clinical intervention and management METHODS: Data from 683 elderly patients with hip fractures in Tianjin Hospital from January 2019 to December 2021 were retrospectively analyzed. All patients were randomly divided into a modeling group (479 cases) and a verification group (204 cases) according to the ratio of 7:3. The 75th percentile of length of stay was taken as the cut-off point (> 28 days), and they were divided into extended hospitalization group and normal group. Single-factor and multifactor Logistic regression analysis and variable importance ranking were used to screen the best prediction model; the nomogram and random forest models were constructed. The prediction efficiency of the two models was evaluated by the receiver operating characteristic curve area, accuracy, sensitivity, specificity, positive prediction value and negative prediction value. RESULTS AND CONCLUSION: (1) Logistic regression analysis showed that bone traction, pneumonia, refolding, multiple trauma, venous thrombosis, lung infection and age-adjusted Charlson Comorbidity Index were the risk factors of prolonged length of stay of hip fracture patients. (2) The random forest model showed that age, bone traction, surgical type, age-adjusted Charlson Comorbidity Index and pneumonia were the first five indexes according to the average reduction of the Gini index, which had an important influence on the prediction of delayed discharge. (3) The area under the receiver operating characteristic curve, accuracy, sensitivity, specificity, positive prediction value and negative prediction value of the Logistic regression model and random forest prediction model were 0.774(95%CI: 0.696-0.853) and 0.708(95%CI: 0.627-0.789), 60.78% and 90.85%, 80.39% and 23.53%, 50.82% and 78.09%, 86.01% and 46.15%, respectively. The results exhibited that the Logistic model had good prediction efficiency. (4) Above findings confirm that the Logistics regression model and random forest model have high predictive value for a prolonged length of stay in patients with hip fractures, which is of great significance for clinical medical staff to identify high-risk patients in time and take effective intervention measures to reduce the length of stay of patients with hip fractures. [ABSTRACT FROM AUTHOR]

Coix seeds from 9 different origins were taken as the research object. An attempt was made to achieve geographical origin traceability of Coix seeds through the combination of excitation-emission matrix (EEM) fluorescence spectroscopy with improved random forest algorithm. The improvements to the random forest algorithm mainly include two aspects, firstly, principal component analysis (PCA) was adopted to reduce the dimension of EEM fluorescent data; secondly, a grid search method was used to identify the optimal number of principal components(PCs) to retain and the hyperparameters of the discriminant model during the PCA dimension reduction process. The results showed that an improved random forest model, incorporating standard deviation normalization and PCA dimension reduction modules, based on Coix seeds EEM fluorescence spectroscopy data, accurately predicted the geographical origin of Coix seed samples from 9 different areas. The optimal model was constructed by combining 100 decision trees with a maximum depth of 3 and a minimum sample size of 1 at the leaf node, using 16 principal components (PCs). This model achieved 100% prediction accuracy for both the validation and test sets, which consisted of a total of 108 samples, outperforming the PLS-DA model constructed by the partial least squares method (96% prediction accuracy). [ABSTRACT FROM AUTHOR]

Objective To investigate the influencing factors of lipoproteins and metabolites in Alzheimer's disease based on machine learning algorithm. Methods A total of 314 subjects diagnosed with cognitive normal (CN) and Alzheimer's disease (AD) in 2012 were selected from the ADNI database, and their lipoprotein and metabolite data were collected. Random forest, Lasso regression, and XG boost algorithm were used to sort and screen the important variables. Using the variables selected by the three methods and combining the gender, age, and marital status of the study population, a random forest model was constructed to predict the important factors affecting the incidence of AD. Results A total of 12 lipoprotein and metabolite variables were screened by three methods, and 15 variables were incorporated into the random forest model combined with the age, sex, and marital status of the study population. The accuracy of the model was 84.13%, the sensitivity was 93.75%, the specificity was 53.33%, the Kappa value was 0.5183, and the AUC (95%CI) was 0.735 (0.600-0.871). The most important variables selected in the random forest model based on Mean Decrease Accuracy and Mean Decrease Gini were phospholipid to total lipid ratio in large very low-density lipoprotein (L_VLDL_PL_PCT), age, triglycerides in cytokinome's and very low-density lipoprotein (XXL_VLDL_TG), and cholesterol to total lipid ratio in large very low-density lipoprotein (L_VLDL_C_PCT). The Mean Decrease Accuracy of the four variables were 6.68, 6.65, 6.10, and 5.49, respectively, and the Mean Decrease Gini values were 7.35, 11.71, 8.77, and 10.08, respectively. Conclusion L_VLDL_PL_PCT, AGE, XXL_VLDL_TG, and L_VLDL_C_PCT are closely related to the incidence of AD. This study can provide new insights for the diagnosis and prevention of AD. In-depth study of the relationship between lipoprotein levels and AD incidence can prevent and delay the occurrence of AD. [ABSTRACT FROM AUTHOR]

Drought is one of the major meteorological disasters in the world. It is urgent to accurately assess the frequency of drought events and the risk of drought disasters in the region, particularly for the warning, prevention and mitigation of regional damage. In this study, the MCI (meteorological drought composite index) was calculated using the daily observation data from the 106 meteorological stations in Xinjiang in western China during 1961-2019. The drought process was identified using the runs theory. The joint distribution of duration and intensity was also established to calculate the return period using the GumbelCopula function. Eventually, the hazard factors were taken as the frequency duration and intensity of drought events. The weights were calculated to assess the hazards of drought using the random forest. The results show that there were 47 drought events in the study area during 1961-2019. Among them, drought events with short duration and low intensity (intensity ≤ 50 and duration ≤ 60d) were frequent, whereas, the extreme drought events occurred less frequently with long duration and high intensity (intensity ≥200 and duration ≥180 d) tend to associtate with longer return periods. Long duration and high intensity tended to occur simultaneously. However, the probability of occurrence decreased with the increasing drought duration and intensity. The drought return period was longer along the Tianshan mountain range than that of other regions under the same duration and intensity conditions. The spatial distribution of drought hazards was performed as the middle low, and the high north-south ends. The hazard was much lower along the Tianshan Mountains. The drought events occurred less frequently on the northern slope, especially with the short duration and the low intensity. The higher hazards can be observed in most parts of the southern region. A similar trend was found in the northwestern part of the Tarbagatay Prefecture, as well as the central and northern of the Altay Prefecture in the northern region. There were different characteristics of drought, although the hazards were high in both the northern and southern regions. The drought events occurred less frequently, but lasted for a long duration with the high intensity in most stations in the southern region. However, the high frequency of droughts was found in the northern region, but for a short duration and low intensity. The northern and southern regions should respond to the drought disasters in different plans. An efficient drought early warning and monitoring system can consider the drought occurrences and development trends in the local conditions. Efficient allocation and utilization of water resources can then be balanced to reduce the risk of drought disasters. [ABSTRACT FROM AUTHOR]

A great challenge has been found to limit the per capita cultivated land and the scarcity of land resources in China. However, the cultivated land can be abandoned across the country with the rapid growth of industrialization and urbanization. It is crucial to accurately and rapidly monitor the abandoned farmland with high accuracy for farmland protection. Land use can also be evaluated to guarantee national food security for the decision-making on modern agriculture. Among them, there is fragmented farmland with a wide distribution in the Loess Plateau of China. The complex terrain has made some challenge to investigate and monitor the abandoned farmland. This study aims to monitor the abandoned farmland on the loess plateau using feature optimization. Data sources were collected from the Chinese satellite images with high spatial resolution and multispectrum. Some feature factors were selected to effectively distinguish the cultivated land from other land types. The prior knowledge was combined with the distinct separability, including the spectral, phenological, topographic and texture features. The spectral feature differences were statistically analyzed from May to October. Spectral time series feature factors were developed to prove the effectiveness and superiority of the schemes. Five types of experimental scheme were also constructed for comparison. The binary particle swarm optimization (BPSO) coupled with the genetic algorithm (GA) was selected to modify the index factor for the optimal selection of feature factors in classification. Random forest (RF) classification and detection were performed post-classification. The spatiotemporal distribution was obtained for the abandoned farmland at field scale between 2011 and 2022. The results showed that the feature optimization improved the accuracy and efficiency in the classification of remotely sensed images with the high spatial resolution. According to an importance assessment for all features, NDVI was found to have the highest MDA score. The final preferred feature was ranked in the order of the green band, red band, near-infrared band, (b1-b2)/(b1+b2), Slope, NDVI, Variance and Contrast. The combination of GA and BPSO exhibited high efficiency in the global search and search speed. The Optimum Index Factor effectively evaluated the overall combination. Feature optimization can be expected to avoid information redundancy and over-fitting issues. The field validation was performed using image interpretation and field investigation. It infers that the abandoned farmland at field scale was recognized from the high spatial resolution images of Chinese satellites with an accuracy as high as 92.48%. There was a 96% overlapping ratio between patches from field investigation and those extracted from images. Among them, the recognition rate of fragmented patches with areas as small as 200-400 m² reached 90.11%. The classification showed that there was abandoned farmland with a total area of 5 492.51 hm² between 2011 and 2022, indicating the maximum abandoned percentage (21.09%) in 2011-2014 and the minimum abandoned percentage (0.99%) in 2020-2022. The large-scale pattern of abandonment of farmland was gradually shifted into a smaller fragmented one by peasant households over this period. As such, the abandoned farmland was monitored at a field scale on the loess plateau region using remote sensing. The finding can provide a strong reference to explore the abandoned farmland using high spatial resolution images of Chinese satellites. [ABSTRACT FROM AUTHOR]

Remote sensing technology combined with sample plot data allows for robust and efficient estimation of growing stock volume, which greatly reduces survey time and costs and facilitates regular monitoring. Taking Wangyedian Forest Farm in Chifeng City, Inner Mongolia Autonomous Region as the study area and a typical temperate coniferous plantation forest as the research object, the forest stock volume of the plantation forest in the whole study area was estimated by the two-stage extrapolation method using the random forest algorithm in combination with the measured sample data, the UAV-LiDAR data and the Sentinel-2 multispectral data. The measured sample plot data were extrapolated to the UAV-LiDAR sample plots to build the Field-UAV-LiDAR model, and the UAVLiDAR estimated forest stock volume generated in the first stage was extrapolated to the full-coverage Sentinel- 2 satellite image to build the Field-UAV-LiDAR-S-2 model. The direct prediction model (Field -S-2) was compared by combining with the measured sample plot data and the feature variables extracted from Sentinel-2. The results showed that the highest accuracy of growing stock volume estimation (R ² = 0. 79, RMSE= 42. 34 m 3 / hm ², rRMSE= 19. 24%) was based on the Field-UAV-LiDAR model, followed by the accuracy of growing stock volume estimation based on the Field-UAV-LiDAR-S- 2 model ( R ² = 0. 68, RMSE = 50. 19 m 3 / hm ², rRMSE= 29. 43%), and the lowest accuracy of growing stock volume estimation based on Field-S- 2 model (R ² = 0. 49, RMSE = 63. 25 m 3 / hm ², rRMSE= 33. 08%). The results of the study verified the validity of combining UAV-LiDAR and Sentinel-2 data to estimate forest stock volume, which provides a reference idea for estimating forest stock volume in large areas. [ABSTRACT FROM AUTHOR]

Northeast China (NEC) has been the major soybean-producing region in China. Hence, it is very necessary to explore the spatial heterogeneity of soybean yield per unit in the NEC, in order to fully meet the current import and export production and demand. In this study, a multi-feature random forest (RF)-based classification was used to extract the spatial pattern of soybeans in 2022 using the Google Earth Engine (GEE) platform. The time series leaf area index (LAI) data was also combined with the field-measured yield. A soybean yield estimation model was established to characterize the spatial heterogeneity of regional soybean yield per unit. A geographic detector model was used to quantitatively explore the influencing factors. The results show that: 1) The overall accuracy of the soybean planting area reached 89.48% after extraction, with the Kappa coefficient of 0.89, and the coefficient of determination R 2 was 0.92 between the soybean planting areas extracted from remote sensing and the statistical data of prefecture-level city. There was a marked spatial decrease in the planting area of soybeans from the northern to the southern NEC. The soybean planting areas were concentrated mainly in the Songnen Plain. Suihua City was found in the center of gravity for the soybean planting areas in the NEC. 2) The average soybean yield per unit was 2 514.08 kg/hm2 in the NEC. The coefficient of determination R 2 was 0.72, compared with the actual measured yield per unit. There was a significantly clustered spatial distribution of soybean yield per unit in the NEC. The areas with the high values were located mainly in the northern part of the NEC, whereas, the areas with the low values were in the southern. 3) Three dominant independent factors with the most pronounced spatial heterogeneity of soybean yield per unit were ranked in the descending order of the soil type, soil pH, and soybean subsidies, with q values of 0.27, 0.24, and 0.24, respectively. The three most significant interaction factors were to explain the spatial heterogeneity in the soybean yield per unit, including the interaction between mean annual rainfall and mean annual cumulative temperature, the interaction between mean annual rainfall and soybean subsidies, and the interaction between soil type and soybean subsidies, with q values of 0.44, 0.40 and 0.40, respectively. Six anthropogenic factors presented the significant impacts on the spatial heterogeneity of soybean yield per unit, namely soybean subsidies, soybean prices, agricultural irrigation area, total power of agricultural machinery, fertilizer prices, and illiteracy rate. Their optimal impact ranges varied significantly, where the optimal impact ranges were from 4 801 to 7 500 yuan/hm2, from 5 601 to 5 800 yuan/t, from 13.6×104 to 26.4×104 hm2, from 252×104 to 436×104 kW, from 2 500 to 2 602 yuan/t and from 1.4% to 1.8%, respectively. There was a significant spatial heterogeneity of soybean yield per unit in the NEC, with an overall decreasing trend from the north to the south. This variation trend can be primarily driven by natural factors also subjected to human intervention. [ABSTRACT FROM AUTHOR]

Objective To analysis the influencing factors of perimenopausal syndrome in women, and providing a reference for perimenopausal syndrome intervention. Methods This was a cross-sectional Internet survey study. Using a snowball sampling method, 468 women aged 45 to 60 years were recruited from Shaanxi province. The random forest method was used to sort the importance and reduce the dimension of the independent variables. The Ordinal logistic regression model was used to analyze the action direction and relative risk (OR value) of independent variables. Results The results of random forest analysis showed that the OOB estimated error rate was the lowest when the number of variables was five, followed by the physical activity level, age, menopausal status, perceived health status and parity. Ordinal logistic regression analysis showed that physical activity level and perceived health status were protective factors for perimenopausal syndrome (OR < 1), menopausal status and parity were risk factors for perimenopausal syndrome (OR > 1). Conclusion The four influencing factors of perimenopausal syndrome were the physical activity level, perceived health status, menopausal status and parity. It is recommended to increase the education of healthy, and strengthen the protective factors and reduce the risk factors to improve perimenopausal women' s quality of life. [ABSTRACT FROM AUTHOR]

Monitoring the maturity of multi-species maize based on remote sensing and thus mastering the optimal harvesting time is crucial for improving its yield and quality. The traditional method to monitor the maturity progress of maize is to use field surveys, and the disappearance of the kernel "milkline" is usually taken as a sign of maturity. However, the traditional field survey method is a labor-intensive activity that is not conducive to high-throughput field monitoring. Therefore, this study aims to construct a maize maturity index (MMI) to quantify the maturity of maize and monitor it through UAV multispectral monitoring, so as to grasp the dynamics of maize maturity stage in the field. Firstly, the UAV platform was used to acquire multispectral images at five time points of the maize maturity stage, and ground-based measured data such as the percentage of milkline, kernel water content and leaf chlorophyll content were collected accordingly. Secondly, based on the weighted analysis of the measured data, the MMI was constructed. Finally, based on the MMI and the vegetation index, a model was constructed using regression models and random forests to realize the UAV multispectral monitoring of corn maturity, and the effects of different varieties on MMI were analyzed. The results showed that: 1) for different varieties of maize at maturity stage, there were differences in the change patterns of leaf chlorophyll content and kernel water content, the leaf chlorophyll content and kernel water content of Zhengdan 958 and Jingjiuqingzhu16 were always higher than that of Jiyuan 1 and Jiyuan 168, while the rate of decline of leaf chlorophyll content and milkline percentage of two varieties of maize at maturity stage was lower than that of Jiyuan 1 and Jiyuan 168. 2) The correlations between MMI and selected vegetation indices in the experiment could reach 0.01 significant level, among which the correlations with normalized difference vegetation index (NDVI) and transformed chlorophyll absorbtion ratio index (TCARI) were highest with correlation coefficients above 0.87, in addition, the wide dynamic range vegetation index (WDRVI) has the most obvious changes, and the variance fluctuates less, which is similar to MMI. 3) The study was verified based on data sets of different combinations. Among them, the random forest model has the highest estimation accuracy of MMI. The test set coefficient of determination (R² ) is 0.84, and the root mean squared error (RMSE) is 8.77%, and the normalized root mean squared error (nRMSE) is 12.05%. In the revalidation scheme 2-1, the RF model test set has high accuracy, in which R² is 0.65, RMSE is 13.02%, nRMSE is 19.17%. In addition, the random forest model has better estimated accuracy of different varieties of MMI. The Jingjiuqingzhu 16 has the best accuracy. Among them, R², RMSE, and nRMSE are 0.76, 10.67%, and 15.88%. The model accuracy proves that the drone platform can be monitored to monitor the maturity of different varieties of corn. The results of the research can provide a reference for the dynamic changes in multi-spectrum drones to monitor the dynamic changes in multi-variety of corn in farmland. [ABSTRACT FROM AUTHOR]

Rice shrimp co-cultivation, as a circular agriculture model that integrates ecological, economic, and social benefits, has been widely applied in China. Accurately obtaining information on the planting area, spatial distribution, and changes of rice shrimp co-cultivation is of great significance for ensuring safe production, social supply of rice and shrimp, and formulating corresponding management and planning policies by government departments. Most of the existing methods for extracting rice shrimp farming fields used medium-resolution temporal images as data sources to analyze the characteristics of water or vegetation changes within a year to construct models. The accuracy of the extraction depended on the results of water or vegetation analysis, the boundaries were unclear and there was a lot of noise. To solve the problems of complex temporal analysis, low accuracy of extraction results, and incomplete boundaries in existing extraction methods, this paper used high-resolution images of a single temporal phase as the data source, without analyzing the temporal features of rice shrimp farming fields. Only the spatial features of rice shrimp farming fields on high-resolution images were used for high-precision extraction. We proposed a deep learning semantic segmentation model named EASNet (edge assisted segmentation network), which was mainly composed of three modules: feature extraction module (FE Module), edge assist module (EA Module), and information fusion Module (IF Module). The FE Module was used to extract multi-level and fused contextual semantic information features of the target object. The EA module used the unique edge "shrimp groove" of rice and shrimp farming fields as an auxiliary information for separate segmentation in the designed EA Module. The IF Module integrated the outputs of the FE Module and the EA Module, enabling the main task to not only enhanced the boundary structure information of rice shrimp farming fields but also learned the unique spatial and semantic information of rice shrimp farming fields. The experimental results showed that with the enhancement of the EA Module, the segmentation results of rice shrimp farming fields were more complete and the boundaries were clearer. The IoU (Intersection over Union) of semantic accuracy and the F1-Score of boundary accuracy were improved by 1.5% and 5.8%, respectively. The recall, IoU, and F1-Score of overall semantic accuracy reached 0.970, 0.964, and 0.930, respectively. The Recall and F1-Score of boundary accuracy reached 0.864 and 0.859, while the Recall and F1-Score of relaxed boundary accuracy reached 0.876 and 0.913. The trained EASNet model was applied to the whole area of Xuyi County, and the spatial distribution map of rice shrimp farming fields in Xuyi County in 2020 was obtained. In comparison with the results of rice shrimp breeding fields extracted by the traditional Water Seasonal Difference method and the Random Forest method, our method obtained the optimal results with an OA(overall accuracy) of 96.71% and a Kappa coefficient of 0.934. The method used in this paper has higher accuracy, lower missing rate, more regular boundaries, and fewer instances of similar integration in extracting results. It can provide a basis for natural resource surveys, and government departments to formulate corresponding rice shrimp breeding management and planning policies. [ABSTRACT FROM AUTHOR]

Real-time and accurate assessment of crop growth and yield is of great significance for grain macro-control measures, crop insurance, and agricultural production planning. However, it is lacking in the comparability for the crops in different growth stages at the same period during traditional monitoring crop growth using normalized difference vegetation index (NDVI). The high/low NDVI value cannot directly represent the high/low yield. In this study, a spatiotemporal comparable statistical index (crop yield index) was designed to dynamically represent the changes in crop growth and yield. The yield, meteorological, remote sensing and crop distribution data were first collected from the Heilongjiang Province of China. The yield estimation zoning was carried out for three crops (maize, rice, and soybeans) by spatial "K" luster analysis by tree edge removal (SKATER). The optimal variables of yield estimation were screened for each yield estimation zone using the random forest importance evaluation and Leave-One-Out. Secondly, the yield estimation models were built and validated for each yield estimation zoning. After that, the yield of three crops was estimated monthly during the growing season of 2022 (June-September), according to the yield estimation models. Taking the historical five years as the base period, the maximum indexation was used to index the predicted yield to obtain the crop yield index. Finally, the monthly analysis was implemented to determine the changes in the yield index that related to the historical reference period and the previous month during the growing season of 2022. The results show that: 1) The whole province was divided into 3, 3, and 4 yield estimation zones for rice, soybean, and maize, respectively. 2) The importance of modeling indicators was ranked in the descending order of the trend yield, remote sensing vegetation, and climate indicators. 3) The importance of each candidate modeling variable was varied in the different yield estimation zones. The different optimal modeling variables were also selected for each yield estimation zone. In the yield estimation models using the optimal modeling variables, the yield prediction accuracy of rice was the highest, with a mean absolute relative precision (MARP) of 95.20%, followed by maize (MARP=93.81%), and finally soybean (MARP=92.73%). 4) The yield indices of three crops from June to September were compared with the calculated using a historical five-year period. There were "flat" state differences in the yield indices of the three crops in most counties (i.e., the difference in the yield index was within 0.05). 5)The monthly yield index differences (the current month's yield index minus the previous month's yield index) of the three crops in most counties are between −0.01 and 0.01. The crop yield index shared excellent spatiotemporal comparability. On the temporal dimension, the crop yield index can be expected to compare the increase or decrease of a certain statistical unit (such as a county, city, or province) that is related to the historical average yield at a specific evaluation time point, or to compare the yield changes of two adjacent evaluation time points. On the spatial dimension, the crop yield index can be compared with the different statistical units at a specific evaluation time point. This crop yield index can provide promising application prospects in growth monitoring and yield forecasting. [ABSTRACT FROM AUTHOR]

Objectives: SSS (sea surface salinity) is an important physical and chemical indicator describing ocean state, and is also one of the most critical variables in global water cycle. In order to achieve largescale and continuous observation of the ocean salinity, SMOS (soil moisture and ocean salinity) mission is launched by European Space Agency to access brightness temperature for SSS retrieval. However, researches have pointed out that the accuracy of SMOS ocean salinity products cannot fully achieve scientific expect. Thus, a novel and effective method is desired for improving SMOS salinity. Methods: This paper selects study area in the Southeast Pacific, where the region is relatively less affected by radio frequency interference. The downloaded array for real-time geostrophic oceanography SSS data, SMOS level2 ocean salinity products and the Auxiliary data in the whole year of 2018 are matched both temporally and spatially. Based on the remote sensing mechanism and radiation transfer theory, five variables, including the first Stokes parameter (TH+TV), SST, UN10, VN10 and Ω are selected as sensitive factors. The ocean salinity retrieval model based on random forest is first established, and GridSearchCV(GS) algorithm is applied to optimize superparameters of the initial model, which is able to improve SSS accuracy further. Results: The experiment results show that： (1) For objective measurement, the mean absolute error (MAE) and root mean square error (RMSE) are adopted to compare the accuracy of the produced SSS from two established models and the SMOS products, in terms of Argo in-situ SSS as reference. According to experiments, MAEs of basic RF(random forest) model and GS‐RF model are both 0.08, which is evidently lower than SMOS SSS product's 0.66. Similarly, RMSEs of two models are also remarkable lower than SMOS's 0.93, while GS RF model is 0.14, indicating its slight advantage over RF model, which is 0.15. (2) All SSS retrieved from RF model and GS RF model distribute concentratedly around 36 psu, while SMOS ocean salinity product occasionally presents abnormal SSS that beyond 40 psu or below 30 psu, which is inappropriate. (3) In the testing set, error of SMOS SSS can reach −12.9 psu and 6 psu, while errors of two established models are no more than ±1.3 psu, noticeably lower than SMOS product. Meanwhile, GS‐RF model also present slight advantage than basic RF model on error measurement. Conclusions: Through experiments and analyses, we found that it is feasible to establish sea surface salinity retrieval methods based on random forest model, by deliberately selecting SSS sensitive factors, and advancedly optimizing super-parameters of model, especially with GridSearchCV algorithm. This paper has achieved reliable and satisfactory results, which have greatly improved the accuracy of ocean salinity data compared with SMOS products, thus benefit for relevant marine studies. [ABSTRACT FROM AUTHOR]

In order to predict the moisture of the mixture, a modeling method based on stacked generalization ensemble is introduced, and a Robust Scaler-Rank Gauss（RS-RG） hybrid algorithm is proposed to process the data input to the stacking model, and then the LGBM-RF-XGB hybrid model is established for sintered mixture moisture prediction based on stacked generalization ensemble, which can predict the moisture value before sinter mixture mixing. The internal mechanism of the LGBM-RF-XGB overlay model consists of generating metadata from the LGBM and RF models to calculate the final prediction using the XGB model. The proposed stacking model was simulated in comparison with several reference models by combining the sintering site data. The results show that the accuracy and error of the proposed stacking model are better than those of the reference models used for comparison, which meets the sintering process requirements. The proposed algorithm can be used for advance prediction of sinter mix moisture in actual production and provides a theoretical basis for automatic control of water addition. [ABSTRACT FROM AUTHOR]

To solve the problems of video freezes and low quality of experience (QoE) in panoramic video transmission, this paper proposed a bitrate adaptive strategy based on viewpoint prediction (VPBAS) by studying the current popular viewpoint adaptive transmission scheme.Firstly, this paper constructed a viewpoint prediction model based on long short-term memory network and full convolutional network.The model concatenated the viewpoint data and video saliency information to realize the mutual complement and correction of different modal data, so as to improve the accuracy of viewpoint prediction.Secondly, the client used the random forest algorithm to predict the current available bandwidth, and selected the bitrate for video segments according to the prediction results and available bandwidth information.Finally, the client sent the selected bitrate information to the server regularly, and the server pushed the panoramic video stream of the best bitrate to the client according to the feedback information.It repeated this interactive process continuously during the video playback until the client finished watching.Experimental results show that VPBAS can effectively improve the user’s viewing experience under bandwidth limitation compared with other schemes. [ABSTRACT FROM AUTHOR]

This study aimed to accurately estimate the organic carbon density of topsoil in Fujian Province and determine the spatial distribution and influencing factors of topsoil organic carbon density. Using 4 350 topsoil samples collected from 2016 to 2019, the spatial distribution of organic carbon density was analyzed by geostatistics and ArcGIS, and the influencing factors of soil organic carbon density were analyzed using a random forest model. The topsoil organic carbon density was 4.06 kg•m-2(range, 0.03-14.68 kg•m-2). The topsoil organic carbon density was spatially concentrated and distributed with a moderate degree of spatial autocorrelation. The spatial variation was mainly dominated by structural factors, showing a spatial distribution pattern that gradually increased from south to north, from coastal to mountainous areas. Precipitation, altitude, soil texture, and pH were the main factors affecting the organic carbon density. Landform type, soil type, and crop type exerted relatively weak effects, and the soil organic carbon density reduced significantly with the increase in chemical fertilizer(phosphorus and potassium) application. These results showed that the topsoil organic carbon density in Fujian Province is not high, and the spatial distribution is concentrated. Natural climate change, terrain factors, and human activities all affect the soil organic carbon storage level, with precipitation, altitude, soil texture, and pH being the dominant influencing factors. [ABSTRACT FROM AUTHOR]

Many kinds of hydrocarbon reservoir damages with complex mechanisms have been encountered in every phase of oil and gas field exploration and development. Conventional core flow test used in evaluating the sensibility damage of a reservoir can give reliable test results, however, this test is both expensive (coring, for instance) and time consuming. Researches have shown that a model established with neural network and random forest algorithm on small-scale samples can be used to save time and money in predicting the properties of samples. In this study, the data of a set of small-scale samples tested in laboratory is obtained from the block X. The training-sets and testing-sets are then selected on the samples. By extensively comparing the results of three algorithms, which are the BP neural network algorithm, the radial basis function neural network algorithm and the random forest algorithm, the random forest algorithm is finally selected as the main method of quantitatively diagnosing the sensitivity damage of hydrocarbon reservoirs. To improve the prediction accuracy, algorithms such as grid search are used in hyperparameter optimization, and data dimensionality reduction is performed based on factor weight. A complete model is finally established based on the studies conducted. The average R² value of the four kinds of reservoir damage model is 0.852, with a prediction accuracy between 90.00% and 95.68%. . [ABSTRACT FROM AUTHOR]

A digital twin construction method based on the reduced order model library and machine learning was proposed for structures under dynamic loads. Firstly, the high-fidelity finite element models were established according to the possible damage states occurring during the service of the physical structures. Secondly, the Krylov subspace order reduction method was used to reduce the orders of the models and the reduced order models were assembled to a library. Finally, the random forest machine learning algorithm was used to train the model selector, infer the current state of the physical structure through the sensor data from the structure, and then drive the digital twin to evolve with the physical structure. A physical frame structure was designed and manufactured to simulate the damages of different degrees at different points, and verify the proposed digital twin construction method for dynamic structures. [ABSTRACT FROM AUTHOR]

In order to construct a high-precision prediction model of forest surface litter moisture content, this paper takes three typical forests in Harbin City, Heilongjiang Province as the research object, and uses a self-made portable meteorological factor detector for real-time detection. Three representative machine learning methods of RF ( Random Forest ), BP ( Back Propagation ), and LSTM ( Long Short Term Memory ) were used to construct litter moisture content prediction models for the three stands. Results showed that, the prediction accuracy of the prediction model constructed by different forest stands under different machine learning methods was different. The prediction accuracy of the three stands in different machine learning models was LSTM > BP > RF. LSTM had the best effect in the study of surface litter moisture content prediction model. The results of this study indicated that LSTM was outstanding in the prediction of surface litter moisture content, which had important theoretical significance and practical guiding value for improving the accuracy of surface litter moisture content model based on meteorological element method. [ABSTRACT FROM AUTHOR]

Given the problem that the communication distance and quality of the vehicle-mounted ultrashort wave radio are affected by ground attachments and topography in the actual combat environment, a radio wave propagation prediction model of vehicle-mounted ultrashort wave radio is established based on ray tracing and machine learning. The integrated modeling of armored combat vehicle and vehicle antenna is established to obtain the antenna radiation pattern, and combined with electronic images, the radio wave propagation simulation model based on ray tracing technology is established. Based on the machine learning algorithm of the random forest and data results for the simulation model, the radio wave propagation prediction model based on the random forest was established. Compared with traditional classical radio wave propagation models such as the Egli and Okumura-Hata models, the radio wave propagation prediction model based on the random forest has higher accuracy. The root mean square error reaches 2.190 1 dB, and the coefficient of determination reaches 0.960 1. It can accurately predict radio wave propagation in the tactical communication environment. [ABSTRACT FROM AUTHOR]

The registration quality assessment is usually given to human experts, which is time-consuming. The commonly used Dice score only focuses on the error at the edge of the tissue, which is difficult to assess the registration result within the tissue. To address these issues, this paper proposed a method to predict registration errors based on machine learning (PREML) in lung CT images.This method firstly constructed three types of features, such as deformation field statistical features, deformation field physiologically realistic features and image similarity features, then expanded the number of features by pooling, and finally used random forest regression to predict non-rigid registration errors. Moreover, it used an adaptive random perturbation to simulate the spatial distribution of lung registration errors to further improve the capability of error characterization of statistical features. The proposed method achieved a mean absolute error of 1.245±2.500 mm from ground truth on lung CT image datasets, outperforming the baseline method. The results show that PREML method has the advantages of high accuracy and robustness, enhancing the safety and effectiveness of registration algorithms in clinical applications. [ABSTRACT FROM AUTHOR]

The nonlinear relationship between the chlorophyll content of Hopea hainanensis, an endangered tree species, and image features was discussed, and the feasibility of estimating the chlorophyll content of H. hainanensis by image color and texture features was studied in this paper, providing a reference for image-based non-destructive monitoring. The canopy chlorophyll content (SPAD value) and RGB images were obtained using a 2-year-old H. hainanensis as the research object. The generalized additive model (GAM) was used to explore the nonlinear correlation between the image color and texture characteristics and the SPAD values. The SPAD values were then estimated using the multiple linear regression, GAM, random forest, and XGBoost models based on the features without multicollinearity filtered by the Lasso algorithm. Among the 22-image color and texture features, the feature with the strongest linear correlation was a correlation standard deviation, which was positively correlated, while the feature with the strongest nonlinear correlation was entropy mean, which was inverse J-shaped and negatively correlated. The Lasso algorithm effectively removed the severe collinearity among multiple image features, and the VIF values of the 15 image features retained after screening were all lower than 10. Based on the results of estimating SPAD values using multi-image color texture features, the test accuracy of the XGBoost model was highest, and the average testing R~2, MAE, and RMSE were 0.757, 2.682, and 3.242, respectively. Therefore, there are certain nonlinear correlations between SPAD values of H. hainanensis and image color and texture features. The image estimation method based on the GAM model with nonlinear interpretation ability can be used as the optimal method for estimating the chlorophyll content of H. hainanensis. [ABSTRACT FROM AUTHOR]

Marketization of rural idle homesteads can be promoted for “three rights separation” of rural homesteads. The perceived risk can be one of the most important identification factors in the farmers’ homestead marketization in the reform of the homestead system. In this study, an evaluation system was constructed to identify the risk of homestead marketization from the perspective of farmers’ risk perception, including the risk of the farmers’ equity, cultivated land protection, collective return, social stability, ecological environment, and cultural inheritance risk. Taking Chongqing Dazu, Yongchuan, and other typical homestead system reform pilots as empirical areas, a Variable Fuzzy set was adopted to evaluate the risk level of farmers’ homestead marketization. Subsequently, an efficient machine learning model (called Random Forest, RF) was employed to identify the risk of farmers’ homestead marketization. Finally, risk prevention and control strategies were proposed under different marketization patterns. The results showed that: 1) An inverted “U” shaped fluctuation trend was found in the risk level of farmers’ homestead marketization. Specifically, the risk probability of direct marketization (43.38%) and green indicators marketization (32.14%) were significantly higher than that of marketization of reclamation indicators (13.73%), and marketization of planning adjustment (12.99%). 2) There were some differences in the various risk factors for the four types of marketization patterns. The farmers’ equity risk (46.74%) was higher in the direct marketization, whereas, the cultural inheritance risk (4.76%) was lower. The social stability risk (35.66%) was higher in the marketization of planning adjustment, while the ecological environment risk (5.85%) was lower. The cultivated land protection risk (29.13%) was higher in the marketization of reclamation indicators, while the collective return risk (2.05%) was lower. The farmers’ equity risk (34.24%) was higher in the green indicators marketization, while the ecological environment risk (1.20%) was lower. 3) There were some differences in the risk mechanism. The same risk factor also shared the heterogeneity in the homestead marketization of four types. 4) The differentiated prevention strategies of marketization risks were considered the timing, and the cooperation of various participants, such as farmers, collectives, and governments. The government, the collective, and the farmer were united to focus on the “prevention in advance” in the direct marketization. The collective was supposed to focus on the “prevention in advance” in the marketization of planning adjustment, while the government was on the “prevention afterward”. By contrast, the government was supposed to focus on the “prevention in the event” in the marketization of reclamation indicators, while the collective was on the “prevention afterward”. Only the government was supposed to focus on the “prevention in advance” in the green indicators marketization. In conclusion, the reform of homestead marketization should set the tone of “steady and prudent promotion”. It was different from the complete marketization of state-owned land. But the marketization was limited using “three rights separation”. Meanwhile, future homestead marketization should avoid homogenization patterns. More importantly, it is urgent to establish an income distribution mechanism and risk aversion system, in order to share the benefits and risks by farmers, collectives, and governments. The findings can provide the decisionmaking on rural residential land consolidation. The homestead risk database can be expected to deepen the reform of the homestead system. [ABSTRACT FROM AUTHOR]

The backwater effect caused by the confluence of the Dongting Lake is a critical factor influencing the hydrological processes in the Jingjiang reach, and an analysis of its changing characteristics is of critical importance for studying flood control safety issues in the middle and lower reaches of the Yangtze River. A quantitative method was proposed to analyze the degree and range of the backwater effect in terms of stage- discharge rating curves in the Jingjiang reach and the discharge ratios of the Dongting Lake inflow from 1990 to 2020. Then, a random forest regression model was constructed to calculate the backwater effect and sequence its four key influencing factors. The results demonstrated that: ① There was a significant positive correlation between the degree of backwater effect and the discharge ratio. In addition, the degree of backwater effect also increased with an increasing discharge from the mainstream. The mean degrees of backwater effect characterized by the water level variation were 0. 59, 1. 33 and 1. 60 m respectively at the Jianli hydrometric station under the conditions of low discharge, bankfull discharge and flood discharge in the Middle Yangtze River during the period 2003-2020. ② The range caused by the backwater effect extended to the upstream with increasing mainstream discharges and discharge ratios. In 2020, with the increasing discharge ratios, the range zones under the three characteristic discharges were the reaches of Shishou- Shashi, Shishou-Chenjiawan and Chenjiawan-Zhicheng, respectively. ③ Furthermore, the contributions of discharge ratio, cumulative scouring depth, water level at the Luoshan hydrometric station, and mainstream discharge to the backwater effect accounted for 28%, 27%, 25% and 20%, respectively. It is varified that the random forest regression model can reflect the degree and range of the backwater effect under various discharge conditions in the Jingjiang reach. [ABSTRACT FROM AUTHOR]

Soil organic carbon (SOC) responds significantly to the surface layer, but also shows different responses with depth. Jianghan Plain plays an important role in the agroecosystem of the Yangtze River Economic Belt, but its SOC baseline remains unclear. In this study, 66 profiles were collected from 2009 to 2012 in Jianghan Plain, and SOC content prediction models were constructed based on nine environmental factors as covariates using the random forest method for three depths：0–30, >30–60 cm, and >60–100 cm. We constructed a baseline map of SOC at 30 m resolution and estimated the SOC stock. The results showed that SOC contents decreased with increasing soil depth, and were generally high in the east-central part and low in the west. The model had the highest prediction accuracy (R² =0.45, RMSE= 3.28 g·kg−1 ) for the topsoil (0– 30 cm) . Temperature, clay content, and precipitation were the three most important covariables for the model. The soil organic carbon storage at 1 m depth in Jianghan Plain was 183.75 Tg, and >30–100 cm accounted for approximately 59% of the total storage at 1 m depth. We conclude that it is necessary to focus on deep soils when evaluating the soil carbon pool and sequestration potential. Our new estimate can serve as a base for both horizontal and vertical variation of SOC. [ABSTRACT FROM AUTHOR]

The geographic distribution data of the three precious hardwood broad-leaved tree species, Fraxinus mandshurica, Juglans mandshurica and Phellodendron amurense in the Maoershan area were collected through investigation and consulting data. And the suitability distribution of three precious hardwood broad-leaved tree species in Maoershan area was studied by using GIS technology and random forest model, combined with the topographic and soil data of the target tree species sub-compartment. Results showed that, the accuracy rates of the random forest models in the ten datasets were all above 70%. When Ntree(number of trees grown)= 500, the average accuracy of the random forest model was 75.2%; when Ntree = 1000, the average accuracy of the random forest model was 75.4%, and the AUC of the group datasets were all above 0.7, and the average AUC was 0.746. In the ten datasets, mean decrease accuracy and mean decrease Gini of altitude were the largest, mean decrease accuracy of slope position was the smallest, and mean decrease Gini of landform was the smallest. The target tree species had the highest growth suitability within the altitude range of 200-300 m. As the altitude increased above 300 m, the growth suitability of the target tree species was lower as the altitude increased. When the slope was 0°-10° and 30°-35°, the growth suitability was the highest in the range. As the slope position changed from top to bottom, the growth suitability of the target tree species was getting higher and higher. The growth suitability in the three directions of southwest, south and north was the highest. The growth suitability was highest in plain areas. When A layer (eluvial horizon) thickness was greater than 15 cm, the growth suitability of the target tree species became higher and higher. When AB layer (eluvial horizon and deposition layer) thickness was in the range of 10-25 cm, the growth of the target tree species was the most suitable. The order of affecting environmental factors importance sorting of the suitability distribution of the three precious hardwood broad-leaved tree species in the Maoershan area was altitude, AB layer thickness, slope, slope aspect, A layer thickness, landform and slope position in descending order. They were mainly suitable for distribution in the southern and northwestern areas of Maoer Mountain, with an altitude of 200-300 m, a gentle slope, a soil layer thickness of 15-25 cm, a north-south aspect, a plain landform, and a downhill position. [ABSTRACT FROM AUTHOR]

In order to explore the optimal feature combination of Sentinel-2 remote sensing image stand type classification, the classification and effect evaluation of broad-leaved forest, masson pine forest, fir forest and bamboo forest were realized. Selecting Changting County, Fujian Province as the study area, 10 original bands（O)were extracted from Sentinel-2 images, and 9 spectral indices（S）, 7 red-edge spectral indices（R）, 8 texture features（Te)were calculated, and 2 terrain feature indices（To)based on digital elevation data were calculated, for a total of 36 features. Using random forest algorithm to analyze the importance of different features in stand type classification, using out of band（OOB)data and average impurity reduction method to optimum individuality combination（OIC）. 6 different experimental protocols（O, O+To, O+To+S, O+To+S+R, O+To+S+R+Te, and OIC)were classified into stand types and the results were evaluated by confusion matrix. The results showed that the importance of the 36 features involved in the classification of stand types was 2.11%-5.43%, the altitude factor was the most important, and the mean and correlation of the red edge band, red edge spectral index, and texture features were also of high importance. Using the original band alone to classify the stand types, the classification accuracy was not high, the overall accuracy was 73.26%, and the Kappa coefficient was 0.64. Based on the original band, other features were introduced. Except the original band, other features can improve the classification accuracy. The optimum individuality combination（OIC)was the top 27 features of importance, including altitude, 8 original bands, 7 red-edge spectral indices, and 3 texture features, the classification accuracy was the highest, the overall accuracy was 83.13%, and the Kappa coefficient was 0.77, which was 0.82%-9.87% higher than the overall classification accuracy of the other five experimental schemes. Using Sentinel-2 images as the data source, the feature combination optimized by the random forest algorithm integrated the features that had an important contribution to the classification of stand types among the multi-type features, thereby improving the classification accuracy. The research results can provide reference for GEE platform Sentinel-2 image extraction of stand type information in forest resource survey. [ABSTRACT FROM AUTHOR]

Hyperspectral remote sensing technology has been widely used in vegetation type mapping, and the integrated use of spectral and texture features of remote sensing data has become an effective way to improve the accuracy of vegetation classification, but the Hughes effect generated by the high-dimensional feature space composed of multi-scale texture features and hyperspectral features reduced the image classification accuracy, and the randomness of feature optimization algorithms such as genetic algorithm (GA) can also cause uncertainty in image classification accuracy. In order to solve those problems, a high-dimensional spectral-texture feature space was established by incorporating spectral features of ZY1 02D Advanced HyperSpectral Imager(AHSI) hyperspectral data and multi-scale texture features extracted from ZY1 02D Visible and Near Infrared Camera(VNIC) data at various window sizes and movement directions, covering the vicinity of Zongjia Town in Dulan County, Qinghai Province, China. Jeffries-Matusita(J-M) distance is employed to assess the sample separability between vegetation types of texture features derived from varying window sizes and movement directions. Based on a combination of feature optimization algorithms (GA algorithm, particle swarm optimization (PSO), generalized normal distribution optimization (GNDO), atom search algorithm (ASO) and marine predators algorithm (MPA)) and classification algorithm (random forest (RF)), GA-RF, PSO-RF, GNDO-RF, ASO-RF, and MPA-RF algorithms were proposed and applied to the vegetation type classification of the high-dimensional spectral-texture feature space. The results show that the texture features extracted from different window sizes and window movement directions showed maximum J-M distance between distinct vegetations. For instance, the J-M distance between wolfberry (old) and wolfberry (new) is highest in the 3×3 window size texture images, while the J-M distance between wolfberry (old) and haloxylon is highest in the 11×11 window size texture images. In the 0° window movement direction texture images, poplar and haloxylon had the best separability to other vegetation types with J-M distances greater than 1.87 and 1.86, respectively. Grass and other vegetation types achieved the best separability with J-M distances greater than 1.18 in the 90° window movement direction texture images. The inclusion of multiscale texture features improved the overall classification accuracy (OA) by 8.02 percentage. The proposed GA-RF, PSO-RF, GNDO-RF, ASO-RF, and MPA-RF algorithms, when compared to the conventional random forest algorithm, led to an improvement in OA of vegetation classification by 1.32 to 2.41 percentage. Among them, the MPA-RF method exhibited the highest accuracy, achieving an OA and Kappa coefficient of 88.92% and 0.86, respectively. The results of this study demonstrate that texture images extracted from different window sizes and window movement directions are useful in distinguishing between different types of vegetation. The accuracy of vegetation recognition is significantly increased by adding multi-scale texture features to the spectral feature set. The combination of feature optimization algorithm and image classification algorithm by iterative optimization alleviates the randomness of the optimization algorithm results, overcomes the Hughes effect of high-dimensional features, and successfully improves the vegetation classification accuracy. This study proposes innovative ideas for feature extraction, feature optimization, and classification algorithm selection for hyperspectral vegetation classification studies. [ABSTRACT FROM AUTHOR]

Objectives: Land cover classification in national parks plays an important role in understanding the status of natural resources, identifying the existing ecological security threats and responding to them quickly. Methods: Two land cover classification methods are developed based on Google Earth Engine (GEE) platform by combining Sentinel active and passive remote sensing data, and spectral indices, textural features and topographic features derived from the data to classify land cover types in the Qianjiangyuan National Park (cropland, forest, grassland, water body, artificial surface and bare land). One used pixel-based random forest (RF) classification algorithm, the other used object-oriented simple non-iterative clustering (SNIC) segmentation in partnership with RF algorithm. Results: The ground experimental results show that the highest overall classification accuracies of the pixel-based method and the object-oriented method are 92.37% and 93.98%, respectively. Furthermore, the integration of synthetic aperture radar (SAR) data can substantially improve the classification accuracy when using the pixel-based method, but there is no apparent escalating effect for the object-oriented method. Conclusions: Land cover classification map generated by SNIC+RF algorithm in GEE platform is more complete and the algorithm requires fewer features and runs quickly in GEE platform. Thus, this algorithm deserves to be popularized in national park management practices. [ABSTRACT FROM AUTHOR]

It is necessary to improve the generalization of tomato maturity grading for better consistent criteria in modern agriculture. In this study, an extraction was proposed for red-colored regions from the fruit surfaces using region brightness correction. The R-G method was used to enhance the red areas on the tomato surface, and then the Otsu method was used to segment them. All the colored areas on the tomato surface were obtained to evaluate the relationship between the contours of each red region of tomato after segmentation by the contour tree structure. The proportion of the area of the red region to the total area of the image was calculated as the main feature of maturity grading. The surface red coloring area ratio, color moments, and color histograms were selected to perform the feature importance analysis, in order to screen out the color features with a significant impact on the maturity grading. The random forest model was used to determine and grade the different maturity stages of tomatoes. Among them, the image in-painting technique based on the fast marching method was used to reduce the high brightness reflection of surface that reduced by the illumination changes. The test results showed that the effect of the area ratio of fruit surface coloring areas on the tomato maturity grading was significantly greater than the color histogram and color moment. The coloring area ratio, G-component third order color moments, and G-component color histogram feature posed the greatest impact on the maturity grading, with importance scores of 0.293, 0.199, and 0.127, respectively. Among all color features, the tomato coloring area ratio as a classification index shared the highest classification accuracy, with an average classification accuracy of 92.96%, which was 6.53 and 20.6 percent point higher than the traditional color moment and color histogram indicators. The domain pixel-weighted sum performed an excellent correction effect on the brightness of the tomato-highlighted areas. After correction, the proportion of the fruit surface colored area of immature, slightly mature, and mature tomato images to the total area of the tomato images increased by 0.06, 0.15, and 0.11, respectively, compared with before correction. The classification accuracy was improved by 17.24, 11.47, and 4.69 percent point, respectively. The area extraction of tomato red areas with the image fast restoration can be used as the recommended extraction of tomato maturity grading. [ABSTRACT FROM AUTHOR]

Objectives: Near surface air temperature (NSAT) is a key parameter in the land-atmosphere interaction process. Sparse NSAT observations from in-situ stations usually cannot fully describe the spatial distribution of NSAT, so estimating NSAT by land surface temperature (LST) and auxiliary variables has become an effective approach to obtain the spatial distribution of NSAT. Although there are some multisource LST products published, the LST from moderate-resolution imaging spectroradiometer （MODIS）, Landsat, and global land data assimilation system (GLDAS), the applicability of each LST product in NSAT estimation still needs further investigation. Methods: Taken the Yellow River Basin Region as the study area, summer NSAT from 2003 to 2019 was estimated based on the Google Earth Engine platform with random forest algorithm in this study, and the mean, maximum and minimum NSAT was estimated in two scales (30 m and 1 000 m) using three LST data sources (Landsat, MODIS and GLDAS). The observed LST from in-situ stations over the Yellow River Basin Region were compared with the estimated NSAT by the ten-cross validation method to evaluate the accuracy of different LST sources when estimating NSAT. Results and Conclusions: The results indicate that the differences of accuracy of the three LST sources are small in terms of mean NSAT. In terms of the maximum and minimum NSAT, the GLDAS LST shows the significant higher accuracy than MODIS and Landsat LST. The root mean squared error of estimated mean NSAT are smaller than maximum and minimum NSAT estimation when using the same LST source. For the spatial distribution of accuracy, the stations with higher error mainly located in the southern or western of the study region. The temporal resolution of LST source is significantly important in NSAT estimation. The GLDAS LST shows the highest accuracy in this study especially for extreme NSAT estimation. Besides, the mean NSAT estimation has higher accuracy than that of maximum or minimum NSAT for each LST source. [ABSTRACT FROM AUTHOR]

Objectives: To address the problem of internal inconsistency of classification targets in existing three dimensional(3D) point cloud data segmentation and classification methods. we propose a high-precision classification method for indoor point cloud jointly optimized by super voxel random forest and long short-term memory (LSTM) neural network. Methods: The method takes into account that the super voxel structure has the characteristics of internal feature consistency, divides the original point cloud into super voxels, and uses super voxels as the basic unit for multivariate feature calculation to build a super voxel random forest classification model for indoor point cloud to achieve coarse classification of point cloud data. On this basis, LSTM is introduced to train and predict the neural network model for the hyper voxel neighborhood connectivity of coarse classification to achieve the optimization of hyper voxel coarse classification results. The validity and accuracy of the proposed classification method are verified based on the open dataset. Results: The results show that the classification accuracy of the proposed classification method can reach 83.2% for 13 types of elements in the open dataset. The training data of the LSTM optimization network proposed in this paper used only the label information of region 1 for model training, while other deep learning frameworks used regions 1-5 for model training, so from the perspective of training data requirements, the point cloud data classification framework proposed in this paper can achieve a relatively better prediction result with a small portion of the training data set. The super voxel-based LSTM optimization method approach has high classification accuracy on objects with obvious set features such as ceiling, floor and wall, however, it is inferior to the deep learning algorithm RandLA-Net in classifying objects with complex structures such as chair, sofa and bookcase. Conclusions: In this paper, we consider the association characteristics between different types of elements embedded in the connection relations among super voxels, and introduce LSTM to train and predict the model for the coarse classification of super voxel neighborhood connection relations to achieve the optimization of coarse classification results of super voxels. The proposed method can achieve better classification accuracy when trained with small samples compared with the classical deep learning framework. [ABSTRACT FROM AUTHOR]

【Objective】 The spatial-temporal variation characteristics of soil pH and its driving factors were clarified to provide a basis for the improvement of soil acidification in southwest cigar area. 【Method】 Based on soil pH data of 26 sample points of the second general soil survey in Sichuan Province in 1984, 391 sample points of soil testing and fertilizer recommendation survey in Sichuan Province in 2009, and 164 sample points of cultivated land quality survey in Sichuan Province in 2019, Multi-statistics and Geostatistics were used to analyze the spatial-temporal variation characteristics of soil pH, and Random Forest model was used to explore the main driving factors of soil pH change.【Result】 The mean values of pH in Shifang City in 1984, 2009 and 2019 were 6.40, 6.24 and 5.97, respectively, and the average annual value decreased by 0.01 from 1984 to 2009 and by 0.02 from 2009 to 2019. From 1984 to 2019, except for Jiandi Town and Yinghua Town, where the average soil pH value showed an upward trend, the rest of the towns were on the decline, with the largest decrease of 30.62% in Hefeng Town and the smallest decrease of 0.30% in Shigu Town. The overall soil pH in 1984 was dominated by weakly acidic and neutral properties, with alkaline soils accounting for 2.59% of the total area planted with tobacco, and the spatial distribution was characterized by higher soil pH in three areas: Luoshui Town, Hefeng Town and Majing Town. In 2009, only weakly acidic and neutral soils existed in the study area, and soil pH was spatially high in the centre and low in the north and south. In 2019, the area of weakly acidic and neutral soils accounted for 95.25%, and the area of acidic soils (pH<5.5) accounted for 4.75%, which were mainly distributed in the towns of Majing, Mazu, Hefeng, Nanquan and Shigu. And soil pH in the south was spatially lower than that in the north. Soils planted with tobacco over the past 35 years have generally shown changes in acidification, with about 53.51% of the soil was acidified (ΔpH < -0.3), mainly distributed in the towns of Majing, Hefeng and Mazu, and about 11.63% of the soil was alkaline (ΔpH ＞ 0.3), mainly in the towns of Yinghua and Jiandi; The pH of 44.98% tobacco growing soils ranged from -1.0 to -0.3, 34.86% ranged from -0.3 to 0.3, and tobacco growing soils with ΔpH ≤ -1.0 accounted for 8.54%. The random forest results showed that mean annual precipitation and soil organic matter were the main factors causing soil acidification.【Conclusion】Over the past 35 years, the acidification of tobacco growing soils in Shifang City was aggravated. Some measures should be taken to adjust the situation in production, such as creating new special fertilizer and applying soil conditioner. [ABSTRACT FROM AUTHOR]

Aimed at the low accuracy of loudspeaker abnormal sound classification, an abnormal sound classification method based on the combination of adaptive variational mode decomposition(AVMD) and residual convolutional neural network(RCNN) is proposed. In AVMD, the grey wolf optimizer (GWO) algorithm was used to optimize the quadratic penalty factor(α) and modal decomposition number(K) of the variational mode decomposition (VMD), and then the VMD was used to extract the features. Finally, the random forest recursive feature elimination(RF-RFE) algorithm was used to extract the optimal feature data, and residual network (ResNet) was introduced by RCNN-3 to classify abnormal sound based on convolutional neural network(CNN). The experimental results show that the proposed method has better classification accuracy and stability, and its average classification accuracy can reach 99.3%. [ABSTRACT FROM AUTHOR]

Accurate segmentation of prostate ultrasound images in the clinic has an important impact on the subsequent diagnosis, and the currently available research findings cannot accurately segment each part. This paper proposed a bimodal shape reconstruction method based on the point distribution model and stream shape learning and segmentation of prostate ultrasound images: target initial localization by random forest indication of the hidden state expression, improvement of the boundary operator to improve the coarse segmentation accuracy, interpolation computation using some gray significant points adjacent to the noisy part of the explicit state expression to recover the shape as a whole. This segmentation method not only reduced the data computation amount, but increased the segmentation reliability. Experiments show that the DSC of the proposed method is 97. 38% and the mIoU is 95.24%, and the accuracy is stronger than the current popular segmentation neural networks. [ABSTRACT FROM AUTHOR]

In the big data environment, the random forest algorithm suffers from excessive redundancy and irrelevant features, the insufficient spatial information content of feature subspace, and low parallelization efficiency. To resolve these issues, this paper presented PRFGRSAE. Firstly, this algorithm proposed a DRNGRSAE, which filtered redundant and irrelevant features of the feature set and extracted features by stacked auto-encoders to reduce the number of redundant and irrelevant features effectively. Secondly, it proposed a SSLF that combined Latin hypercube sampling and normalized correlation degree, which formed feature subspaces with high spatial expression by performing multi-layer division sampling on the feature set, and ensured the feature subspace information content. Finally, it proposed a reducer allocation strategy DSVLA combining with variable action learning automata, which allocated each cluster to reducers for processing evenly and improved the parallelization efficiency effectively. Experimental results show that compared with IMRF, KSMRF, and GAPRF algorithms, the speedup ratio and accuracy of the PRFGRSAE algorithm are significantly improved. Therefore, the algorithm can obtain higher accuracy and parallel efficiency when applied to process large data, especially for data sets with more features. [ABSTRACT FROM AUTHOR]

[Objectives] The paper aimed to explore the effects of microbe-consuming protists, fungi and the microbe-consuming protists and growth-promoting fungi co-inoculation on cucumber growth and soil microbial community. [Methods]A greenhouse experiment was designed to decipher the role of keystone protists (Cercomomnas lenta) in interaction with plant-promoting Trichoderma guizhouense on the soil microbiome and cucumber performance. Four treatments were designed: no-inoculation control(CK), T. guizhouense inoculation treatment(T), Cercomomnas lenta inoculation treatment(C),and C. lenta and T. guizhouense co-inoculation treatment(T+C). Plants and soil samples were collected after two weeks of greenhouse experiment to evaluate the effects of each treatment on cucumber performance and soil bacterial and fungal communities. [Results] Compared to the control, plant height of T. guizhouense inoculation treatment, C.lenta inoculation treatment and C. lenta and T. guizhouense co-inoculation treatment increased by 6.0%,12.5% and 21.5%, respectively, while plant biomass increased by 5.6%,12.9% and 24.0%, respectively. This indicated that the growth promotion effect of C.lenta and T. guizhouense co-inoculation was stronger than those of individual inoculation of C. lenta and T. guizhouense. Principal coordinate analysis showed that three inoculation treatments significantly changed the bacterial and fungal community compositions; random forest models revealed that bacterial and fungal community compositions were the main predictors of plant biomass; and the LEfSe analysis showed that the T. guizhouense inoculation, C. lenta inoculation and C. lenta and T. guizhouense co-inoculation treatments stimulated 22,51 and 82 potential key bacterial OTU, and 2,6 and 16 potential key fungal OTU, respectively. In total 21 key bacterial and 11 key fungal OTU were only stimulated in the C. lenta and T. guizhouense co-inoculation treatment potentially contributing to cucumber growth. The bacterial OTU mostly contributing to cucumber growth promotion were assigned to Parasediminibacterium, Fluviicola and Roseimaritima, and the fungal OTU mostly contributing to cucumber growth promotion were assigned to Dactylella, Melanoleuca and Chaetomium. [Conclusions] C. lenta and T. guizhouense co-inoculation can effectively enhance plant performance through co-regulating indigenous bacterial and fungal communities and stimulating potentially beneficial microorganisms. [ABSTRACT FROM AUTHOR]

To explore the spatial distribution patterns of traffic accidents at urban road intersections and find out the key factors affecting the severity of accidents, 1 758 valid intersection accident samples and 9 categories features were extracted from the traffic accident database of Harbin. The spatial distribution of intersection accidents was visualized, and the density analysis algorithm was used to obtain the spatial distribution characteristics of intersection accidents considering the road network density, intersection density and severity respectively. The accident severity was divided into fatal accidents and non-fatal accidents. The random forest was used to rank the significant factors affecting the accident severity in the whole urban area, low density area and medium-high density area. The results show that the spatial distribution of intersection accidents differs considering road network density, intersection density and severity respectively. Season and weather are the significant factors for the whole urban, low density and medium-high density area, while intersection type, accident type and time of day are the respectively significant factors for the three types of areas. [ABSTRACT FROM AUTHOR]

【Objective】Through the effects of functional traits of artificial Caragana korshinskii forest on the productivity in the windy-sandy region of northwest Shanxi province, the mechanism of ecological function degradation of artificial C.korshinskii forest was revealed in order to provide theoretical guidance for the sustainability of vegetation management and ecological security in the region.【Method】The effects of plant functional traits on productivity of artificial C.korshinskii forest stand of different ages were studied by general regression model and random forest regression model.【Result】（i）With the increase of stand age, the productivity and aboveground biomass of artificial C.korshinskii forest indicated a trend of increasing first and then decreasing, and the degradation occurred after 18 years;（ii）Artificial C.korshinskii forest stem-leaf ratio（R²=0.277, P=0.000）, branch diameter（R²=0.404, P=0.000）, branch number（R²=0.269, P=0.000）, plant height（R²=0.395, P=0.000）, longitudinal cross-sectional area（R²=0.211, P=0.001）, leaf area（R²=0.314, P=0.000）, leaf carbon content（R²=0.284, P=0.000）, leaf dry weight（R² =0.233, P=0.001）, branch dry weight（R²=0.211, P=0.001） were all extremely significantly positive correlated with productivity.（iii）Stem-leaf ratio（R²=0.287, P=0.000）, longitudinal cross-sectional area（R²=0.318, P=0.000）, leaf dry weight（R²=0.366, P=0.000） and branch dry weight（R²=0.992, P=0.000） were significantly correlated with aboveground biomass, but there was no significant correlation between other functional traits and aboveground biomass（P＞0.05）.（iv）The results of the random forest regression model indicated that stem-leaf ratio（11.59）, branch diameter（9.82）, and branch dry weight（9.18） were the main functional traits that affected its productivity. The branch dry weight（22.59）,leaf dry weight（12.87） and longitudinal cross-sectional area（4.58） were the main functional traits that affected aboveground biomass.【Conclusion】Functional traits related to shoots were one of the main mechanisms affecting the productivity and aboveground biomass of C.korshinskii forest in windy-sandy region of northwest Shanxi province, while the effects of leaf traits were relatively weak. [ABSTRACT FROM AUTHOR]

Relying on the Hongshengsha-Yufengwei Section Tunnel Project of the second phase of Guangzhou Metro Line 7, this paper used the particle swarm optimization random forest method (PSO-RF) to establish a sealed cabin pressure prediction model in the dual-mode shield soil pressure boring mode. Through the correlation analysis of shield boring parameters, the tunneling parameters that had a great influence on the pressure of the sealing chamber were screened out, including screw speed, screw torque, cutterhead speed, propulsion speed, penetration degree, cutterhead torque, and total thrust, and the screened tunneling parameters were used as the input parameters of the prediction model, and the pressure of the sealing chamber was used as the output parameter of the model to predict the pressure of the sealing chamber. The results showed that the random forest PSO-RF prediction model improved by particle swarm algorithm could effectively predict the pressure of the dual-mode shield sealed cabin, and compared with the traditional neural network prediction model, the PSO-RF model had higher prediction accuracy, the average absolute error was within 10%, and the predicted value and the actual goodness-of-fit R2 were 0.901 4, which was significantly superior to the BP neural network in terms of prediction accuracy and generalization ability of the model. [ABSTRACT FROM AUTHOR]

BACKGROUND: Selecting key technical indicators to guide special training or competition is the key to improving competition performance. There are obvious differences in the techniques of male and female athletes and the key technical indicators of athletes of different sexes have not yet been known. OBJECTIVE: To select key technical indicators that affect the competition performance of elite speed climbing athletes, thereby providing direction and theoretical support for the scientific training of speed climbing. METHODS: The videos of speed climbing finals in the 2017-2021 World Championships, World Cup and China Climbing League were selected to obtain the competition samples of the top four male and female athletes (109 men and 117 women). Reaction time, contact time and segmental speed were collected and then random forest and XGBoost models of male and female speed climbing athletes were established using two-dimensional kinematic analysis. Accordingly, the influence of each technical indicator on the competition performance was calculated. RESULTS AND CONCLUSION: For male athletes, the root-mean-square error values of random forest and XGBoost models were 0.224 and 0.265, respectively, and r2 was 0.765 and 0.686, respectively. The key technical indicators were total contact time of the left hand, motion frequency of the right hand, total contact time of the right hand, speed of holds 3-6, motion frequency of the left foot, total contact time of the right foot, speed of holds 11-13, speed of holds 13- 18, speed of holds 8-10, total contact time of the left foot, number of total holds of the left hand and motion frequency of the right foot. For female athletes, the root-mean-square error indicator of random forest and XGBoost models were 0.066 and 0.055, respectively, and r2 were 0.846 and 0.887, respectively. The key technical indicators were total contact time of the left hand, total contact time of the right hand, motion frequency of the right hand, motion frequency of the left foot, total contact time of the left foot, speed of holds 8-10, total contact time of the right foot, and speed of holds 10-11 and 13-18. To conclude, the technical indicators of the upper limbs (contact time sum and action frequency) on speed rock climbing competition are more influential than those of the lower limbs. The number of holds has no high influence on the competition results. The key segments for winning in male and female elite athletes are holds 3-6 and 8-11 respectively. Athletes should focus on optimizing the upper limb contact time of the key segments for winning and then optimize the lower limb contact time, so as to improve the competition performance. [ABSTRACT FROM AUTHOR]

Soil total nitrogen (STN) content can be accurately and rapidly estimated to better reflect the response relationship between spectrum and STN content. In this study, an unmanned aerial vehicle (UAV) equipped with a hyperspectral sensor was used to obtain the soil hyperspectral images in farmland. The original spectral reflectance (R) was then transformed into the reciprocal of reflectance (RR), logarithm of reflectance (LR), the first derivative of reflectance (FDR), the first derivative of reciprocal reflectance (FRR), and the first derivative of logarithm of reflectance (FLR). Grey correlation degree and Pearson correlation coefficient were also selected to extract the sensitive band of STN content in each spectrum. The hyperspectral inversion model of STN was finally constructed using the sensitive band, the partial least squares regression (PLSR), ridge regression (RR), and random forest regression (RF). The determination coefficient R², root mean square error RMSE, and mean absolute error MAE were used to evaluate the accuracy of the model. After that, the model with the highest accuracy was selected for the inversion mapping of STN content in the study area. The availability of the model was tested, according to the distribution of STN content. The optimal model was selected to invert and map the STN content. The results showed that: 1) The sensitive band (996-1 003 nm) of the RR spectrum was concentrated in the near-infrared long wave range, according to the gray correlation degree and Pearson correlation coefficient. The sensitive bands in the FRR spectrum (39-459, 469, and 472-1 003 nm), and FLR spectrum (398-459, 463-973, and 978-1 003 nm) were distributed in the visible and near-infrared range. The sensitive bands (615-625, 632, and 666-670 nm) in the FDR spectrum were concentrated mainly in the red range of visible light. 2) Pearson correlation coefficient was used to better reflect the response relationship between spectrum and STN content. The STN inversion model R², RMSE, and MAE were in the range of 0.058-0.693, 0.226-0.477, and 0.171-0.416 g/kg, respectively, in terms of Pearson correlation coefficient. In grey correlation degree, the STN inversion model R², RMSE, and MAE were within the range of 0.693-0.859, 0.123-0.276, and 0.107-0.209 g/kg, respectively. The accuracy of the model with the Pearson correlation coefficient was higher than that with the grey correlation degree, indicating that the Pearson correlation coefficient performed better on the response relationship between spectrum and STN content. 3) The RF-FDR model was used to estimate the STN content in the field. Among the 12 inversion models of STN content, the RF-FDR model shared the highest accuracy, with R² of 0.859, RMSE of 0.143 g/kg, and MAE of 0.114 g/kg. The inversion mapping of STN content showed that the STN content in most areas was in the range of 1.50-2.00 g/kg, according to the RF-FDR model. There was the consistency with the average value of STN content in the 72 soil samples, together with the actual situation of planting in one season, low soil fertility consumption, and annual fertilization. As such, the RF-FDR model can be expected to estimate the STN content in fields. Therefore, the Pearson correlation coefficient can be used to extract the sensitive bands for the soil spectrum from UAV hyperspectral. The inversion model of STN content can be constructed with higher accuracy for the effective estimation of STN content. [ABSTRACT FROM AUTHOR]

Soil management zoning can normally use weighted environmental variables to obtain the weight of environmental factors. This approach relies mainly on the crop yield to acquire the weights of environmental variables, in order to establish the relationship model between the crop yields and environmental variables. Thus, the current zoning cannot fully meet the requirement of modern agriculture in recent years, particularly when the crop yields are unavailable or there is no relationship between them. Fortunately, the normalized difference vegetation index (NDVI) can be expected to reflect the soil condition, which was significantly influenced by the environmental variables. In this study, an accurate and efficient approach was developed to obtain the weights of environmental variables using NDVI. The relation model between NDVI and environmental variables was then established in this case. Three steps consisted of this approach. Firstly, the key environmental variables were selected to significantly impose the influence on NDVI. The relation model was then established between NDVI and the selected key environmental variables. Finally, the weights of environmental variables were acquired from the established relation model. A case study was applied to test the effectiveness of the approach. Specifically, the weights of environmental variables were acquired for a rubber plantation in Danzhou County, Hainan Province, China. Among them, the candidates were taken as the 10 terrain attributes, 19 bioclimatic variables, and 5 parent materials types. Furthermore, 15 key environmental variables were determined to construct the relationship model with the NDVI using the random forests (RF). The importance index (IncMSE) was extracted from the fitted relation model, and then calculated the weight of each key environmental variable. These acquired weights were also used to weigh the environmental variables, and then to serve as the input variables for the K means clustering. Finally, six groups of soil management zones were generated for the rubber plantation. A total of 1296 top soil (0-20 cm) samples were employed to verify the soil management zones. Results indicated that the management zones better distinguished the abundance and deficiency levels of different soil nutrients (soil organic matter (SOM), total nitrogen (TN), available phosphorus (AP), and available potassium (AK)). Zone 1 presented the high AP content, the normal range of pH value and SOM content, as well as the TN and AK in deficiency. Zone 2 shared the high AP content, pH value suitable for the rubber tree growth, but the SOM, TN, and AK in deficiency. Zone 3 demonstrated a very high AP content and a pH value within the normal range. Nevertheless, the contents of SOM, TN, and AK were all lower than the normal range in Zone 3. More importantly, the contents of SOM, TN, and AK in Zone 4 were similar to those in Zone 3. However, the values of pH and AP were significantly lower than that in Zone 3. In Zone 5, the values of pH and AK were the highest in the normal range among the six zones, while the values of SOM and TN were the lowest in deficiency. In Zone 6, the soil TN content was the highest among the six zones within the normal range. The pH value and AP were also within the normal range, but the AP content was close to the lower limit of the normal range, indicating the insufficient content of SOM and AK. At the same time, the mean differences were all significant at P < 0.05 level in 15 selected environmental variables among different management zones. Consequently, the soil management zones further verified the effectiveness of the approach, where the weights of environmental variables were acquired by the NDVI. Anyway, the NDVI acquiring the weights of environmental variables can be applied to much wider ranges, because the NDVI can be much easier to access than the crop yields in practice. [ABSTRACT FROM AUTHOR]

Fishway is one of the common measures to maintain the continuity of rivers and protect fish resources in the world. Among them, the hydraulic characteristics of the early fishway cannot match the swimming ability and habits of the domestic fish, particularly in the early stage of construction. Fortunately, the internal structure can be optimized to improve the efficiency of fishways operation in recent years. However, the complicated physical experiment depends greatly on seasonal, temperature, light, and external factors. Additionally, the test cost is ever increasing in a large amount of fish and feeding. Therefore, it is a high demand to predict the fish migration routes in the field of fish conservation using mathematical models. Taking Schizothorax prenanti as the research object, this study aims to simulate fish migration in the vertical slot fishway. The original data set was divided using upstream data monitored in the previous test, according to the swimming characteristics of Schizothorax prenanti. Random Forest (RF), XGBoost, K-Nearest Neighbors (KNN), and Gradient Boosting Decision Tree (GBDT) were used to establish the multiple Behavior classification model (BCM), and Swimming speed regression model (SSRM). According to the evaluation indexes of the classification and regression model, the optimal model was selected to establish the mapping relationship between the fish behaviors and hydraulic indexes using the Eulerian-Lagrangian-agent method (ELAM) framework, and then the fish migration model was constructed. The improved model also considered the influence of velocity (magnitude, vector, and velocity gradient) and inertia factors on the movement of fish. Specifically, the perception area of the fish body was represented by a two-dimensional circular plane with the inductive distance as the radius, particularly with the center of the fish body as the origin. Finally, the simulated and the observed track were compared to verify the reliability of the fish migration model. The results showed that the RF model performed the best to predict the swimming behavior classification and swimming speed regression. The classification accuracy of swimming behavior was 0.804, the recall rate was 0.794, and the F1 score was 0.798. The R2 values were all greater than 0.75 for the swimming speed regression using the three data sets. The predicted trajectories of the migration model were better agreed with the actual. Most virtual fish successively passed through the mainstream and the right backflow area, indicating better consistency with the actual fish passing experiment. A comparison was made to obtain the characteristic track of the numerical simulation and the actual fish test. Consequently, the fish migration model can be expected to fully reflect the basic behavior characteristics of the target fish, and then to better predict the migration route of Schizothorax prenanti. The finding can also provide a strong reference to evaluate the fishway flow field for the protection measures in the fish industry. [ABSTRACT FROM AUTHOR]

Soil organic matter (SOM) is one of the most important parts of the soil carbon pool. The carbon-containing organic matter in soil mainly includes animal and plant residues, microorganisms, and various organic matter decomposed or synthesized in agriculture. Among them, the SOM content is one of the important indicators to measure the soil fertility level. Accurate measurement of SOM is of great significance for soil fertility evaluation, environmental protection, agricultural and forestry development. Therefore, accurate prediction of SOM content is extremely important so far. Previous research on the SOM prediction of random forest (RF) usually only uses one spectral input without considering the complementarity between different spectral inputs. Therefore, it is necessary to select an appropriate method for the noise reduction of the reflection spectrum, in order to reduce the influence of spectral noise. Among them, discrete wavelet transform can be used to reduce high spectral noise, while preserving the effective information for the SOM prediction. In this study, the combination of different spectral inputs and discrete wavelet transform was used to predict the SOM with the optimized spectral input using RF. The stochastic forest model was also used to predict the SOM. Firstly, the original spectral reflectance of 204 soil samples from Baoqing County was analyzed using discrete wavelet transform. Secondly, the spectral characteristic parameters and principal components were extracted from the decomposed characteristic spectral curves, in order to construct the spectral indices. Finally, the three spectral inputs were substituted into the RF model to explore the optimal combination of spectral inputs for the SOM prediction. Meanwhile, the variation trend of different spectral inputs was obtained under different wavelet decomposition scales, in order to provide a new idea for the selection of spectral inputs for the SOM hyperspectral prediction. The RF model was better to predict the SOM in this case. The optimal combination of different spectral information was obtained to predict the organic matter and the optimal decomposition scale of the discrete wavelet transform. Finally, the combination with the highest accuracy was obtained among all the inputs at all decomposition scales. The results show that: 1) The accuracy of SOM prediction under different spectral inputs was higher than that of direct spectral reflectance modeling. The highest verification accuracy of the principal component in the single spectral index, similar to the combination of spectral characteristic parameters and principal component, was higher than that of the principal component modeling alone, indicating that combining different spectral inputs improved the prediction accuracy. However, simply stacking spectral inputs was not enough to improve the prediction accuracy. 2) There was also a different variation trend of prediction accuracy of different spectral inputs, with the increase of decomposition scale. The variation trend of prediction accuracy of different spectral input combinations was changed with the different spectral inputs in the combination, indicating the variation characteristics of spectral inputs. 3) The highest verification accuracy was found in the combination of the spectral characteristic parameters and principal components with the decomposition scale of 6, R² reaching 0.78, and RMSE reaching 1.32%, indicating an excellent prediction ability. Anyway, it is feasible to predict the organic matter using the spectral input combined with discrete wavelet transform modeling. The finding can provide a reliable idea and theoretical support for the dynamic monitoring of SOM under temporal and spatial changes. [ABSTRACT FROM AUTHOR]

To reveal the spatial distribution of flue-cured tobacco, 11 Sentinel-1 images obtained in the growing season of flue-cured tobacco in Yuxi City, Yunnan Province in 2020 were utilized in combination with field survey data. First, the polarization of the synthetic aperture radar (SAR) with different land use types were compared. Next, the random forest model was trained with 70% field survey data to make a land-use map. Finally, the identification accuracy of the flue-cured tobacco was validated with 30% field survey data. The results indicated that vertical transmit-vertical receive（VV) polarization distinguished land-use types better than the vertical transmit-horizontal receive (VH) polarization. In the four forms of polarization (i.e., VH, VV, VV+VH, and VV+VH+VV/VH), VV+VH produced the most accurate land-use map with a total accuracy of 87.6% and a Kappa coefficient of 0.847. Specifically, the producer accuracy of the fluecured tobacco was 96.3%, and the user accuracy was 89.4%. We conclude that the identification accuracy of the flue-cured tobacco distribution based on the multi-temporal SAR images could fulfill regional application needs. [ABSTRACT FROM AUTHOR]

Aiming at the problems of various forms of fraud in the industry, hidden operations, and extremely unbalanced data distribution, this paper adopts the ADASYN algorithm to adaptively move the classification decision boundary to difficult instances to achieve data augmentation, to solve the over-fitting problem caused by unbalanced data. The Sequence Forward Search Strategy algorithm based on the Random Forest is used to filter out the optimal feature subset to detect fraud, reduce the impact of noise data added by the ADASYN algorithm on the determination of classification boundary, construct a fraud detection model, and use LIME to make local interpretation of the model detection results to improve the use of the model. The experiments show that the model can better overcome the defects of traditional fraud detection models in misclassifying most classes of samples, and help to improve the efficiency of transaction fraud identification in the industry. At the same time, the random samples detected by the model are effectively analyzed through LIME, which is convenient for decision-makers to make empirical analyses on the detection results of the algorithm model and plays an obvious early warning and decision-making reference value. [ABSTRACT FROM AUTHOR]

BACKGROUND: Radiomics can be used to quantify image heterogeneity. Whether radiomics can be used to screen out features such as fingerprint from MRI images of osteoporotic vertebral bodies to predict the occurrence of new fracture is worth studying. OBJECTIVE: To explore the feasibility of constructing a machine learning prediction model for thoracolumbar refracture after vertebral augmentation through combining MRI radiomics features and clinical information. METHODS: This study retrospectively collected the data of patients who were diagnosed with osteoporotic vertebral compression fracture by MRI and treated with percutaneous vertebral augmentation in Chengdu First People’s Hospital from May 2014 to April 2019. PyRadiomics was used to extract the imaging features of T1 sequences of vertebral MRI at the T11-L2 segments before percutaneous vertebral augmentation. All models were constructed in the training set, and prediction performance evaluation was performed in the validation set. Feature dimension reduction was conducted by applying least absolute shrinkage and selection operator regression. The corresponding refracture prediction models were constructed by multivariate logistic regression, random forest and adaptive lifting algorithm analysis using clinical parameters, selected features or the integrating of both. The diagnostic efficacy of the model was evaluated using the receiver operating characteristic curve. The decision analysis curve was used to compare the clinical value of each model. RESULTS AND CONCLUSION: (1) A total of 336 vertebrae were included in 135 patients, of which 67 vertebrae had refractures. 1 746 features were extracted from each vertebra, and 13 important features were obtained through dimension reduction. (2) Among the three models, area under curve of the combined model in the training set and validation set was significantly higher than that of the clinical model (P < 0.05), and the decision analysis curve also showed that the net benefit of the combined model in predicting thoracolumbar refracture was higher than that of the clinical model in most threshold intervals. (3) The results indicated that it was feasible to construct a refracture prediction model based on MRI T1 sequence imaging and clinical information, which could help to identify the vertebrae with high risk of refracture at early stage. [ABSTRACT FROM AUTHOR]

Broken-winged chicken carcasses can be one of the most common defects in broiler slaughter plants. Manual detection cannot fully meet the large-scale production, due to the high labor intensity with the low efficiency and accuracy. Therefore, it is a high demand to rapidly and accurately detect broken wings on chicken carcasses. This study aims to realize the rapid inspection of broken-winged chicken carcasses in the progress of broiler slaughter, in order to improve the production efficiency for the cost-saving slaughter line. 1053 broiler carcass images were collected from a broiler slaughter line using a computer vision system. Rapid identification was then constructed for the broken wing defects. Specifically, the front view of the chicken carcass was obtained in the machine vision system. The preprocessing was then deployed to obtain the chicken carcass images without the background, including the weighted average (graying), two-dimensional median filtering (denoising), and iterative (threshold segmentation). The code was also written in the MATLAB platform. After that, a total of 11 characteristic values were calculated, covering the exact distance starting from the left and right ends of the chicken carcass image to the centroid and the difference (d1, d2, and dc), the heights of the lowest point in the two wings and their difference (h1, h2, and hc), the areas of the two wings and ratio of them (S1, S2, and Sr), squareness (R), and width-length ratio (rate). As such, the eight principal components were achieved in the principal component analysis after the reduction of several dimensions. Separately, the principal components and characteristic values were imported into the specific model of linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), random forest (RF), support vector machine (SVM), and BP neural network. Among them, the input parameter of the VGG16 model was from the RGB maps of the chicken carcass with the removed background. Finally, a comparison was made for the F1-scores and total accuracy of each model. Thus, the highest recall rate of 91.13% was achieved in the RF model with the characteristic values as the input parameters among the shallow learning models. The second higher recognition accuracy and F1-score were 96.28% and 92.58%, respectively in the SVM model with the characteristic values as input parameters. The highest total accuracy of model recognition was achieved in the quadratic discriminant and SVM models with the characteristic values as the input parameters, both up to a proportion of 91.78%. Moreover, the F-score and total accuracy of the VGG16 model were the highest among the total model combinations, with respective rates of 94.35% and 93.28%, respectively. In terms of the prediction time of models, the shortest prediction time was obtained in the SVM model with the principal components as the input parameters. Specifically, the capacity was found to determine 353 sample images in 0.000 9 s, with an average speed of 3.92×105 images per second. By contrast, the longest prediction time was observed in the VGG16 model, where 24.46 s to determine 253 sample images, with an average speed of 10.34 images per second. In conclusion, the VGG16 model can be expected to serve as the best classification of broken wings in chicken carcasses. [ABSTRACT FROM AUTHOR]