Search

Autonomous navigation is an important guarantee for intelligent agricultural machinery to complete harvesting operations. Navigation line recognition is an important step in achieving autonomous navigation of intelligent agricultural machinery. Real-time and accurate detection of harvesting navigation lines can effectively accelerate the progress of operations and reduce crop loss rates. Visual sensors have the advantages of low cost and rich information acquisition, making them widely used in route detection. Frequent changes in lighting under cloudy weather make it difficult for a single feature to cope with wheat field environments. In this study, a wheat harvesting boundary detection method based on multi texture feature fusion was proposed. In the harvested area, the characteristics of low stubble and no top wheat ears result in differences in uniformity, density, and other aspects when harvesting under both light and backlight compared to the non-harvested area. At the same time, due to the exposed wheat stubble in the harvested area, it appears more regular on the image, and its gradient direction is more consistent than that in the non-harvested area. Therefore, a two-dimensional feature vector composed of image entropy and directional gradient was constructed to classify the harvested and unharvested areas of the wheat field, and then the harvesting boundary was extracted. Due to the susceptibility of image entropy features to perspective phenomena, this paper compared the entropy features of different regions in the image when fixing the camera installation angle and height, and ultimately selected the middle region of the image as the region of interest. After analyzing the characteristics of image entropy feature extraction, a histogram statistical method based on sliding windows was proposed to accelerate the speed of image entropy feature extraction. By further dividing the window to be calculated into several sub windows, sliding and combining these sub windows significantly reduced the computation. The entropy feature extraction in this paper took 0.53 seconds, which is 49.52% faster than the traditional method of directly extracting entropy from the entire window. After extracting twodimensional features, based on the distribution characteristics of the feature histogram and combined with the maximum entropy threshold segmentation algorithm, the wheat field image was preliminarily classified. For entropy features, low noise points affected the threshold segmentation effect. By comparing the impact of discarding different proportions of low entropy points on the segmentation threshold, this paper discarded 10% of low entropy points, enhancing the threshold segmentation effect. The method of removing small connected regions was used to eliminate the misclassified regions in the initial classification binary image. Then, the Canny operator was used to extract edge contour points distributed near the harvesting boundary. Due to the consistent direction between the harvesting boundary and the agricultural machinery's forward direction, the Ransac algorithm was used to restrict the fitting line in the region, obtaining the accurate harvesting boundary. In order to verify the feasibility of the method proposed in this article, wheat field images under different lighting conditions were collected, including 2 200 weak light images, 758 local strong light images, and 1 134 strong light images. After selecting the region of interest, the size of the image to be processed was 240 pixels×1 280 pixels. The algorithm in this paper took an average of 0.88 s to process each image on laptops. The detection rates under weak light, local strong light, and strong light were 90.41%, 88.26%, and 89.68%, respectively, with an average of 89.45%. Compared with the traditional method of using Adaboost algorithm for machine learning, the detection speed of this algorithm was improved by 73.89%, and the detection rate was increased by 46.19 percentage points, 46.00 percentage points, and 49.64 percentage points under weak light, local strong light, and strong light, respectively. Compared with traditional algorithms, the method proposed in this paper significantly improved real-time performance and detection rate, which can provide reference for the detection of field agricultural machinery navigation lines. [ABSTRACT FROM AUTHOR]

Fruit picking has been one of the most important operations in modern agriculture. Manual picking cannot fully meet large-scale apple production in recent years, due to the labor-intensive and time-consuming task. Alternatively, robot picking can be expected to replace manual work. However, the conventional semi-automatic picking machine can seriously vibrate the branches and damage the fresh apples during batch harvesting in the existing orchard. It is a high demand to detect apples under complex environments, and then pick them without damaging the fruit or trees. The picking robots are also required a compatible auxiliary platform during operations in an orchard. The performance of picking robots can be assessed using the picking speed, picking success, and fruit damage rates. These parameters depend on three key technologies: target recognition, localization, and fruit separation. In this paper, an overview was provided of the history and current status of commercial apple-picking robots. Specifically, the performance and robustness were highlighted, in terms of the picking efficiency, structural composition, the type of orchard, and adaptability to the external environment. The main technical features were then summarized from the application of apple-picking robots. Since the apple picking was operated in an unstructured orchard environment, the most serious challenge remained in the application of target recognition, localization, and fruit separation techniques. Therefore, the recognition techniques were reviewed for the apples in the complex natural environments with changing light, branch and leaf shading, overlapping fruit, night-time environments, and same-colored apples. Deep learning-based apple recognition algorithms were the far-reaching prospects in complex environments, particularly in multiple scenarios. The accuracy and speed of apple recognition were improved using deep learning recognition algorithms in the field of apple picking. The second most challenge was the application of the apple localization for the fruit occlusion and overlap, while the fruit displacement by wind during the imaging process. Therefore, four aspects were introduced in the apple localization, including 2D information acquisition, depth information acquisition, apple pose acquisition and oscillating fruit localization. Point cloud processing algorithms were highlighted to achieve the precise localization of occluded apples. In addition, the applications of end-effectors were reviewed in the apple-picking robots with high efficiency and non-destructive picking. Three types of the end-effector structure (such as hand-jaw, vacuum, and shear) were summarized under fruit separation. It was found that the hand-jaw was the most commonly-used type of end-effector. Current research on the end-effectors was focused on the flexible materials for the finger surfaces, under-driven, flexible structures or control systems, such as the soft hands and perception-based servo control. Therefore, it is necessary to improve the picking speed and success rate with the low fruit damage and cost saving in the apple picking robot, in order to accelerate the process of apple picking automation. Future development was also proposed for the apple-picking robot in the urgent commercial need of the pioneering innovation in the combination of agricultural machinery and agronomy. Some optimization can be expected to be performed on the recognition algorithm, multi-sensor fusion, multi-arm cooperation, human-machine collaboration, extending device versatility, and integrating 5G and IoT technology. This finding can provide a strong reference for the current status, future development trends, and equipment innovation of apple-picking robots, together with the key technologies. [ABSTRACT FROM AUTHOR]

Agriculture is crucial for economic development. Losses in agricultural production were caused by a variety of circumstances, but pests and illnesses are among the most important. The use of plant protection machinery to spray pesticides is still the most effective method of pest control at present. Traditional agricultural sprayers were inefficient and time-consuming, which caused pesticide waste and environmental contamination. In the process of spraying pesticides, farmers are directly exposed to pesticides, which is easy to cause bodily harm or even poisoning. With the steady progress of China's agricultural modernization, robots are widely used in agricultural plant protection work. Intelligent pesticide spraying robots are an effective method to solve the problem of applying pesticides in complex environments such as hills, orchards, fields, and greenhouses, which is more efficient and accurate. Intelligent spraying robot not only saves labor but also reduces the improper use of resources and environmental pollution. It is conducive to our better reaching of the goal of agricultural sustainable development. In this study, the research progress analysis of an intelligent pesticide spraying robot focuses on four respects: mobile chassis design, spraying the device design, navigation technology, and target detection technology. This paper discusses the factors limiting the development of intelligent pesticide-spraying robots for the primary technologies applied to various operation scenarios. Firstly, the mobile chassis design of the spraying robot was analyzed from the ground spraying robot and aerial spraying robot. The study and development of wheel-track composite robot chassis can integrate the advantages of wheeled chassis as well as a tracked chassis, which will be the research trend of ground intelligent spraying robot. Plant protection UAV is unrestricted by terrain, and can be applied to a variety of scenarios, which will be the key orientations for future research. Secondly, the spray devices required by the spraying robot in different spraying scenarios were summarized. It mainly includes an adjustable boom sprayer device, electrostatic spraying device, targeted spraying device, air-assisted spraying equipment, and profile modeling spray equipment. The most commonly used navigation methods are analyzed, including Global Positioning System (GPS) navigation, visual navigation, inertial navigation, multi-sensor fusion navigation, laser radar navigation, odometer, geomagnetic navigation, etc. Additionally, the detection methods of various pesticide spraying robots are evaluated from the perspectives of crop detection and pest detection. The deep learning algorithm can automatically extract the image features of crops, make use of the information, and exactly identify the location of crop pests and diseases, to quickly respond to achieve accurate spraying. Finally, the development trend of intelligent pesticide application robots for precise application is presented. Six requirements or prospects were proposed for the intelligent pesticide spraying robot development in the future, including to improve the response speed of the spraying robot, to improve detection accuracy, to design lightweight chassis, to reduce the development cost of the spraying robot, and to integrated air-ground spraying technology. In conclusion, intelligent and precise application robots have the potential to significantly improve the quality of precision agriculture operations by facilitating intelligent decision-making on operational parameters before actual operation. [ABSTRACT FROM AUTHOR]

Soluble solid content (SSC) has been widely used to realize the prediction of apple grades in recent years. It is very necessary to establish a non-destructive prediction model for the mapping from the near-infrared spectra to the SSC. However, the single classification model has seriously limited the prediction accuracy and application, particularly with the hard partition of the instance space. In this study, an evidence theory-based multi-model fusion was proposed to deal with this issue. The evidence theory provides a flexible framework to represent and reason with uncertainty. After that, the evidence theory was adopted to aggregate the predictions of the two models. Moreover, the mass function generation was also treated for the fusion model. The classification knowledge was then better provided by the SSC, Therefore, a much more accurate classification was achieved using the improved model. Some processes were firstly carried out to improve the prediction performance, including the information collection, data pre-processing, and spectral feature selection. Specifically, the near-infrared spectra of apple instances were collected by the self-developed WY-6100 fruit online nondestructive testing system. Meanwhile, the SSC was measured using physicochemical technologies. 439 Red Fuji apple instances were collected from Yantai City, Shandong Province, China. The training and test set were then randomly divided by the 7:3 ratio. Specifically, the 307 randomly-selected instances were used for the model training, whereas, the rest 132 instances were used for the performance testing. A variety of procedures were selected to pre-treat the near-infrared spectra, in order to achieve the inputs of instances. Some anomalous instances were also eliminated using Principal Component Analysis-Mahalanobis distance (PCA-MD). A Savitzky-Golay convolutional smoothing filtering was used to remove the noise caused by the equipment, environment, and external factors. The degree of baseline drift was reduced by the standard normal variable transformations. The characteristic wavelength was then extracted from the spectrum by the genetic algorithm (GA), in order to preserve the most useful information. Partial least squares (PLS) and Extreme learning machine (ELM) models were also established in this paper, with classification accuracy of 93.80% and 92.25% respectively, as shown by the experimental data. Two prediction models were fused by the Evidence theory. The uncertainty of each prediction model needed to be quantified with a mass function, particularly for the higher classification performance of the fusion model. A triangular mass function generation was proposed to balance the distance between the predicted value of SSC and the classification boundary. The misclassifications were attributed to the apple instances, where the predicted values of SSC fell within the area near the boundary of apple classification. Thus, a novel strategy was used to assign the value of masses into a precise class, where this class was set to contain the adjacent one. Once the function was generated, the mass functions of ELM and PLS models were combined to obtain the fusion prediction using Dempster's combination. The focal element with the maximum mass value in the combined mass function was selected as the final decision of apple grades. Finally, the experimental results showed that the triangular mass function generation was more reasonable than that using the hard partition. The classification accuracy of the multi-model based fusion reached 95.35%. In summary, the mass function generation was modified to more precisely depict the Evidence theory-based fusion model for the uncertain classification information. Therefore, a much more accurate and intuitive classification was achieved during this time. Moreover, the improvement can also be applied to the fusion of other prediction models. The better applicability was also obtained without the hard partition, compared with the single prediction model. [ABSTRACT FROM AUTHOR]

With the advancement of remote sensing technology, the high spatial resolution remote sensing image contains rich special information with a great detail. At the same time, the complexity of high spatial resolution remote sensing images also requires higher the classification technology of remote sensing images. However, in the face of high spatial resolution remote sensing image more obvious geometrical structure and the more rich texture characteristics, how to design rational system of characteristics, select the appropriate sorting algorithms to accurately and quickly grasp the number of rural land of building and its distribution status, are of great significance to balance urban and rural areas, save land, and realize sustainable development. This will help in exploring the application of deep learning model in high spatial resolution remote sensing image building extraction, and have research significance for improving the classification accuracy of high resolution remote sensing image. In this paper, the semantic segmentation model (SegNet) was used for extracting buildings. SegNet is mainly composed of encoder network, decoder network and pixel-wise classification layer. The encoder network transforms high-dimensional vectors into low-dimensional vectors, enabling low-dimensional extraction of high-dimensional features. The decoder network maps low-resolution feature maps to high spatial resolution feature maps, realizing the reconstruction of low-dimensional vectors to high-dimensional vectors. The softmax classifier separately classifies each pixel, which outputs the probability that each pixel belongs to each class. In this paper, a 3000 pixel × 3000 pixel and two 2000 pixel × 2000 pixel slices were taken from the global remote sensing image of Bazhou City, Hebei Province as training samples, and a 3000 pixel × 3000 pixel slice was taken as the verification sample. In this paper, five comparative experiments were used to extract the buildings, including PSPNet, support vector machine, random forest, ISO clustering and maximum likelihood method. The confusion matrix of each classification method was obtained by calculating the difference between the classification results of the comparison experiment and the real value. From the traditional classification algorithm to the shallow learning algorithm to the deep learning algorithm, the Kappa coefficient and overall accuracy of classification kept constantly increasing, among which SegNet semantic segmentation algorithm based on the deep convolutional network performed better than the other five algorithms in extracting buildings from high spatial resolution remote sensing image. The Kappa coefficient and the overall accuracy of SegNet semantic segmentation algorithm were 0.90 and 96.61%, respectively, and the ground truth value was basically the same as the classification result. The F1Score of building extraction of SegNet semantic segmentation algorithm based on deep convolution network was 0.91, but the other five algorithms were below 0.87. SegNet had the lowest error rate of 9.71% for buildings, indicating that the ability to identify buildings of semantic segmentation algorithm from high spatial resolution remote sensing was superior to traditional classification algorithms, shallow layer learning algorithms based on machine learning, and PSPNet semantic segmentation algorithm based on deep convolution network. The Kappa coefficient and overall accuracy of the remaining five classification algorithms were respectively below 0.83 and 94.68%, and the difference between the ground truth value and the classification result was relatively large. SegNet can not only make use of spectral information but also make full use of abundant spatial information. During SegNet training, more essential features can be learned, and more ideal features suitable for pattern classification were finally formed, which can enhance the ability of convergence and generalization of the model and improve the classification accuracy. Traditional classification algorithms, such as ISO clustering and maximum likelihood method, failed to make use of the rich spatial information of the high-resolution remote sensing image, so the accuracy was relatively low. Due to limited computing units and large amount of high spatial resolution remote sensing image data, shallow layer learning algorithms based on machine learning such as support vector machines and random forest cannot effectively express complex features of ground objects, so their advantages are not obvious in building extraction from the high spatial resolution remote sensing images. The experimental results showed that the SegNet based on deep learning has the best performance, and it has important theoretical significance to explore the application of deep learning model to remote sensing image classification methods. At the same time, the research results also provide a reference for improving the classification accuracy of high resolution remote sensing images. [ABSTRACT FROM AUTHOR]

Northern Xinjiang is a significant base of agricultural production in China. Finding out the cropping structure of main crops is important for food security. Accurate acquisition of field crop planting information is the basis of agricultural production and the key to accurate estimation of yield. This paper selected some farmlands of the 8th Division of Xinjiang Production and Construction Corps as the different experimental area and verification area. In August 2017, the CW-20 fixed-wing drone was equipped with a Parrot Sequoia camera, which captured the visible light image. The red, green and blue bands were employed to color space conversion and texture filtering, the texture and structural features of the image were deeply explored, and the automatic extraction of crop planting information was realized. Firstly, color space transformation and gray level co-occurrence matrix texture filtering were used to obtain 27 texture features of hue, saturation, and brightness. Secondly, classification weights and correlation coefficients of 24 texture features and 3 low-pass features were calculated by ReliefF-Pearson feature reduction method. The redundant features with weak classification ability and high correlation were rejected. Hue convolution low pass, hue entropy, intensity correlation, hue homogeneity, intensity convolution low pass, intensity entropy, intensity homogeneity, intensity dissimilarity, saturation convolution low pass, saturation correlation, and intensity variance were the optimal features for visible waveband image classification. Based on the optimal classification features, K-nearest neighbor (KNN), support vector machine (SVM), Naive Bayes (NB), artificial neural network (ANN), and decision tree (C5.0) classification algorithms were used to train model parameters and assess the classification effect. The overall classification of all methods is quite satisfactory, and the classification accuracy rate is over 80%. Among them, the SVM method has the highest classification accuracy, and the accuracy of test area classification is 83.77%, followed by the ANN and KNN methods. Finally, the SVM, KNN, and ANN classification methods were used to perform pixel-level supervised classification of images in the test and verification areas, and the classification effects were evaluated based on visual interpretation maps. The results show that the SVM, KNN, and ANN classification methods are satisfactory in the classification of crop category, and the boundary of the field block is relatively complete, but the classification effect of the non-crop category is unsatisfactory. From the confusion matrix, the SVM method has the highest classification accuracy. Its overall accuracy is 80.74% and Kappa coefficient is 0.75. The classification accuracies of cotton, corn, marrow, and alfalfa crops are 82.78%, 85.49%, 92.65% and 80.84%, respectively, while that of the trees and other categories are only 64.58% and 59.28%, respectively. The overall accuracies of KNN and ANN methods are above 74%. In summary, based on the optimal color-texture features selected by ReliefF-Pearson, the SVM classification method is the most stable and reliable in the crop category classification and can be utilized to classify the main crops in northern Xinjiang. However, all classification methods are built on the pixel level. There are some differences in the canopy structure of the same crop, and some internal areas may be misclassified as others. Using object-oriented segmentation can improve the classification accuracy. In addition, there are differences in the growth period of different kinds of crops, and the difference in the optimal identification time differs greatly in the accuracy of each category. Using crop phenology information combined with the time-series images can get the expected classification effect. The method of this paper has a certain applicability for the UAV (unmanned aerial vehicle) visible waveband images to obtain planting information of the main crops in northern Xinjiang and provides a methodological reference for a large-scale crop planting information survey. [ABSTRACT FROM AUTHOR]

Temperature and humidity prediction plays an important role in various fields of the national economy. In view of the poor predictions of temperature and humidity and the failure to achieve multimodal prediction, this paper uses the Gaussian mixture process (GPM) model to predict the multimodal temperature and humidity. GPM model was developed from the Gaussian process (GP) model. It was proposed by Tresp in 2000. The main idea of GPM is to fit the samples in various input regions with different GP models in order to fit different patterns in these regions. However, the existing GPM models are rather complicated and thus the learning algorithms involve various approximation schemes, which can lead to much computation or rough prediction. So, in order to improve the model learning efficiency, a variable-hidden posterior hard-cut iterative training algorithm is proposed for the GPM model. This algorithm improves the EM (expectation maximization) learning with a new approximation strategy, which achieves the optimal grouping of samples through iterative learning and using the maximum a posteriori estimation. Prior to the experiment, the autocorrelation function and partial autocorrelation function were used to verify the nonlinear and non-stationary property of humidity and temperature sequences. Largest Lyapunov, saturation correlation dimension and recursive graph were used to verify the chaotic characteristic of humidity and temperature sequences. Finally, the GPM proposed in this paper is compared with the kernel regression (K-R), the minimax probability machine regression (MPMR), the linear regression (L-R), and GP models. K-R is a kind of prediction model based on kernel function. By adjusting the optimal window width, K-R gradually gets the corresponding prediction result. K-R has been the basis of predictive model because it can predict the linear and non-linear time series simultaneously. MPMR does not need to make specific assumptions about the model distribution. It only needs to know the mean and covariance matrix of the data distribution. This model improves the neural network easy to fall into the minimum and over-fitting. L-R is a very traditional model, using the quantitative relationship between variables to obtain a linear expression and then predict the test sample. The results show that not only GPM can perform multimodal prediction, but also the accuracy of the model is significantly better than other traditional models. In the humidity sequence prediction, the minimum of RMSE can reach 0.0620 and the maximum of R2 can reach 0.9362. In the temperature sequence prediction, the minimum of RMSE can reach 0.0426 and the maximum of R2 can reach 0.9666. In the prediction time consumed, GPM is a little more than the other 4 traditional models. The main reason is that GPM requires iterative learning to achieve the optimal grouping of samples. Finally, humidity prediction needs 113.4175 s and temperature prediction needs 90.0049 s. The research in this paper not only can promote atmospheric temperature and humidity forecasting, but also has some reference value for indoor and solid surface temperature and humidity forecasting because GPM is the prediction model without environment factors input. [ABSTRACT FROM AUTHOR]

The rapid acquisition of soil water content (SWC) in field crop root zone is significant for drought supervision and precision irrigation. The UAV multispectral remote sensing system has the advantages of obtaining high spatial-temporal resolution of crop phenotype data, and has a wide application prospect in soil moisture monitoring. In order to obtain SWC accurately and timely at a farm scale, in this paper, the field maize with different water treatments is taken as the research object, and the multispectral remote sensing monitoring of summer maize is carried out by using the UAV remote sensing platform, and the soil water content of different soil depth in maize root zone is collected synchronously. Based on the UAV multispectral remote sensing image data sets of jointing stage, tasseling-silking stage and milky-maturity stage of summer maize in 2018, the soil background is removed by support vector machine classification, the spectral reflection of maize canopy is extracted, and the 10 vegetation indices are calculated, then the sensitivity analysis of soil water content in different depth is carried out by using full subset screening method for different wave bands and vegetation indices, and the soil water content in different depth is analyzed respectively, ridge regression and extreme learning machine are used to construct quantitative estimation models of soil water content at 0-20, 20-45 and 45-60 cm soil depth after full subset selection. The test area is located in Zhaojun Town, Dalate Banner, Ordos, Inner Mongolia, China(40°26'0.29" N, 109°36'25.99" E, elevation 1 010 m). The sowing time of maize is on May 11, 2018, the emergence time is on May 18, and the harvest time is on September 8, 2018. The total growth period is 114 days. The UAV multispectral remote sensing images and ground data collection dates are July 8, July 12, July 17, July 21, July 26, August 2, August 28 and September 7, 2018. It is collected once a day and tested 8 times in the whole growth period. July 8-21 is the jointing stage, July 26-August 2 is the tasseling-silking stage, August 28-september 7 is the milk-maturity stage. The flight altitude of the UAV is 70 m, and the flight time is 11:00-13:00 local time (11:44-13:44 Beijing time). Firstly, the UAV multispectral canopy images of field maize with 5 different irrigation treatments (TRTs) are acquired through the six-rotor UAV equipped with a RedEdge multispectral camera ( MicaSense, USA), and the multispectral images of diffuse reflector (reflectivity 58%, size 3×3 m) are collected at the same height to perform radiometric correction in the meantime, and then the spectral reflectances of the field maize are acquired. Secondly, the support vector machine (SVM) is used to eliminate the multispectral image of soil background in ENVI and ArcGIS software, then the maize canopy spectral reflectance is extracted and 10 vegetation indices (VIs), such as Normalized Difference Vegetation Index (NDVI), Normalized Green Difference Vegetation Index (GNDVI) and Transformed Chlorophyll Absorption In Reflectance Index(TCARI), etc, are calculated. Finally, the full subset selection method based on Bayesian Information Criterion (BIC) is used to analyze the sensitivity of SWC at different depths for different spectra bands and vegetation indices in R3.5.1 software, and then Ridge Regression (RR) as well as Extreme Learning Machine (ELM) are used to construct a quantitative estimation model of SWC at soil depths of 0-20, 20-45 and 45-60 cm at different growth stages, respectively. The results show that the full subset selection method based on BIC can effectively select the optimal spectral subset, and the selected variables generally pass the significance test and the independent variables number is small; the effects of the ELM model outperformed the RR model almost under all the same conditions; the optimal monitoring soil depth of maize at jointing stage, tasseling-silking stage is 0-20 cm, and the optimal monitoring soil depth of milk-maturity stage is 20-45 cm; the ELM inversion model at 20-45 cm soil depth at milk-maturity stage has the best effect, the decision coefficients of modeling set and verification set are 0.825 and 0.750, respectively, the root mean square error are 1.00% and 1.32%, respectively, and the normalized root mean square error are 10.85% and 13.55%, respectively. The combination of full subset selection method and machine learning can improve the inversion accuracy and robustness of SWC. This study provides a new way for rapid and accurate monitoring of SWC in farmland and precise irrigation. [ABSTRACT FROM AUTHOR]

Nitrogen is a key plant nutrient and its deficiency or surplus could inhibit plant growth and reduce crop yield. Over the past decade, remote sensing has been increasingly used to diagnosis nitrogen deficiency in crop. Taking rice grown in cold region in northeast China as an example, this paper studies the relationship between nitrogen content in japonica rice and the difference between spectral reflectance based on data measured from field. This relationship was used to inversely estimate nitrogen deficiency in the rice based on hyperspectral images. In our analysis, the nitrogen content producing the highest yield was defined as standard nitrogen content and its associated spectral reflectance was defied as standard spectral reflectance. The difference between real nitrogen content and the standard nitrogen content, as well as the difference between the real spectral reflectance the standard spectral reflectance, were calculated respectively. The difference in spectral reflectance was dimensionally reduced using the discrete wavelet multi-scale decomposition, continuous projection method (successive projections algorithm, SPA) and vegetation index construction. The characteristic bands screened by SPA were 459、460、475、 671、723、874 和 996 nm. Analysis showed that when the discrete wavelet multi-scale decomposition was used to reduce the dimension, the Sym8 wavelet mother function worked best when it was decomposed at the seventh layer. Comparing DVI, NDVI and RVI vegetation index found that the determination coefficient of the DVI index and nitrogen deficiency was significantly higher than that of NDVI and RVI index. The three indexes were used as input to the partial least squares (PLSR), the extreme learning machine (ELM) and genetic algorithm optimization extreme learning machine (GA-ELM). The GA-ELM model was most accurate with the R² being 0.7062 for the training set and 0.7594 for the verification set; their associated RMSE was 0.5099mg/g and 0.4276mg/g respectively. The GA-ELM model based on the optimal vegetation index was least accurate, with the R² for the training set and the verifying set being 0.6615 and 0.6509 respectively; their associated RMSE was 0.4415mg/g and 0.5312mg/g. Overall, GA-ELM improved stability and predictability of the model compared with PLSR and ELM. It can thus be used as a new method to detect nitrogen content in rice leaf, and has important implication in precision fertilization. [ABSTRACT FROM AUTHOR]

Estimating soil salinity is imperative for scheduling irrigation and remediating saline soil but difficult at large scales. Remote sensing can bridge this gap because of its advantages in low cost and large-area coverage; it has become an efficient method for assessing soil salinization in field. One issue in use of remote sensing to assess saline soil is the presence of plastic film mulch and bare soil because of their difference in reflecting waves in the spectral bands. In order to investigate the effect of plastic film mulch on soil salinity inversion using UAV multispectral remote sensing, we studied four plots with plastic film mulch at the Shahaoqu Irrigation area in the Hetao Irrigation District, Inner Mongolia of China. From each plot, we took soil samples and measured their salt contents from May to July. We also flew a drone to simultaneously take multispectral images of the sampling sites and extracted the spectral reflectance to calculate the spectral indices. Correlation analysis found that the S4, S6, SI1, SI2, SI3 and BI indices can be used to calculate soil salinity. Six-band spectral reflectances and six spectral indices obtained from different datasets were used as independent variables to calculate the salt content with the support vector machine (SVM), the back propagation neural network (BPNN) and the extreme learning machine (ELM), respectively, before and after the mulch film was removed. We compared the three models based on their determination coefficient (R²), root mean squared error (RMSE) and relative error (RE). The results showed that plastic film mulch did impact on soil salinity inversion. Although all three models could adequately estimate the soil salt contents before and after the film removal, they worked better after the film removal than before the film removal. Models based on the spectral indices were more accurate than those based on the spectral reflectances, and the accuracy of the inversely calculated salt content varied with sampling time and treatment. The inversion results based on monthly data differed from those based on by pooling all data. After the film was removed in June, the salt content estimated using the model was most accurate, with its associated R² and RMSE being 0.695 and 0.182 respectively for the spectral reflectance-based method, and 0.663 and 0.191respectively for the spectral indices-based method. The salt content estimated by BPNN was least accurate in May, with its associated R² and RMSE being 0.766 and 0.161 respectively for the spectral reflectance-based method, and 0.769, 0.162 respectively for the spectral indices-based method. Comparison of the three models revealed that ELM was most accurate, followed by SVM and BPNN, although their errors were within the tolerable range. In summary, this paper provides an effective method to inversely calculate soil salinization at large mulched farmland using UAV multispectral remote sensing. [ABSTRACT FROM AUTHOR]

Soil salinization is the main problem of land degradation in the Yellow River Delta of China. Remote sensing technology can gain soil salinization information quickly and accurately, which is of great significance to the sustainable development of agriculture. In this paper, a typical salinization area in Kenli County of the Yellow River Delta was chosen as the study area. In order to retrieve soil salinity from hyperspectral imagery with high accuracy, image fusion and machine learning were used in this study. HSI (hyperspectral imaging radiometer) hyperspectral imagery of HJ-1A satellite of China and OLI (operational oand imager) multispectral imagery of Landsat 8 of USA (United States of America) were preprocessed, including radiometric calibration, atmospheric correction and image registration. After that, the 2 kinds of images were fused with the hyperspherical color space resolution merge algorithm. This algorithm was designed for 8‑band data of Worldview‑2 sensor, and it works with any multispectral data containing 3 bands or more. The fused image has 30 m spatial resolution and 4.32 nm spectral resolution, in which saline soil can be identified better than that in the original image. The feature bands were selected according to spectral analysis of different levels of saline soil and the PLSR (partial least squares regression) regression coefficients between soil salinity and imagery bands. Two types of models, i.e. statistical model and machine learning model, were built. The statistical model includes multi linear regression model and PLSR model, while the machine learning model includes BP (back propagation) neural network model, support vector machine (SVM) model and random forest (RF) model. These models were built with soil salinity data as retrieval target and feature bands of images as input variables. In this process, natural logarithm function was adopted for soil salinity data to obey the normal distribution. The research gained the following results. Firstly, the retrieval model based on fused images is overall better than HSI images, and the latter is better than OLI multispectral images, which shows that both spatial and spectral resolution have important effects on the retrieval results. The retrieval accuracy of fused image is obviously better than that of HSI and OLI images. The main reason is that the fused image not only has both high spatial and spectral resolution but also has fewer mixed pixels. Secondly, with regard to performance of the models, the machine learning models are superior to the classic statistical models. This is because classic statistics often require sufficient samples, while machine learning is designed for small sample data and has more advantages over retrieval problems. In general, BP neural network model is better than SVM model and RF model. For the retrieval of fused images, the correlation coefficients of the 3 models are all higher than 0.82, and thus all of them achieve desirable results. Thirdly, the results also indicate that the accuracy of the models can be improved to some extent by proper preprocessing of the data, such as natural logarithm function which can let the data obey normal distribution. Either the classic statistical analysis method or the new machine learning method is based on the training data to explore the relationship between the retrieval target and the input variables. We then conclude that: 1) Despite the differences both in time and in wavelength between OLI and HSI images, image fusion can significantly improve the accuracy of remote sensing retrieval of soil salinity; 2) Machine learning model is better than traditional statistical model for soil salinization retrieval; 3) The main factors that affect the retrieval accuracy in our study include the number of measured samples, the quality of remote sensing data, the data preprocessing, fusion and modeling methods, and so on. Therefore, this study provides useful results and has positive theoretical and practical significance to the soil salinity retrieval in the typical area of the Yellow River Delta with remote sensing method. [ABSTRACT FROM AUTHOR]