Search

Efficient detection of pepper clusters can be crucial to realize the automated harvesting in the development of pepper industry. However, the existing detection of pepper cluster cannot fully meet the large-scale in recent years, due to the low efficiency, generalization, and real-time algorithm. In this study, an improved YOLOv5s model was proposed to accurately detect the pepper clusters in real time. 400 images were collected from the Xuande Pepper Industrial Park, Qinglong County, Guizhou Province, Qianxinan Buyi and Miao Autonomous Prefecture, Guizhou Province, China. 148 images were also downloaded from the Internet, and then screened to enhance data diversity. The collected 548 images were then labeled using the LabelImg tool. In addition, the data enhancement was performed on each image and labeled bounding box. Two new images were randomly generated using random rotation and brightness adjustment. As such, a total of 1644 images were obtained as the dataset. Firstly, the EfficientViT network was optimized using the lightweight convolutional neural network (CNN) MobileNetV3 Block Convolution (MBConv) module and the ReLU with lightweight self-attention mechanism. CSPDarkNet53 network was used to replace the backbone of YOLOv5s. The global receptive field of model was enhanced to extract the important features. The better performance was suitable for the targets with the different scales, shapes, and background information. Secondly, Optimal Transport Assignment (OTA) was selected as the label assignment strategy. The optimal label assignment plan was obtained from the global perspective to achieve the effective label assignment. The robustness and accuracy of the model were improved for the complex scenes and target distribution. The WIoU loss function was also used as the bounding box localization of the detector, instead of the CIoU one. At the same time, the high-quality anchor frames were achieved to generate the benefit gradients for the better overall performance of the detector. A common experiment of target detection was performed on a homemade dataset, in order to verify the effectiveness of the improved model. The experimental results show that the improved model was achieved in a mean accuracy (mAP) of 97.3%, a parameter count of 5.9 M, and a detection speed of 131.6 frames per second (FPS). Specifically, the mAP and detection speed of the improved YOLOv5s model increased by 1.9 percentage points, and 14.5%, respectively, whereas, the number of parameters decreased by 15.7%, compared with the original. In addition, the mAP of the improved YOLOv5s model was 8, 16.9, 8.6, and 1.5 percentage points higher than that of EffcientDet-D1, SSD512, RetineNet-R50, and YOLOXs, respectively. The better performance of the improved YOLOv5s was achieved in terms of detection accuracy, detection speed, and number of model parameters. Therefore, the improved YOLOv5s model can be expected to accurately and rapidly detect the pepper clusters. [ABSTRACT FROM AUTHOR]

Overlapping occlusion has seriously limited the yield estimation of bagged grapes in recent years, due to the ever-increasing grape volume after bagging and the large surface area of grape leaves. The unstable speed of manual video shooting can also lead to the loss of bagging grape targets. In this study, a yield estimation was proposed for the bagged grape using self-correcting Non-Maximum Suppression (NMS)-ByteTrack. Firstly, the bagged grapes were detected in the video using object detection (YOLOv5s). The NMS operation was also post-positioned to the tracking stage to retain the fruit detection boxes that filtered under the occlusion. Specifically, the detection boxes of bagged grapes were detected by object detection (YOLOv5s), whereas, the prediction boxes of bagged grapes were predicted to calculate the intersection over union using the Kalman filter. If the intersection over the union was less than the given threshold, the detection box was filtered to reserve the detection boxes closer to the predicted position. Then, the final detection box was obtained using the NMS operation. Secondly, the camera motion compensation and improved Kalman filter algorithm were added to automatically correct the position of the fruit prediction boxes and track them using ByteTrack. Specifically, the camera motion compensation was used to first extract the background key points in the bagged grape pictures of the previous frame and the current frame except for the tracking target. The sparse optical flow was used to match the extracted background in the key points of the bagged grapes. Then the affine transformation matrix of background motion was calculated by the RANSAC algorithm, and the Kalman filter was utilized to predict the bagged grape prediction boxes of the current frame. Finally, the affine transformation matrix was obtained to convert the prediction boxes in the coordinate system from the previous to the current frame, in order to realize the position information correction for the prediction boxes of the current frame. In addition, the improved Kalman filter algorithm can be used to change the aspect ratio in the state vector, and then to express the position of the bagged grape tracking boxes in the Kalman filter directly as the width, in order to more accurately estimate the position of the bagged grape tracking boxes. As such, a line-counting strategy was proposed to automatically count the bagged grapes. The strategy was also to set the counting line in the middle of the video, and also retain the idle frames without the bagged grapes in the first few seconds of the video. Furthermore, the bagged grapes with the same ID were only counted once to avoid repeated collision line counting. The successful collision between the center point of the tracking box of bagged grapes and the counting line was achieved in the automatic counting of bagged grapes. The dataset was collected from the Paidengte Agricultural Science and Technology Demonstration Park, Bishan District, Chongqing of China. The mobile phone cameras of Redmi K40 and OPPO Reno6pro+ were selected to capture the pictures of the same bagged grape at 8:00, 12:00, and 18:00 from different angles, such as frontal, sideways, and overhead shots. The total shooting time was about 6 h, the shooting height was about 1.5 m from the ground, and the shooting route was line by line. A total of 500 images and six videos were obtained for the bagged grapes. Among them, 500 images were expanded to 2000 images after saturation, brightness enhancement, brightness reduction, and mirror operations. Then, 2000 images were randomly divided into the training and validation sets in the ratio of 8:2, where six valid videos were used as the test set. Relevant experiments were conducted using this dataset. The experimental results showed that the yield estimation achieved a significant improvement in the tracking performance of bagged grapes using self-correcting NMS-ByteTrack. The multi-object tracking accuracy and precision, as well as the identification F1-score, were 64.6%, 82.4% and 80.8%, respectively, which increased by 1.7, 1.0 and 4.1 percentage points, respectively, compared with the ByteTrack. The number of ID switch was reduced by 3 times. Then, the average counting accuracy reached 82.8% in terms of counting performance, compared with manual counting. In addition, a comparison was also made with the five tracking methods. A better tracking performance was achieved, compared with the rest. The applicability of this estimation was also verified in tracking and counting bagged grapes. Therefore, the yield estimation can be expected to effectively promote the tracking and counting of bagged grapes in real scenarios using self-correcting NMS-ByteTrack. More accurate yield estimation of bagged grapes was also achieved in this field. [ABSTRACT FROM AUTHOR]

Straw mulching has been a common measure for black land conservation in the northeastern plains of China. It is a great significance of an effective solution to reduce soil erosion for environmental protection and sustainable development in modern agriculture. However, it is difficult to calculate the coverage rate at one time, due to the wide area of crop cultivation. Therefore, it is a high demand for the rapid acquisition of image information on a large area of farmland, in order to improve the efficiency and accuracy of straw mulching. Fortunately, aerial drone images and videos can be widely used in various fields, such as agricultural monitoring, weather forecasting, and geographic mapping. Among them, the application of remote sensing technology focuses mainly on the crops and soil in agriculture. The spectral patterns of the ground can be utilized to monitor the crops and soil in the field. Specifically, the spectral characteristics of the ground are used to monitor the growth of crops, crop quality, crop pests, and diseases. The high-resolution images can be collected by unmanned aerial vehicle (UAV) aerial photography. The panoramic stitching can be selected for rapid and accurate access to the large-area farmland image information. Therefore, it is of great practical significance to agricultural intelligence, particularly for the high efficiency and accuracy of straw cover monitoring of conservation tillage. In this study, a fast-stitching method was proposed for the aerial panoramic images of the fields using the optimized scale-invariant feature transform (SIFT) algorithm. Firstly, the high-resolution image was down-sampled to effectively detect the overlapping regions of the image. The gradient normalization-based feature descriptors were then used to match the feature points. The false matches were removed by the progressive sample consistency algorithm, in order to accurately calculate the stitching conversion model. Finally, the multi-resolution fusion algorithm was used for the high-quality stitched image using the best stitching line. The experimental results show that the improved algorithm reduced the number of feature points by 97% and 90%, the running time by 94% and 69%, and the average matching efficiency by 65.17% in the image alignment stage, respectively, compared with the traditional SIFT and Speeded Up Robust Feature (SURF) algorithm, which was about 4 times that of SIFT algorithm, and 9 times that of SURF algorithm, respectively. Compared with the as-projective-as-possible (APAP), the shape-preserving half-projective (SPHP), and the adaptive as-natural-as-possible (AANAP) algorithms, the new algorithm significantly improved the information entropy, the average gradient, and image contrast of the stitched image. More importantly, the stitched image was of higher quality with more than a 90% reduction in the running time, particularly with better overall performance. The systematic evaluation demonstrated that the layered stitching method significantly improved the clarity and fusion effect of the panoramic stitched images, compared with the traditional. Moreover, there was no misalignment and ghosting that appeared in the traditional. Specifically, the information entropy and average gradient of the panoramic images were slightly improved to increase the contrast by about 20%, indicating a higher stitching efficiency than before. The overall stitching time was also shortened by about 90% or more. The research structure can also provide a scientific reference to monitor the conservation of tillage with straw returning. [ABSTRACT FROM AUTHOR]

Apples are produced in the large quantities each year, particularly for as the largest economic fruit in China. It is highly required for the rapid picking within the harvesting period. Therefore, the automatic apple picking is essential to the apple harvesting in intensive farming. An accurate and rapid identification of fruit can be fundamental for the automatic picking. However, some environmental factors surrounding the fruit can pose a great interference in the fruit identification under the natural, complex, and variable backgrounds, such as the light intensity, occlusion, and overlap of the fruit. In this study, an apple recognition was proposed using two-level segmentation and region-marked Hough transform. Experimental results show that the robust and practical performance was achieved for the apple recognition under different illumination, branch and leaf occlusion, as well as the fruit overlap. Specific steps were as follows. Firstly, the front camera (NikonD90) of the information acquisition robot was used to capture from 600-800 mm away from the fruit tree under the conditions of nature natural light and backlight, respectively. The brightness adaptive correction algorithm was then used to correct the brightness of apple images with the uneven distribution of surface brightness, in order to enhance the image details. The Cr component images of YCbCr color space were extracted as the preprocessing samples to combine with the feature of the color of the apple. Secondly, the improved Otsu algorithm was utilized to obtain the binary image of the apple target for the initial segmentation, in order to accurately extract the contour of the target fruit under different growth states (mainly including single and double fruits with the overlap and occlusion). A morphological open-close reconstruction filter was also introduced to the Otsu algorithm to remove the background noise. The traversal range of the gray level was reduced to shorten the complexity and running time of the algorithm for the high segmentation rate. Thirdly, the watershed algorithm was combined to perform the secondary segmentation of the segmented fruit region using distance transformation. The conglutinated and overlapping apples were separated to effectively extract the apple target contour. Finally, the gradient Hough circle transformation was selected to identify the number of apples. But the algorithm traveled through the whole image for the computational complexity, time time-consuming, and easy to produce the false identification. Therefore, the minimum circumscribed rectangle outside the contour was set as the effective area for the gradient Hough circle transform in the effective area, particularly for the recognition speed and accuracy. The experimental results show that: 1) The improved Otsu algorithm was achieved in the higher segmentation accuracy of fruit targets, especially with the less segmentation time. The improved algorithm was also filled the tiny holes in the apple to suppress the noise in the background branches and leaves, further to more clearly segment the target region more than before. The average segmentation time was 1.458 s, which was 0.643 and 1.060 s shorter than the Otsu and K-means algorithm before the improvement. In the adhesion of partial apple regions in the binary images, a watershed segmentation with the distance transformation was used for the quadratic segmentation to effectively separate the sticky apples for the full apple target boundary. 2) The gradient Hough circle transform was used to recognize the 200 apple images under different lighting conditions, where the recognition accuracy was 90.75% in the nature natural light and 89.79% in the backlight, which were improved by 15.03 and 16.41 percentage points, respectively, compared with the traditional. The average recognition time was 0.665 and 0.693 s, which was 0.664 and 0.643 s shorter than before. Therefore, the proposed algorithm can meet the requirements of apple-picking robots, in terms of recognition speed and accuracy. The findings can provide a strong reference for the fast recognition of spherical fruits, such as apples. [ABSTRACT FROM AUTHOR]

Digitization has been one of the significant directions in the rural land governance reform in the future. However, it is still lacking in the logical concepts and framework for the digital governance of rural land. In this study, a three-dimensional system framework of “demand-baseline-function” was established using the perspective of the transaction cost. An emphasis was also put on the construction of the digital system for land acquisition and relocation. Firstly, the basic, core and conflict needs of the participants were clarified ranging from the government, land users, and the original land stakeholder. Secondly, the bottom line was defined as the legal, spatial, and regulatory governance. Thirdly, the digital system was classified to design various digital realities. The future functional modules were selected to entirely reduce the transaction costs, thereby integrating the multiple sharing systems for the overall operational efficiency. In addition, Zhejiang Province of China was selected as the experimental area for the system design. The reason was that Zhejiang Province has been the representative region in the digital reform of land governance in recent years. Nevertheless, there was a high transaction cost in the process of land acquisition and relocation. It is also a high demand to effectively build a digital system during this time. As such, the actual system was designed to combine with the improved the digital framework. The result showed that the “demand-baseline-function” framework effectively guided the development and design of the “intelligent land acquisition and demolition” digital system in the actual application situation. The transaction costs were also reduced to fill the logical blank for the previous system of land digital governance. The demand analysis demonstrated that the improved system provided the port to fully meet the basic and core needs of the government, land users, collectives, and farmers. The information asymmetry, discourse game, and conflict transmission effectively reduced the transaction cost of the game between the subjects. The bottom line was clarified for the implementation boundary of the law and space of the land requisition and relocation process. The whole supervision system of land requisition and relocation was constructed to stabilize the land requisition and relocation environment. The functional analysis revealed that the improved system coordinated the actual and upgrade function of the “intelligent land acquisition and demolition” digital system. The logical framework effectively reduced the negotiation and decision cost, while the implementation and supervision cost, as well as the management and sunk costs in the land acquisition and demolition. Therefore, the “demand baseline function” framework can provide a strong reference for the transaction cost in the digital reform of land governance. Anyway, the underlying construction logic of the framework can also be applicable to the other areas without the digital construction guidance, providing for the most general transaction cost in the rural land governance. In addition, the logical framework of the system can also provide a strong reference for the digital transformation of rural land governance in the future, in order to improve the digital level of land space and the modernization of governance capacity. [ABSTRACT FROM AUTHOR]

t: Stalk set-row planting has been one of the conservation tillage systems to minimize the frequency or intensity of tillage operations in northeast China. The agricultural management approach can integrate the tillage and planting operations to cover at least 30% of the soil surface with crop residues. A narrow row zone can be sown to clear out from the stalk-covered surface in the wide leisure zone. However, the pesticide liquid cannot reach the stalk strip surface, particularly when weeding the crop during the seedling stage under stalk mulching in the traditional full-coverage spraying mode. In this study, a precision spraying device was designed to promote the utilization rate of pesticide liquid using image recognition. Firstly, a digital camera was mounted in front of the plant protection machine to real-time capture the stalk set-row planting images in fields. A Digital Signal Processor (DSP) was used as the host computer to pre process the images. Then, an optimized multi-threshold segmentation was performed on the images of stalk and seedling strip parts. A series of morphological operations were utilized to correct the dividing edges, where an external moment was selected to frame and separate the seedling strip areas of the field. As such, the position of the seedling strip center was evaluated to mark out the line of the seedling strip. Secondly, an STM32 microcontroller was used as the lower computer. Specifically, the controller drove the stepper motor slide to move the nozzle on the top of the seedling strip for spraying, according to the line coordinates of seedling strip positioning. At the same time, a multi-subdivision drive was used to improve the operation precision of the stepping motor. Nevertheless, some unfavorable conditions occurred in the horizontal and vertical kinds of airflow during spraying, whether the droplets drifted into the stalk rows, or the uneven spraying was caused by the air. Therefore, the drift rate of the droplet was calculated under different velocities of airflow using Fluent fluid simulation software. After that, the fuzzy controller was set up, with the horizontal and vertical airflow speed as the input, and the spray volume as the output. The spray volume was changed to adjust the opening degree of the electric valve for a better uniform settlement of the chemical liquid. Finally, the precision spraying device on a platform was tested to verify the simulation in the field. The test results showed that the accuracy of the self-developed device was about 88.3% for the positioning line of the seedling strip. A WFS-II integrated test bench was used to evaluate the spray performance. The spraying of a single nozzle was measured on the rows to save the liquid by about 64.8% more than before. The qualified rate of spraying on the rows was about 85.8% than before. The variable spraying test was conducted to maintain the liquid coverage at 28.5%-33.8% in the variable spraying mode. The field trials demonstrated that the precision spraying saved about the pesticide liquid of 52.84% in the stalk set-row planting mode, compared with the traditional full-coverage one. Furthermore, the uniformity level of liquid deposition was also improved significantly. Consequently, this device can be expected for the plant protection machinery in the precision application of seedling strips during the stalk set-row planting. The finding can provide a strong reference for the precision spraying to improve the liquid utilization rate and environment, particularly for the technical support to promote the conservation tillage. [ABSTRACT FROM AUTHOR]

Litchi has been one of the characteristic fruits in Guangdong Province of South China. Traditionally, the litchi can be inspected manually in the orchard, due to its susceptibility to the weather, diseases, and insect pests. The number of fruits in the litchi tree can be counted to determine the subsequent agricultural operations, such as pouring nutrient solution, or the removal of insect pests. Nowadays, an unmanned aerial vehicle (UAV) low-altitude remote sensing has been a promising way to observe the litchi, particularly for safety, efficiency, cost-saving, and easy operation. But the litchi images taken by the UAV are characterized by the smaller target size and fewer features. In this study, an improved Single Shot MultiBox Detector (SSD) model was proposed to detect the small litchi fruits using multiple feature enhancement and feature fusion (MFEFF). Firstly, 160 litchi images with a resolution of 4000×3000 pixels were collected in the orchard using the UAV. A sliding window of 512 pixels was also applied to the original litchi images, according to the target pixels between 20×20 and 30×30 pixels. As such, 3 590 images of 1 024×1 024 pixels were captured to boost the feature information of litchi, such as the size. Secondly, the data enhancement was implemented to improve the robustness of the model using some operations, including the image flip, color space transformation. Thirdly, the last two convolution layers of Vgg16 were deleted to reduce the unnecessary computation. The Receptive Field Block (RFB) was used on the Conv8 and Conv9 layers, and the feature map of Conv3_3 layer was added for the feature extraction, where much more features of the litchi were expanded the receptive field for the detailed information. The Efficient Spatial Pyramid (ESP) Block was also applied on the enhancement of the shallow features in the maps. Finally, the improved Path Aggregation Network (IPANet) was used for the multi-scale features fusion at the Conv3_3, Conv4_3, and fc7 layers. A Squeeze and Extension (SE) module was also introduced in the first two feature layers, further to improve the detection accuracy. The channel attention network was also used the global information of litchi images to selectively enhance the weight of the litchi channel, but to efficiently suppress the useless feature information, such as green leaves against the background. At the same time, the size and quantity of anchors were adjusted to match the size of the small target litchi. A training detection was carried out to verify the model for the total of 3 590 labeled images, where 3 to 114 litchi targets were set in each image. Among them, 2 907 images were distributed as the training set, 324 as the validation set, and 359 as the testing set. The precision indexes were the mean Average Precision (mAP), recall, precision, and F1-Score (F1). The results showed that the RFB module was significantly improved the detection, compared with the original, where the mAP, Recall (R), Precision (P), and F1 increased by 2.61, 0.48, 1.49, and 1 percentage points, respectively. The IPANet detection was better than that of the feature pyramid network (FPN), where the mAP value increased by 0.44 percentage points. The SE module was better than the Convolutional Block Attention Module (CBAM) and Efficient Channel Attention (ECA), indicating the best score of the three modules. The ESP detection was superior to the atrous spatial pyramid pooling (ASPP), in which the mAP, R, and P increased by 2.51, 0.38, and 1.13 percentage points, respectively. Consequently, the MFEFF-SSD had improved the mean average precision by 30.62, 14.58, 44.46, and 15.93 percentage points, respectively, compared with the SSD, Yolov4-tiny, Faster-RCNN, and CenterNet models. Anyway, the MFEFF-SSD model can be widely expected to more accurately and effectively detect the litchi images taken by UAV. This finding can also provide a strong reference for the detection of small target fruits. [ABSTRACT FROM AUTHOR]

Lameness is an abnormal gait or stance of an animal, indicating a more severe disorder of the locomotion system. The disturbance in the gait and body has posed a great threat to the welfare, health, and production of dairy cows in the herds. An early observation is typically performed on the sudden change in gait during day-to-day dairy practices. However, the current detection of lameness was widely used in the 2D cow video, particularly susceptible to light and environmental inferences. In this study, a novel approach was proposed to extract the movement features of the lameness cows using thermal infrared video under deep learning and image processing, in order to improve the detection accuracy and robustness against the complex illumination and environmental disturbances. Taking the multiparous lactating Holstein cows as the study objects, the datasets were collected at the Dingyuan farm in Hebei Province of China in September 2020. The gait features of lameness were related to the head, hoof, back, shoulder, and hip of cows. The performances of different target detection were then evaluated on the dataset of locomotion features from the thermal infrared and 2D images of cows walking. The results showed that the average accuracies of deep learning for the detection of cows locomotion features were 90.84%, 86.68%, and 79.68%, respectively, in the thermal infrared, daylight, and daylight images, whereas, those in the traditional image processing were 74.26%, 54.31%, and 48.95%, respectively. The test results demonstrated that the target detection of deep learning and image processing performed better on the thermal infrared images, compared with the 2D images. The thermal infrared images had effectively reduced the influences of external factors, such as the light and background. In addition, a lameness detection test was conducted on the walking cows using the thermal infrared video and 2D video, according to the arched curvature. It was found that the detection accuracies of cow lameness in the thermal infrared and 2D video were 90.0% and 83.3%, respectively, where the values of Macro-F1 were 0.90 and 0.83, respectively. Consequently, a better performance was achieved for the application of thermal infrared cameras in cow lameness detection using computer vision. This finding can provide a new potential strategy for the timely diagnosis and prevention of abnormal gait in cows in the dairy farming industry. [ABSTRACT FROM AUTHOR]

The slique is one of the most important organs to store the energy matter in an oilrape plant, a typical oil crop in China. It is necessary to measure the seed number per silique of oilrape for the breeding traits, even for the estimation of rape yields. But the current manual counting cannot fully meet the requirement of modern breeding, due to the time-consuming, labor-intensive, and less accuracy for the seed number per silique of oilrape, particularly by manually breaking the pod of rape. In this study, Image recognition of the silique length was proposed to count the seed number per silique of oilrape. Three varieties of oilrape were also used as the experiment materials. The siliques were first selected from the upper, middle and lower parts of rape plants. After that, a digital calliper was utilized to measure the length of silique pod without considering the fruit stalk and beak part. A correlation model was then established for the length of silique and seed number per silique. The images of siliques were also captured by a scanner in the tiled state. The image preprocessing included the gray scale, binarization, hole filling, and the removal of small areas. Meanwhile, the fruit stalk and fruit beak of silique were removed from the images of siliques using the opening operation of mathematical morphology. After that, the images were segmented to the sub-images of siliques with the single connected domain. Each sub-image of siliques was then thinned to a line with one pixel. The endpoint and intersection point of siliques images were finally obtained using hit-or-miss transformation. The Density Based Spatial Clustering of Applications with Noise (DBSCAN) was also used to remove the redundant intersection points. The endpoints of silique images were classified and paired to recognize the single silique. As such, the number of image pixels was obtained for the length of a single silique. The actual length of silique was calculated by a relative formula between the single pixel and actual length. Correspondingly, the seed number per silique was finally calculated from the images using the correlation model between the actual length of silique and seed number per silique. The determined coefficient for the correlation model between the length of silique and seed number per silique in the three varieties were 0.891, 0.881, 0.887, respectively, indicating a better performance of the estimation on the seed number per silique using the length of the silique. Furthermore, the average recognition accuracy was 97.25% for the length of silique, and the average prediction accuracy of seed number per silique was 83.87% using image processing. Consequently, image processing can be widely expected to realize the automatic recognition and counting of the seed number per silique for higher efficiency of rape production in modern agriculture. [ABSTRACT FROM AUTHOR]

BACKGROUND: Currently, forensic anthropology, ergonomics and footprint studies have all found that there is a significant correlation between foot information (such as foot length and foot width) and body height and mass. However, there is no enough quantitative research on barefoot morphological characteristics contact area (BMCCA) as well as human body height and mass. OBJECTIVE: To analyze the correlation between BMCCA and body mass and height, and to innovate the theoretical systems of forensic anthropology and ergonomics, thereby providing reference for individual recognition by barefoot footprint. METHODS: Samples were collected using traditional plane footprint ink printing method, and digital software MATLAB was then used for image processing. A total of 100 healthy young men with a height distribution between 165-190 cm were randomly selected as subjects. Body height and mass were recorded. BMCCA was obtained under normal walking for statistical analysis. The implementation of the study complied with the Declaration of Helsinki and the relevant ethical requirements of Criminal Investigation Police University of China. RESULTS AND CONCLUSION: For healthy young men with normal arch of the foot (excluding interrupted arch of the foot and expanded arch of the foot), there is a significant positive correlation between BMCCA and height and mass (P < 0.01). The relationship between BMCCA and human body height can be applied for height analysis in forensic anthropology, ergonomics, and footprint. Compared with body mass, body height has a more significant impact on BMCCA and is more directly related. [ABSTRACT FROM AUTHOR]

This study aims to improve the detection precision of plant root phenotyping in the images, particularly with broken roots. An algorithm of radial growing repair was proposed to apply to the evaluation of maize seed’s resistance to damage. A soil culture experiment was also conducted on the Pukou campus of Nanjing Agricultural University in China every month. After that, four varieties of corn seeds were selected: Yufeng 303 (ordinary control group), Zhengdan 958 (drought resistance group), Liyu 16 (water resistance group), and Zhengdan 958 (salt resistance group). Four kinds of resistant corn seeds were planted in trough-shaped and flat containers for 14d, including ordinary, drought-resistant, water-resistant, and salt-resistant corn seeds. Subsequently, the root system was taken out to rinse the residual soil with tap water, and then placed on a solid-color background plate to level out. Prior to image acquisition, the root length and diameter were measured by a ruler, and the number of main and lateral roots was counted to record. The specific procedure of image processing was as follows. Firstly, a series of operations was used to preprocess the collected images for the extraction of root systems from complex backgrounds, such as adaptive contrast enhancement, histogram grayscale searching, and pepper-salt denoising. As such, the discrimination of root images was improved to remove the noise during image acquisition, such as reflections and water stains. Secondly, the tips of main roots in maize images were detected by training the YOLO-V3 neural network. Finally, the radial growth repair algorithm was presented, including the direction discrimination of main roots in bifurcation points, and adaptive repair in end points. These strategies greatly contributed to extracting phenotypic parameters from main and lateral roots. Maize root datasets were also selected to evaluate the practicality and accuracy of radial growth repair. The results demonstrated that the phenotypic accuracy of repaired main roots lengths and diameter increased from from 83.6% and 84.4% to 97.4% and 94.8%, respectively, compared with that processed by region growth algorithm. The phenotypic parameters extracted by radial growth repair algorithm was more precise than that extracted by region growth algorithm, which indicated that radial growth repair algorithm was suitable for extraction of maize root system parameters in different stress environments. The accuracy of radial growth was improved more obviously in the salty environments and drought environments. The results in this study demonstrated that the proposed radial growth repair algorithm could improve the accuracy of root image phenotype detection and could be efficient for maize resistance evaluation, which provided reference for the rapid extraction of root system phenotype. [ABSTRACT FROM AUTHOR]

As an important economic tree species in China, tea trees play a critical role in promoting economic development. However, the expanding area of tea plantations in China over the recent years has reduced the biodiversity of the ecosystem and accelerated ecological problems. Identifying the spatial distribution of the tea plantation accurately is important for the monitoring of the tea plantation as well as the development of the tea industry. In this study, Wangzhai Town, located in the central of Wuyi County in Zhejiang Province, was chosen as the study area, and Gaofen-1 (GF-1) data on December 28, 2017, and all available Sentinel-2 data in 2017 were downloaded to study the tea plantations extraction method. The spectral and texture features of tea plantations are similar with crops on the high spatial resolution remote sensing image which results in the mixture of the tea plantations and crops. The Normalized Difference Vegetation Index (NDVI) time series data can reflect the phenological changes of the tea trees and crops, which can be used to improve the separability of tea plantations and crops. In this study, a phenology-based approach was applied to the NDVI time series to distinguish tea plantations from crops by comparing the similarity of the seasonal changes of NDVI time series. Due to the influence of cloud and fog on the satellite image, the complete Sentinel-2 time series data is difficult to be obtained. Therefore, the high temporal resolution Moderate-resolution Imaging Spectroradiometer (MODIS) reflection product (MCD43A4) in 2017 was used to reconstruct the complete Sentinel-2 NDVI time series data (5 m spatial resolution, 5 d temporal resolution) based on Bayesian Maximum Entropy (BME) spatial-temporal fusion algorithm. Then, the Dynamic Time Wraping (DTW) distance was calculated based on the DTW algorithm to characterize the similarity of NDVI time series between the tea plantation samples and other pixels to be classified. Combining the advantages of GF-1 in spatial details and the high temporal resolution of reconstructed Sentinel-2 NDVI time series in capturing the difference of growth process of tea trees, the tea plantation was identified based on random forest algorithm. Two identification schemes: 1) spectral and texture features based on GF-1; 2) spectral, texture features based on GF-1 and DTW distance were set up to investigate the impact of NDVI time series on the tea plantation identification. The results showed that: 1) the reconstructed NDVI time series data derived from MODIS and Sentinel-2 based on spatio-temporal fusion algorithm is able to capture the seasonal dynamics of different objects, and the DTW distance can capture the difference of NDVI time series data between crops and tea plantations and can be used for tea plantations extraction; 2) the accuracy、 error rate、precision、recall and F1score were 96.91%、3.09%、89.00%、83.09% and 0.86 respectively combining texture and spectral information of GF-1 and Sentinel-2 NDVI time series information, which were 94.72%、5.28%、73.09%、84.61% and 0.78 respectively without Sentinel-2 NDVI time series information. Overall, the results of combining GF-1 and Sentinel-2 time series data performed better than that only based on the spectral and texture features of GF-1. These results confirmed that combining high spatial resolution with high temporal resolution time series remote sensing data is an effective method to improve the identification accuracy of tea plantations. [ABSTRACT FROM AUTHOR]

Wheat is one of the main food sources in China, and wheat yield is very important to China's food security. Lodging is a common agricultural disaster in wheat production. Lodging often causes changes in the population structure of wheat, affects the protein synthesis and nutrient transport of wheat, and greatly influences the quality and yield of wheat. To extract the lodging area of wheat quickly and accurately, this study adopted the Unmanned Aerial Vehicle (UAV) remote sensing platform to obtain the visible light images (including the red band, the green band, and the blue band) and the Digital Surface Model (DSM) images of the wheat canopy at the late stage of grain filling. The Excess Green (EXG) index was also calculated. Fusion images of DSM+RGB as well as fusion images of DSM+EXG were created by ArcGIS image mosaic tools. By using the maximum likelihood method and random forest method, the two feature fusion images were supervised and classified for extracting the wheat lodging area, and the results were also compared with the extracted results only based on the RGB images and the DSM images. The results of using four images to extract the lodging area of wheat were evaluated by the confusion matrix and Kappa coefficient. The results showed that the two methods had the same overall trend of wheat lodging area extraction on four images, and the extraction effect of the maximum likelihood method was generally excellent. Based on the random forest method, the overall accuracy of RGB image, DSM image, DSM+RGB fusion image, DSM+EXG fusion image to extract the lodging wheat area was 77.21%, 93.37%, 93.75%, 81.78%, and the Kappa coefficients were 0.54, 0.86, 0.87, 0.64, respectively. According to comparative analysis, the image data extraction accuracy after fusion of the DSM and RGB parameters was the highest, and the overall accuracy was up to 93.75%, the Kappa coefficient was 0.87. The difference between the two images was improved due to added elevation information. The extraction effect of the lodging area was the worst which was only based on the RGB images with an overall accuracy of 77.21% and a Kappa coefficient of 0.54. The reason was that the wheat with normal plants but yellow leaves were misclassified. Compared with the fusion images of DSM+RGB, the classification accuracy based on the DSM images alone was not much different, indicating that DSM effectively reflected the difference between lodging wheat and normal wheat. It was an effective feature, while the accuracy of the latter was higher than that of the former, indicating that the fusion of unreasonable image features effectively improved the image difference and the extraction accuracy of the lodging area of wheat. In this study, the fusion of different image features preserved more detailed information of the image, which enhanced the difference of image features and made the fused image suitable for image classification. This research had shown that the method of image feature fusion could effectively extract the information of lodging wheat and provide a method for accurate identification of wheat lodging. [ABSTRACT FROM AUTHOR]

Buckwheat is widely cultivated in the regions of high altitude and cold mountains, such as northern and southwest China. The occurrence of crop diseases has posed a great threat to the quality and yield of buckwheat. Disease surveillance is an important measure to ensure the healthy development of the buckwheat industry. Since artificial intelligence has been extending to precision agriculture in recent years, machine learning and pattern recognition are beneficial to image classification, detection, and recognition for high accuracy and efficiency, while reducing the overhead in the detection of crop diseases. However, the existing deep learning cannot consider the complexity of disease images collected in the field, such as leaf overlap, uneven lighting, and shadow coverage. Therefore, it is necessary to accurately extract key features in a complex imaging environment for accurate recognition of crop diseases. In this study, a feasible multi-layer feature extraction in deep learning was proposed to extract the features of buckwheat diseases. A surveillance system of disease was then established according to the classified characteristics, thereby automatically identifying the categories of buckwheat diseases. Firstly, a combination of maximally stable extremal regions (MSER) and convolutional neural network (CNN) was used to detect the feature regions of buckwheat disease, where the disease and disease-free areas were separated to precisely locate the disease position. A two-level inception structure was then added to the traditional CNN. The first layer was used to extract the contour features of the disease, where the dimension of the feature was represented by shallow information. The second layer was selected to obtain more detailed features, so that the number of parameters was relatively small in each layer, avoiding the gradient disappearance in the training process. This structure was normally used to extract the features for the low-quality images of buckwheat disease, particularly for the higher accuracy of classification. An improved convolution based on cosine similarity was utilized rather than the traditional convolution operation, in order to reduce the sensitivity of illumination during the sampling process. After that, the positions with similar features to the convolution kernel behaved higher activation values in the feature map. In addition, the difference between features was also reduced to prevent the interference of sample noise, thereby achieving better feature extraction for samples with uneven illumination. Finally, a systematic dataset was established, including some images for eight types of buckwheat diseases. The results showed that the combined framework of MSER and CNN with the two-level inception recognition was an effective way for region detection. In the identification for the presence or absence of disease, the accuracy, recall, and F1-measure reached 97.54%, 96.38%, and 97.82%, respectively. Furthermore, the mean for the recognition of disease categories reached 84.86%, 85.78%, and 85.4%, respectively, indicating excellent performance in recognition accuracy and speed. The finding can provide promising technical support for the automatic recognition of buckwheat diseases. [ABSTRACT FROM AUTHOR]

Automation processing has become particularly important for the Camellia oleifera industry in Southern China, as the agricultural economy is ever increasing. Fruit shelling of Camellia oleifera is a very critical link in the production line. There are still some problems so far in the sorting and recognition system for the Camellia oleifera sheller, such as single-feature recognition method, great disturbance by target color, and relatively low adaptive function. This study aims to propose a multi-features intelligent sorting algorithm, combining the artificial immune network (aiNet) and Support Vector Machine (SVM), in order to fully utilize the multi-feature clustering feature of immune algorithm, and the dichotomy feature of SVM algorithm. Six morphological and color characteristics of shell kernel in a Camellia oleifera were extracted, including elongation, roundness, completeness, R component, G component, and B component of color feature. These characteristics were used to sort and identify the shell and kernel of Camellia oleifera. The collected images were first preprocessed, then three morphological features were integrated into the aiNet algorithm for multi-features comprehensive identification, finally three-color features were input into the SVM algorithm for the recognition of color features. Since the color of fruit shells and seeds varied in different storage periods, 3 and 12 days were selected to obtain the obvious color characteristics of Camellia oleifera fruits, considering the influence of temperature, and humidity, on the picking Camellia oleifera fruits. In the experimental test, the multi-features immune network combined with SVM algorithm significantly reduced the complexity of multi-dimensional operation while saved the operation time. The results showed that the sorting efficiency of Camellia oleifera fruit reached 97.4% in 3 days, and 76.6% in 12 days, indicating a high separation efficiency. The recognition time reached an average of 600 ms and a minimum of 510 ms, where the recognition time was the sum of the consumption time of two algorithms, and the ratio of time consumed by aiNet and SVM algorithm was 2.3:1. A comparation was made in the recognition rate and time, including the multi-dimensional aiNet, the multi-dimensional SVM algorithm, the color threshold method, and the morphological threshold method. Although the conventional algorithm of color morphological threshold had a short execution time, it does not have multi-features adaptability, as its simple structure. Nevertheless, the usage of multi-features immune algorithm can easily lead to the "dimension disaster" of long recognition time, particularly when to recognize six features of shape and color. The multi-dimensional SVM algorithm was not suitable for the multi-feature recognition, due to its binary structure. An improved algorithm can also lead to the problem of long sorting time, due to the complexity of structure. The recognition rate decreased, when the color difference was not obvious during the storage period of 12 days. The combination of artificial immune and SVM can be used to enhance the efficiency and real-time performance, particularly better than other methods in the shelling and sorting production line of Camellia oleifera fruit. The finding can verify the algorithm with innovative and practical characteristics, thereby to improve the production of Camellia oleifera fruit. [ABSTRACT FROM AUTHOR]

In China, the current field survey methods of pest damage symptoms on rice canopy mainly rely on the forecasting technicians to estimate the sizes and numbers of damage symptoms by visual inspection for estimating the damage level of pests in paddy fields. The manual survey method is subjective, time-consuming, and labor-intensive. In this study, an improved RetinaNet model was proposed to automatically detect the damage symptom regions of two pests (Cnaphalocrocis medinalis and Chilo suppressalis) on rice canopy. This model was composed of one ResNeXt network, an improved feature pyramid network, and two fully convolutional networks (one was class subnet and the other was regression subnet). In this model, ResNeXt101 and Group Normalization were used as the feature extraction network and the normalization method respectively. The feature pyramid network was improved for achieving a higher detection rate of pest damage symptoms. The focal loss function was adopted in this model. All images were divided into two image sets including a training set and a testing set. The training images were augmented by flipping horizontally, enhancing contrast, and adding Gaussian noise methods to prevent overfitting problems. The damage symptom regions in training images were manually labeled by a labeling tool named LabelImg. 6 RetinaNet models based on VGG16, ResNet101, ResNeXt101, data augmentation, improved feature pyramid network, and different normalization methods respectively were developed and trained on the training set. These models were tested on the same testing set. Precision-Recall curves, average precisions and mean average precisions of six models were calculated to evaluate the detection effects of pest damage symptoms on 6 RetinaNet models. All models were trained and tested under the deep learning framework PyTorch and the operating system Ubuntu16.04. The Precision-Recall curves showed that the improved RetinaNet model could achieve higher precision in the same recall rates than the other 5 models. The mean average precision of the model based on ResNeXt101 was 12.37% higher than the model based on VGG16 and 0.95% higher than the model based on ResNet101. It meant that the ResNeXt101 could effectively extract the features of pest damage symptoms on rice canopy than VGG16 and ResNet101. The average precision of the model based on improved feature pyramid network increased by 4.93% in the detection of C. medinalis damage symptoms and the mean average precision increased by 3.36% in the detection of 2 pest damage symptoms. After data augmentation, the mean average precision of the improved model increased by 9.13%. It meant the data augmentation method could significantly improve the generalization ability of the model. The improved RetinaNet model based on ResNeXt101, improved feature pyramid network, group normalization and data augmentation achieved the average precision of 95.65% in the detection of C. medinalis damage symptoms and the average precision of 91.87% in the detection of C. suppressalis damage symptoms. The mean average precision of the damage symptom detection of 2 pests reached 93.76%. These results showed that the improved RetinaNet model improved the detection accuracy and robustness of pest damage symptoms on rice canopy. It took an average time of 0.56 s to detect one image using the improved RetinaNet model, which could meet the realtime detection task of pest damage symptoms on rice canopy. The improved RetinaNet model and its results would provide the field survey data and forecasting of damage symptoms of Cnaphalocrocis medinalis and Chilo suppressalis on the rice canopy. It could be applied in precision spraying pesticides and pest damage symptom patrol by unmanned aerial vehicles. It would realize the intelligent forecasting and monitoring of rice pests, reduce manpower expense, and improve the efficiency and accuracy of the field survey of pest damage symptoms on rice canopy. [ABSTRACT FROM AUTHOR]

Rice is a staple crop in China. Controlling rice pests and diseases is important to safeguard its sustainable production, in which classification and identification of rice planthoppers plays an important part. While there has been an increased interest over the past few years in image-classification of the rice planthoppers, currently, this method is not automatic and susceptible to faulty recognition and low efficiency. To circumvent these shortcomings, a sparse-representation image-based classification method was proposed based on the K-SVD and OMP. A field insect collection device was used to collect insect images, in which a high-pressure mercury lamp was used to attract the insects to the collection workbench based on their phototaxis characteristics. A PLC was mounted on the top of the three-phase adjustment device to control high-definition industrial cameras to take images of the insects. The images were then segmented using the maximum inter-class variance method (OTSU threshold segmentation method) to extract the image of the insects. Overall, 1186 single insect images were obtained from the field experiment. Two insect characteristics were selected as initial over-complete dictionary and they were extracted from 500 images. K-SVD dictionary learning algorithm was used to iteratively update the over-complete dictionary of the rice planthopper image features and the number of iterations was set at 500. 200 images of the rice planthoppers were selected as the original input comparing images, and 150 and 50 images of the rice and non-rice planthopper were used as testing and verifying sets respectively. The appropriate sparsity is an important factor for improving efficiency and accuracy, and the errors of the testing images was calculated from the OMP algorithm when the sparsity was 6, 12 and 18 respectively. Comprehensive analysis of the errors and convergence rate showed that the optimal sparsity was 12. Finally, classification features of the rice and non-rice planthoppers were sparsely reconstructed in line with the updated reconstruction dictionary, and the reconstruction error was calculated for the sparsity of 12. Hundreds of experiments revealed that the classification threshold of 0.1 was quick and effective to classify the rice and non-rice planthopper. Using the same experimental data, we compared the proposed method with the traditional image-based classification algorithms, SVM and BP neural network. The results showed that the accuracy and classification speed were 65.5% and 0.5 frames/s respectively for SVM, and 78.0% and 1.0 frames/s respectively for the BP neural network. In contrast, the proposed method improved the accuracy to 93.7% and the classification speed to 6.0 frames/s. [ABSTRACT FROM AUTHOR]

Ample sunshine, together with scarce rainfall and large variation in diurnal temperature, makes Xinjiang a unique place for producing tasty red jujube in China. Grading is an important parameter for storing and processing the jujube to maximize its market value, and needs to measure its volume and surface area. Traditional methods for measuring fruit volume are spheroid-like to measure the volume of water the fruit displaces when being immersed into water, with the surface area measured by peeling or slicing. These methods are inefficient and cannot be used for real-time measurement. The aim of this paper is to present a real-time multi-contour model to estimate the volume and surface area of the red jujube. We assessed the effect of contour angles, projection heights and diameters on the ultimate results. In the proposed method, 2D images of the targeted jujube were captured on a rotating circular table using a camera, and the contour of the images was then extracted using image processing. A 3D multi-contour model was developed based on the extracted 2D contour, and it was then used to estimate the volume and surface area of the targeted jujube. The result showed that the diameter of the target estimated by the multi-contour sphere model did not change, and that with an increase in the relative errors between the contours angle and the projection height, the volume estimated by the model increased (with the minimum relative error being 6.0%) while the error of the estimated surface area decreased (with the minimum being 1.0%). The angle between the contour and the projection height in the multi-contour sphere model had a prescribed value, and the relative error of the volume and surface area estimated by the model varied with the diameter in that the smaller the diameter was, the bigger the errors were. The average mean square error and the average relative error of the volume and surface area estimated by the model were 2.45 cm³ and 10.2%, and 3.65 cm² and 7.09%, respectively. An increase in grading appeared to increase the average relative errors of the estimated volume but had no noticeable impact on other factors. In summary, the multi-contour model for real-time measuring the volume and surface area of the red jujube offers an alternative to grading the jujube although further improvement is needed to reduce the errors. [ABSTRACT FROM AUTHOR]

Sausage is an emulsification-type, popular meat product, because of its unique flavor, high in nutrition and easy to store procedures. According to national standards of China (GB/T 20712-2006), sausage can be divided into three types of grades (general, excellent and premium). Traditional sausage grade detection methods are laborious and time consuming.So, it is imperative to develop a rapid and simple detection method. In this research, ultrasound imaging system was evaluated as rapid and precise detection method to differentiate between different grades of sausage. And, the texture of the sausage was also analyzed simultaneously. A total of 120 sausage samples from 2 different manufacturers were collected from local supermarkets of Zhenjiang, Jiangsu, China. From each sausage, small sample (2.5 cm×1.5 cm) were obtained for ultrasound imaging, moisture, starch and protein measurement. These measurements were utilized to divide sausages into general, excellent and premium quality grades. Ultrasound imaging system worked with the UTEX 320 equipment in pulse echo mode. The parameters of ultrasound imaging system were as follow: the pulse voltage; 300 V, the pulse repetition frequency; 800 Hz, the gain; 35 dB, and the scanning speed was 5 mm/s. Total of 240 ultrasound images (2 brands 3 grades, each had 40 samples) were collected by ultrasound imaging system. Images generated from different grades had obvious difference, however, different brands’ images with the same grade were similar. Grey level co-occurrence matrix (GLCM) was generated in 0, 45, 90 and 135° directions, respectively. The commonly used angular second moment (ASM), contrast (CON), correlation (COR) and homomorphity (HOM) were extracted in all directions, and a total of 16 texture feature variables were generated. Combined with the average average image (AVG), variance (VAR) of the image, 18 texture feature variables were finally obtained. Furthermore, the textural features of different grades had significant difference (P<0.05). All the texture feature values were randomly divided into calibration set (162 samples data) and prediction set (78 samples data) to build calibration model and predication model. Principal component analysis (PCA) was performed to simple variable because that texture feature always carried redundant data and examined the qualitative difference of these sausage grades using the first 3 score vectors. From the results of PCA, all the samples of sausage were divided into three classes according to the grade of the sample. However, the brand of the sausages failed to be distinguished. The 3 groups of different class of sausages were almost apart from each other in the space of the first 3 principal components (PCs), although there were some overlaps among the groups, because the emulsification-type meat product hardly to achieve well-distributed in each part. Results from PCA was in accordance to the results of image analysis. The first 3 PCs accounted for the all variations of 97.54%, representing all the information of the variables. Therefore, all the samples were divided into 3 classes based on different grades. The linear discriminant analysis (LDA), used in statistics, pattern recognition and machine learning to find a linear combination of features that characterizes or separates two or more classes of objects or events, and support vector machine(SVM), as a learning algorithm used for classification and regression tasks, was used to get the identification model. All the models relatively had high recognition rate. The identification results of the SVM were compared with the LDA. From the comparison, it showed that the discrimination accuracy of all the 3 classes of sausages using the SVM was up to 100% in prediction set and 100% in calibration set, respectively. From the results, it can be concluded that the ultrasound imaging technology can be used as a powerful and attractive tool to identify and discriminate different grades of sausages. The study could provide a reference for ultrasonic imaging technology’s appalication in discriminating different grades of sausages. [ABSTRACT FROM AUTHOR]