Search

Droplets deposition characteristics estimation is helpful to know the pesticides deposition distribution on crops, which is related to the crop spray quality directly, and especially the fast detection method will provide the basis for improving the pesticides spraying technology. Droplets image processing based on the water sensitive paper is one of the most common methods to detect the droplets deposition characteristics. The droplets coverage percentage and coverage density were selected to evaluate deposition distribution in this paper. With the development of science and technology, a smart phone was selected as image acquiring tool to replace the special image acquisition system, and the improved image processing algorithm of the water sensitive paper was developed. There are 3 steps in the image processing algorithm to obtain each droplet stain, which are image preprocessing, droplets area segmentation, and overlapped droplets segmentation. Firstly, the image was enhanced and calibrated by a calibration board, and then the image of water sensitive paper was segmented from the R channel and B channel image and transferred to a gray scale image. Secondly, the blue droplets stain area was segmented from yellow paper background, and the dynamic threshold method based on the pixel position was used to solve the problem of the influence of uneven brightness in this step, in which the gray mean value of an area of 25×25 pixels was calculated as the segmentation threshold for the middle pixel, and the droplets coverage percentage was calculated by the stains pixel number divided by the pixel number of water sensitive paper area. Thirdly, the circulatory segmentation method based on region circularity was designed to segment the multiple-droplet overlapped regions. Based on the statistical analysis, 0.6 was selected as the circularity threshold, less than 0.6 was considered to be overlapped droplets, and first erosion and then dilation based on a circle with the radius r were performed; next the segmented area circularity was recalculated, the area with the circularity of greater than 0.6 was selected again and segmented by the circle with the radius (r+1), and the erosion-dilation operation would be repeated over and over until the circularity of all stains was greater than 0.6. Finally, the identified droplets were marked as circle, and the droplets coverage density was calculated by droplets number divided by the water sensitive paper area. Experiments were conducted to test and verify the detection advancement of the proposed image processing algorithm. Experimental results showed that the dynamic threshold segmentation method is not affected by the uneven brightness and can extract 92.64% of droplets, and the droplet coverage percentage detection result is 12.57% and 8.74% greater than constant threshold and partitioned threshold method respectively. Moreover, the proposed overlapped droplets segmentation algorithm can segment successfully more than 2 droplets and the long and thin droplets, the accuracy of droplets identification is 97.2%, and the coverage density detection results showed that the relative error between the algorithm in this paper and manual counting is only 3.31%. The results indicated that the proposed image detection algorithm of droplets deposition characteristics is efficient and convenient, and can completely fulfill the demand of droplets deposition characteristics detection in the field, and the corresponding smartphone applications are in development. [ABSTRACT FROM AUTHOR]

Deep learning has been widely applied to the recognition and segmentation of plant leaf disease. However, traditional recognition models of plant leaf disease usually lack transparency, since these end-to-end deep classifiers are susceptible to shadow, occlusion, and light intensity. To address this drawback, in this paper, we proposed a new recognition and segmentation model of plant leaf disease based on deconvolution-guided, called Deconvolution-Guided VGGNet (DGVGGNet). An encoder-decoder architecture with symmetric convolutional-deconvolutional layers was applied to DGVGGNet so that the plant leaf disease recognition and segmentation can be carried simultaneously. Our model consists of three main phases: recognition, inversion, and deconvolution. In the recognition phase, we first fed the VGGNet with plenty of plant leaf disease images, and then Categorical Cross-Entropy Loss was utilized to train the recognition model. VGGNet was made up of the convolution layers of VGG16 and 2 fully connected layers, and the weights of convolution layers were pre-trained on ImageNet. Ten kinds of tomato leaf disease images in the PlantVillage dataset were used in this paper, concretely, 30% of the pictures were used for training, and the rest 70% were used for the test. Besides, on-the-fly data augmentation was also exploited during the training stage, such as flipping the images and corrupting the origin images by brightness, saturation, salt noise, and Gaussian noise. In the inversion phase, two fully connected layers were used to reverse the category prediction vector. Simultaneously, two skip connections were used to reinforce the decoder by adding the vector of the VGGNet’s fully connected layers. Then, the feature vector was reshaped to feature maps to input into the deconvolution module. In the deconvolution phase, the feature maps were fed into the deconvolution module to acquire the segmentation result of the disease area, where each pixel was trained with the Binary Cross-Entropy Loss. The deconvolution module consists of the upsample and convolution operation. Meanwhile, five skip connections were used to fuse the multiple features, which can refine the segmentation results. Besides, only a few samples of the training set were given pixel-level labels of plant disease to supervise the output of the deconvolution module. At the end of the deconvolution module, the reconstruction layer was used to smooth the segmentation edge. To explore the influence of the number of pixel-level labels used in the model, 9 and 45 pixel-level labels were used to supervise the segmentation results, respectively. To simulate the natural conditions, different kinds of interference were added to the test data, such as translation, fruit occlusion, soil occlusion, leaf occlusion, and brightness reduction with different percentages. Experimentally, we evaluate our recognition module by exploring the performance of VGGNet, DGVGGNet-9, and DGVGGNet-45 on different interference datasets, respectively. We also evaluate our deconvolution module by exploring 4 different evaluation metrics, i.e., PA, MPA, MIoU, and FWIoU, compared with 3 popular semantic segmentation models, i.e., FCN-8s, U-Net, and SegNet. The results show that DGVGGNet-45 has the highest recognition accuracy as well as with the highest PA, MIoU, and FWIoU among the four segmentation evaluation metrics, which are 94.66%, 75.36%, and 90.46%, respectively. Compared with VGGNet, the deconvolution module of DGVGGNet-45 can guide the recognition module to pay more attention to the actual area of disease, which is effective in improving the segmentation accuracy. The recognition results demonstrate that DGVGGNet had strong robustness in tough conditions. Furthermore, DGVGGNet only took 12 ms to identify a single picture on the GPU, which can meet the real-time requirements. [ABSTRACT FROM AUTHOR]

Purple soil which riches in mineral nutrients and is the main farming soil in Southwest China. Purple soil color images collected by machine vision have complex background, including crops, weeds, scattered small soil blocks and surface soil because of its stochastic field scenes. In order to avoid the interference of background on further processing and recognition of purple soil with machine vision, it is a primary task to segment the purple subsoil region from its background adaptively. Clustering algorithm has achieved good results in image segmentation and is widely applied. However, the selection of parameters of some classical clustering algorithms are sensitive, which need to be manually set and cannot satisfy to the adaptive segmentation of purple soil. Thus, aiming at the problem of adaptive segmentation, a segmentation methool of purple soil color image based on adaptive density peaks clustering was proposed in this paper. The specific segmentation algorithm was as follows: firstly, in order to calculate the density peaks conveniently and increase the separability between soil region and its background, a color to gray transformation method was carried out. The entropy-based similarity matrix was constructed, then the optimization model based on maximizing the between-class variance and minimizing the withinclass variance criterion was established with the similarity matrix. The optimization model was solved to obtain the gray matrix, which enhanced the separability of the gray value for clustering. Secondly, in the light of the shortcomings of density peaks algorithm, the density formula was improved and a measure was designed to determine the clustering centers adaptively, which realized the adaptive segmentation of purple soil region with the clustering algorithm based on the density peaks clustering framework, and improved the initial segmentation accuracy. Finally, post-processing algorithms of boundary extraction and region filling were designed to remove both the discrete small soil blocks in the background area and the internal voids in the soil area of the image obtained by initial segmentation. Therefore, the purple soil region image could be completely obtained and the segmentation accuracy of soil region was improved. From purple soil color images collected by machine vision in the field, 60 images with normal illumination, no surface soil and no shadows around the subsoil were randomly selected as normal test images and divided into 20 groups.we picked out randomly 60 images with the characteristics of normal illumination, then another 60 images were randomly selected as robustness test images, which were characterized by scattered subsoil and some shadows around the topsoil. These images were divided into 20 groups. The results of contrast test showed that the proposed method could automatically segment the purple soil color images of normal test and robustness test. For 60 normal test images, average segmentation accuracy of the proposed algorithm were 93.45%, which were 11.54, 7.96, 3.16 and 0.85 percentage points higher than that of FRFCM (fast and robust fuzzy C-Means) algorithm, H-threshold algorithm, DPC (density peaks clustering) algorithm and DFDPC (data field based density peaks clustering) algorithm, respectively. For 60 robustness test images, average segmentation accuracy of the proposed algorithm were 87.40%, which were 11.75, 5.2, 12.47 and 3.09 percentage points higher than that of FRFCM algorithm, H-threshold algorithm, DPC algorithm and DFDPC algorithm, respectively. The results proved that the proposed algorithm in this paper was superior to the other four comparison algorithms. Furthermore, boundary extraction algorithm and region filling algorithm further improved the average segmentation accuracy of soil region, in the post-processing stage. In the final exact segmentation results, average segmentation accuracy of normal test images were 96.30%, with average time-consuming of 0.36 s，and average segmentation accuracy of robustness test images was 91.63%, with average time-consuming of 0.35 s. In conclusion, the proposed algorithm was effective, and it can provide reference for the segmentation and extraction of purple soil region from the computer vision image. [ABSTRACT FROM AUTHOR]

Green crop is generally used as the foreground in image segmentation of agricultural visual navigation system because of its obvious chromaticity difference from the basic image background. However, for the segmentation of maize stubble row, there are many backgrounds in the field harvested by combine harvester, such as naked land surface, maize residues, which has little color difference with the maize stubble row. To achieve accurate and quick segmentation of maize stubble row, an image segmentation approach of maize stubble row based on genetic algorithm and threshold filtering noise was proposed in this paper. Firstly, the RGB color space was selected to accomplish the segmentation, which is a basic color space and widely used in machine vision automatic guidance systems. Secondly, to reduce the difficulty of image segmentation, the region of interest (ROI) was selected by calculating the maximum of column gray value accumulation. Thirdly, the gray-scale image of the maize stubble row was obtained through the optimized gray-scale operator. Besides, the genetic algorithm was often used in global optimization, which was used to optimize the gray-scale operator. Specially, the maize stubble row image’s pixel was divided into 3 classes: land surface, residues in rows and stubble tip incision, and the 3 classes were selected as the sample to optimize the gray-scale operator by genetic algorithm. Then, the segmentation of single threshold method was used to segment the gray-scale image. Furthermore, the segmentation of single threshold method is widely used in image segmentation because of its high efficiency and convenience. Lastly, the morphological corrosion treatment (MCT) and threshold noise filtering algorithm (TNFA) were applied to guarantee the integrity of the maize stubble row region and the noise points on both sides of the maize stubble row removal. In order to verify the effect of the method proposed in this paper, 200 test images were captured from Zhuozhou Experimental Station of Conservation Tillage Research Center of Ministry of Agriculture in Baoding City, China in each October during 2014－2016. The acquisition was often on sunny day, aiming at collecting images under different natural lighting conditions, different positions in maize stubble row field. The results showed that the average relative object area error (ARAE) by our method was only 24.68%, while the AREA of the iteration method and the OTSU method were 90.67% and 86.42%, respectively. The average processing time of a test image based on this paper presented algorithm was 0.16s, which was much more time-consuming than the OTSU method (0.07s), while the processing time of our method was less than the iteration method (0.25s). Therefore, the method presented in this paper achieved better performance than other methods in maize stubble row segmentation, and was effective to segment the maize stubble row in the complicated backgrounds. The presented method can provide precise basis for detection of guidance line in maize straw covering field. [ABSTRACT FROM AUTHOR]

Mechanical weeding can reduce the use of pesticides and is of great significance to ensure high yield of crops. Real-time and accurate identification of crops is a key technical problem needs to be solved in mechanical weeding equipment. Because of the subjectivity of feature extraction process in weed recognition, the accuracy of traditional methods in actual field environment is low. In recent years, the method of weed identification based on convolution neural network has been widely studied. Although the accuracy is obviously improved, there are still problems such as large parameters and poor real-time performance. In order to solve the above problems, a four-channel input image is constructed by collecting near infrared and visible images of sugar beet in the field, and a lightweight convolution neural network based on codec structure is proposed. In this paper, Sugarbeet and weed images collected from a farm in Bonn, Germany, in 2016 were used as data sets, which covered images of different growth stages of sugar beet, and 226 pictures of which were randomly selected as training sets, and the remaining 57 pictures were used as test sets. The experimental data set was composed of three channels of visible light image and one channel of near infrared image, which are merged into a four-channel image by pixel level superposition, and the depth-wise separable convolution was used in the deep model. Firstly, the input feature image was convoluted in 2 dimensions convolution kernel and the number of channels was expanded. Then, the 1×1 convolution kernel was used to make the 3 dimensions convolution which combined channel feature and compressed the channels to enhance the nonlinear mapping ability of the model. In order to avoid the problem of the gradient disappearing, the residual block was used to connect the input and output of the depth-wise separate convolution. Finally, the coding and decoder structure was designed and the shallow features were combined with deep features to refine the segmentation effect. Due to the imbalance of pixel proportions of soil, crops and weeds, the weighted loss function was used to optimize the model. The segmentation accuracy, parameters and operating efficiency of models at different input resolutions and different width factor were introduced to evalute the model. When the width factor was 1, the segmentation accuracy of the model increased with the increase of the input image resolution, the model accuracy of four channel input was higher than that of the model based on original visible image input, which showed that the near-infrared image features can compensate the defects of ordinary RGB images to some extent, and make the model more suitable for the dark environment. Under the same input image resolution, the model with a width factor of 2 or 4 performs better than the model with a width factor of 1. With the increases of width factor, the parameters of the model increase greatly. The amount of calculation is related to the size of the input image, so the frame rate gradually decreases with the increase of the size of input image. The experimental results show that the optimal model in this paper is a four channel input model with a width coefficient of 2, and the average intersection union ration is 87.58%, the average pixel accuracy is 99.19%, the parameters are 525 763 and the frame rate is 42.064 frames/s. The model has high segmentation and recognition accuracy and good real-time performance, and can provide theoretical basis for the development of intelligent mechanization weeding equipment. [ABSTRACT FROM AUTHOR]

Mechanical weeding can reduce the use of pesticides and is of great significance to ensure high yield of crops. Real-time and accurate identification of crops is a key technical problem needs to be solved in mechanical weeding equipment. Because of the subjectivity of feature extraction process in weed recognition, the accuracy of traditional methods in actual field environment is low. In recent years, the method of weed identification based on convolution neural network has been widely studied. Although the accuracy is obviously improved, there are still problems such as large parameters and poor real-time performance. In order to solve the above problems, a four-channel input image is constructed by collecting near infrared and visible images of sugar beet in the field, and a lightweight convolution neural network based on codec structure is proposed. In this paper, Sugarbeet and weed images collected from a farm in Bonn, Germany, in 2016 were used as data sets, which covered images of different growth stages of sugar beet, and 226 pictures of which were randomly selected as training sets, and the remaining 57 pictures were used as test sets. The experimental data set was composed of three channels of visible light image and one channel of near infrared image, which are merged into a four-channel image by pixel level superposition, and the depth-wise separable convolution was used in the deep model. Firstly, the input feature image was convoluted in 2 dimensions convolution kernel and the number of channels was expanded. Then, the 1×1 convolution kernel was used to make the 3 dimensions convolution which combined channel feature and compressed the channels to enhance the nonlinear mapping ability of the model. In order to avoid the problem of the gradient disappearing, the residual block was used to connect the input and output of the depth-wise separate convolution. Finally, the coding and decoder structure was designed and the shallow features were combined with deep features to refine the segmentation effect. Due to the imbalance of pixel proportions of soil, crops and weeds, the weighted loss function was used to optimize the model. The segmentation accuracy, parameters and operating efficiency of models at different input resolutions and different width factor were introduced to evalute the model. When the width factor was 1, the segmentation accuracy of the model increased with the increase of the input image resolution, the model accuracy of four channel input was higher than that of the model based on original visible image input, which showed that the near-infrared image features can compensate the defects of ordinary RGB images to some extent, and make the model more suitable for the dark environment. Under the same input image resolution, the model with a width factor of 2 or 4 performs better than the model with a width factor of 1. With the increases of width factor, the parameters of the model increase greatly. The amount of calculation is related to the size of the input image, so the frame rate gradually decreases with the increase of the size of input image. The experimental results show that the optimal model in this paper is a four channel input model with a width coefficient of 2, and the average intersection union ration is 87.58%, the average pixel accuracy is 99.19%, the parameters are 525 763 and the frame rate is 42.064 frames/s. The model has high segmentation and recognition accuracy and good real-time performance, and can provide theoretical basis for the development of intelligent mechanization weeding equipment. [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]

In this paper, a soil pore segmentation method based on fully convolutional network (FCN) is proposed to improve the accuracy of pore segmentation in soil image and provide technical support for the research of soil science. Taking the soil of typical black soil as the research object, the soil computed tomography image were obtained by scanning and cutting. Based on the FCN network, the soil image and the calibrated image of pore structure were input for convoluting, pooling and deconvoluting operations, and the error between the prediction image and the calibration image was used as feedback to complete the forward inference operation. Then, the weight value was updated by the back propagation algorithm to establish the soil pore segmentation model. Fully considering the pore geometry and spatial distribution characteristics, the pore model can accurately output the soil pore binary image. Meanwhile, the commonly used segmentation methods in the literature, such as Otsu method, watershed method, regional growth method and Fuzzy C-means method (FCM) were adopted for the comparative experiments on soil computed tomography images with low pore density (0-0.03), medium pore density (0.03-0.1) and high pore density (0.1-1) which were defined by porosity of soil. The experimental results showed that the watershed method and the regional growth method overestimate the pore structure of different geometries, including cracks between the pores, whereas the Otsu method and FCM method tended to overestimate the macropores and underestimate the micropores. Compared the five methods, the FCN method can accurately extract the pore structures with vary topologies from the complex soil computed tomography images with low, medium and high pore density. Moreover, the segmentation accuracy rate, over-segmentation rate, and under-segmentation rate were used to evaluate the soil pore segmentation performance of five methods. Based on 1487 soil computed tomography images, the average segmentation accuracy of FCN pore segmentation method was 98.1%, which was 25.6%, 48.3%, 55.7% and 9.5% higher than that of Otsu method, watershed method, regional growth method and FCM method. The average over-segmentation rate of the FCN pore segmentation method was 2.2%, which was only 33.8% of the suboptimal method (FCM method), respectively. And the average under-segmentation rate of the FCN pore segmentation method was 1.3%, which was only 23.6% of the suboptimal method (FCM method). In total, the FCN method can accurately extract the pore topology, restore the spatial distribution of pores and its application can make up for the shortcoming that the traditional segmentation method only uses the low-level features (gray and edge) when extracting the pore structure. Owing to the multiple convolution layers in the network, the FCN method can obtain the vary features of pore structure, so it has strong generalization ability and robustness of pore segmentation for different types of soil images. This paper will has a good reference for the microscopic process simulation , 3D reconstruction and soil structure analysis on the pore scale, and can provide a more intelligent technical method for soil science. [ABSTRACT FROM AUTHOR]

The efficiency and accuracy of corn ear test are two of the key factors restricting the breeding efficiency seriously. Corn ear test includes the measurement, records, statistics and analysis of parameters such as ear weight, ear length, ear width, number of ear rows, kernels per row, average thickness of kernel, kernels per ear. In this paper, a corn ear parameter extraction method based on 4 cameras was proposed based on the high-throughput automatic measuring device which has been developed previously. Four high-resolution color cameras were evenly distributed around the ear with the interval of 90° to get the corn ear images from 4 directions at the same time. Every image from the corresponding camera was processed including image preprocessing, projection model building, and parameters extraction of corn ear. During image preprocessing process, center part of the original image with the length of 7/9 of the original image length, the width of 1/2 of the original image width was chosen as the processed area. Binarization processing was applied to the area to obtain binary image, and the binary image was processed by image denoising, hole filling and other morphological transform. An AND-operation was then applied between the processing result and the original image to access the corn ear images without background. The projection model was constructed after image preprocessing process, which considered ear cross-section circular, and kernels were distributed on ear cross-section as point on the circumference of a circle. Thus, number of ear rows can be easily calculated according to the relationship between number of ear rows and circumferential angle of those rows. Procedure such as kernels area acquisition, kernels center position acquisition, kernels at edge removal, reserved kernels tracking and corn ear parameters calculation are operated based on the projection model. Since there are 4 images for each ear, the final ear parameters including ear length, ear width, average thickness of kernel, number of ear rows, kernels per row, kernels per ear are calculated based on parameters measured from each image. The ear length and width are represented by the maximum length and width of the smallest external rectangle of the 4 images. Number of ear rows in each image is calculated from the valid row number and the circumferential angle which can be obtained on the basis of the projection model. Kernels per row are acquired by tracking the kernel area for each ear image, the maximum number of kernels in a row for each image is calculated as well as the average value, and the round-of number is considered as kernels per row of the ear. Kernels per ear are calculated from the valid row number, kernel number of the valid rows and corn ear rows. Average thickness of kernel is calculated according to the tracked kernel number and the total tracking path. Experiments are carried out with the high-throughput automatic measuring device for corn, and results show that the determination coefficients (R2) of ear length, ear width and average thickness of kernel achieve 0.997 3, 0.984 and 0.997 3 respectively between the values obtained by the proposed method in this paper and that measured artificially. The measuring accuracies of number of ear rows and kernels per ear are 98.63% and 95.35%, respectively, which meet the requirements of corn parameters measurement during maize breeding. The proposed method also provides a new train of thought for the extraction of corn ear parameter, and it also lays a solid foundation for the realization of automatic high-throughput device for corn. [ABSTRACT FROM AUTHOR]

Ochotona curzoniae is an endemic species and key species in the alpine meadow of the Tibetan Plateau and also it is a main kind of organism that destroys the grassland ecology of the plateau. In order to prevent the dangers of the ochotona curzoniae, we need to study the living habits of ochotona curzoniae and investigate the degree of harm of ochotona curzoniae, and then we can control the number of ochotona curzoniae through the effective preventive measures. With the development of sensing technology and image processing, we can provide an objective basis through intelligent monitoring system to control the damage of ochotona curzoniae. The object detection of ochotona curzoniae is a key technology in the intelligent monitoring equipment because it can provide the object contour feature for behavior analysis of ochotona curzoniae. The object detection of ochotona curzoniae is very difficult, because the ochotona curzoniae video possesses the characteristics of complex background, low contrast, the object color with intensity inhomogeneity, diversity and mutability. The traditional object detection method cannot extract the object contours accurately. This paper presents a fast object detection method based on space-time domain. Firstly, the centroid position of the object in the current frame image is determined by the background subtraction, and then the rough segmented image and the initial contour are obtained based on the centroid position. The rough segmented image is segmented by the improved Chan-Vese model, and then we can obtain the object contours. In view of the fact that the level set function needs to be initialized in the process of improved Chan-Vese model, and the initialized computation is enhanced with the increase of the image scale, the centroid of the object is taken as the center to intercept the image containing the object as the roughly segmented image. Then, the improved Chan-Vese model is used to segment the roughly segmented image, so as to reduce the time consumption of Chan-Vese model segmentation. In addition, as Chan-Vese model can’t fully segment the image of object whose color is diverse and mutable, we use the improved Chan-Vese model to segment the roughly segmented image. The internal pixels of image evolution contours were processed by K-means clustering, and the clustering center point values were obtained. The internal fitting values of Chan-Vese model were constructed by the clustering center point values and the image mean filtered intensity information, thereby improving the adaptability of Chan-Vese model for complex object image segmentation．In addition, rectangular Dirac function was used to replace regularized Dirac function in the energy function of Chan-Vese model, and the calculation of level set evolution equation could be limited to the zero level set so as to avoid the influence of the image background disturbance on the segmentation result. In this paper, the video processing with 50 frames of images shows that the time consumption of this method is only 15.25 s, the average value of Dice similarity coefficient is 0.852 929, and the average value of Jaccard index is 0.744 57. In summary, the object detection method proposed in this paper can accurately extract the object contour and has a high real-time performance. [ABSTRACT FROM AUTHOR]

Effective recognition method of crop and weed is the basis for promoting the development of intelligent mechanization weeding pattern. Summarizing the previous research, we found that the accuracy and stability of the recognition model could be improved by natural and diversified feature presentation, but there are still 2 main problems. On the one hand, feature presentation of the natural property of target was difficult to be obtained by the hand-engineered feature extractor. The spatial consistency of the obtained features was bad, and the real-time performance of recognition system was reduced for the complex feature extraction algorithm. On the other hand, the effect of image preprocessing has important influence on recognition results, especially the overlapping segmentation of crop and weed. For overlapped objects, it is usually difficult to segment them without affecting their respective feature presentations, resulting in low recognition accuracy and stability. In order to solve the main problems in the current research, we explored the way to improve the recognition accuracy, stability and real-time performance, and a recognition method of crop and weed based on multiscale hierarchical feature learning combined with superpixels segmentation was proposed. The main research contents of this paper were as follows: 1) Excellent internal features of image are hierarchical. In this research, the multiscale hierarchical feature is a scene level feature with invariance and consistency in scale space. Multiscale convolutional neural network was built to extract multiscale hierarchical feature. Multiscale convolutional neural network contains multiple copies of a single CNN (convolutional neural network) that are applied to multi-scale Gaussian pyramid of the input image. The CNN model as feature extractor in this paper includes 3 stages. In the first 2 stages, it contains a bank of filters (convolution kernels) to produce dense feature maps, a point to point nonlinear mapping activation function, and a spatial pooling module for sub-sampling of each feature map, While the last stage only contains a bank of filters. Each filter is applied to the input feature maps through 2-dimensional convolution operation, in which local feature presentations are detected at all pixel locations on the input image. For each pixel, the CNN model is used to collectively encode the internal information in a large sense region around the location of given pixel. The CNN is fed with raw pixels and trained with back propagation method. With complete training, this CNN model can automatically extract hierarchical feature representations from the input image, thereby decreasing the need for hand-engineered features extracting. A series of feature maps for multiscale regions centered on each pixel of the image are produced by the multiscale convolutional neural network; these representations contain shape, texture and sense information. Therefore, the multiscale hierarchical feature is learned to allow the recognition of the class of all pixels in the image. The average pixel recognition rate is 93.41%. 2) In this research, multiscale hierarchical feature is used to produce the class distribution for every pixel through a 2-layer MLP (multi-layer perceptron). But recognizing the class of pixel independently from its surrounding regions could produce some interference at the boundary of targets. Accurate boundary segmentation for each target in the image is not provided. In this paper, the superpixel method is used for generating an over segmentation of the image, where each segmentation component (superpixel) is an irregular pixel block composed of similar pixels in texture, color and brightness characteristics. Object in the image is composed of these superpixels, and has the exact original boundary. The superpixels segmentation is computed following the simple linear iterative clustering method described in this paper. 3) We proposed the recognition strategy of multiscale hierarchical feature learning combined with the superpixels segmentation. Firstly, an over segmentation of the original image was produced through the superpixels segmentation. At the same time, each pixel location of the image was classified densely based on the multiscale hierarchical features. These predictions of pixels in each superpixel were aggregated to produce the class prediction of superpixel, through computing the average class distribution within the superpixel. Adjacent superpixels with the same class were merged to obtain the final target class prediction and image segmentation. The accurate image segmentation was achieved while recognizing the target in the image by this recognition method, which effectively avoided the problems caused by targets overlapping, and the recognition results were more stable and accurate. Maize seedlings at 2-5 leaves stage and weed during the same stage were used as research object, and the recognition method of multiscale hierarchical feature learning combined with superpixels segmentation was tested. The results showed that the average target recognition rate with this method was 98.92%, and the standard deviation was 0.55%. The average target recognition rate with the method in previous research was 98.36%, and the standard deviation was 1.05%. So, the accuracy and stability of recognition results by this method in this paper were improved with different degrees. In aspect of real-time performance, the average time to recognize a single image was 1.68 s with this method, and compared with the method proposed in previous research, it was decreased by 1.58 s. The real-time performance of this method can be further improved by GPU (graphics processing unit) hardware acceleration, and the average time to recognize a single image was only 0.72 s. Therefore, the recognition method based on multiscale hierarchical feature learning combined with superpixels segmentation can effectively achieve accurate, stable and efficient recognition of maize and weed. The research results provide reference for the development of precision weeding. [ABSTRACT FROM AUTHOR]

Image segmentation algorithm can effectively extract target information, thus providing an important precondition for the application of intelligent agriculture. How to retain the external features of field crops as well as meet the requirements of farmland work speed in a complicated and changeable farmland environment is one of the important problems to be solved in the current farmland image segmentation algorithm. At present, color segmentation method is the most commonly used one for field crop image segmentation. There are obvious color feature differences between crops and soils, so the most intuitional color information of the image can be used for effectively extracting information of the crops. However, the commonly used color model can keep the shape of the crop but can't meet the requirements of speed and different illuminations at the same time. In recent years, machine learning methods have been applied in the crop image segmentation in some studies to improve the traditional segmentation algorithms. Though these methods can segment field image more precisely in different light conditions, the algorithm results rely on a data training process which needs a long calculation time, so it is difficult for these methods to be applied in real-time work. Therefore, this paper proposes a field crop image segmentation method based on reverse mutation - particle swarm optimize (RM-PSO) algorithm to solve the mentioned above problem. This method is divided into offline and online parts. The optimal coefficients of color are obtained by training data samples offline. The coefficients can be used to segment field image in real time. The method proposed in this paper can improve the process speed. The paper focuses on offline part and puts forward a strategy of RM to improve the PSO algorithm by training sample data. Firstly, the reverse particles are generated from the initial ones by using the reverse space strategy, and we compare the fitness of the reverse particles with that of the initial ones to eliminate the particles of poor quality and construct a new initial particle swarm, thus improving particle diversity and the quality of initial particles. Secondly, when the evolutionary particles are in a passive state during the process of particle iteration, we mutate the passive particles to avoid premature convergence of the algorithm and running into local optimum. Thirdly, we introduce a satisfaction function to evaluate the optimal color coefficient of a single sample in the color searching space, and take the global optimal color coefficient as the final result. As for the online part, the images collected in real time will be processed with the global optimal color coefficient, the between-class variance of the processed gray-level images will be calculated with the automatic threshold segmentation method, and then the maximum between-class variance threshold will be used as the optimal one to obtain the segmenting images. By comparing this method with frequently-used color segmentation methods, the experimental result shows that under different light conditions and interference in different sceneries, the method in this paper can make crops maintain complete external feature while ensuring the processing speed. According to the statistical data of 30 test images, for the algorithm segmentation accuracy, the method in this paper has an average wrong segmentation ratio of 4.8%, which is lower than the HSI (hue, saturation, intensity) algorithm (11.3%), the excessive green algorithm (19.5%) and traditional PSO (7.6%); for the algorithm stability, the algorithm in this paper has a wrong segmentation ratio standard deviation of 3.1%, which is lower than the HSI algorithm (7.2%), the excessive green algorithm (14.7%) and traditional PSO (7. 9%), so it is of higher reliability; for the processing speed, the method in this paper takes nearly the same processing time as the excessive green algorithm and traditional PSO, namely 0.311, 0.303 and 0.319 s respectively. Therefore, the method in this paper provides a reliable technology basis for field crop identification and possesses high practical application value. [ABSTRACT FROM AUTHOR]

The breeding environment plays an important role in healthy growth and development of the pigs, and it is also a guarantee of the excellent traits. The growth of pig is the integrated result of their own traits and external environmental factors. For different environments, pigs will show different behaviors. It is found that the appropriate growth environment can be reflected by their degree of dispersion in the pens. When the environment is suitable, the pig group will lie down together. When the temperature is low, they will huddle together; and when the environmental temperature is high, the pigs will scatter around. Image segmentation is the important link to analyze pig behavior status based on computer vision technology, and pig image segmentation is an image representation that provides easy-to-understand and analysis for the behavior feature extraction, parameter measurement, image analysis, pattern recognition of individual pig. The accurate and effective image segmentation algorithm is for pig behavioral intelligent analysis, understanding and environmental intelligent decision. The traditional Graph Cut algorithm uses the energy function to compute all the pixels in the image, it takes a lot of time. It has the low segmentation efficiency and it cannot accurately segment specific pig. Aiming at the limitation of traditional Graph Cut algorithm and the characteristic of image itself, on the basis of the interactive watershed algorithm, in this paper, we proposed an interactive image segmentation method based on improved Graph Cut algorithm aiming at better use of interaction information provided by the user and effectively control of the number of super pixels. To some extent, the algorithm could also avoid the problems of over segmentation or under segmentation. The image was divided into regions based on the interactive watershed algorithm, and the foreground and background were specified. Each region block was regarded as the super pixel, the gray scale of super pixel was used as the vertex. Establish edges between adjacent vertices and the new network map was constructed instead of the traditional weighted graph. In order to reconstruct the energy function to complete the effective segmentation of the foreground and background, the maximum flow, and minimization cut algorithm were used to cut the image. Results of experiments showed that the method can effectively segment pig target, it had high segmentation quality and high efficiency. In this paper, the peak signal to noise ratio (PSNR) and structural similarity (SSIM) were used as objective evaluation standard. The range of the PSNR was [30, 40] and the range of the SSIM was [0.9, 1] in this method, which was the most accurate segmentation algorithm in the shown segmentation algorithm. It indicated that two evaluation criterions were consistent with the subjective evaluation, and the stability and the reliability of this method were improved. The average time consumed in this algorithm was only 33.7% of in the traditional Graph Cut algorithm. It greatly improved the efficiency of segmentation, and it had highly robust for the segmentation of specific target pigs. The results of this paper can provide technical support for subsequent target identification, detection, tracking and monitoring, evaluation of external environments, such as alarms. [ABSTRACT FROM AUTHOR]

The application of multi-rotor unmanned aerial vehicle monitoring system for forest pest information collecting has many advantages, such as low running cost, operating flexibility, easy access to data, high image resolution etc. It has been regarded as a quick access to forest insect pest information collecting. By use of unmanned aerial vehicle system, valid segmentation and extraction of pest images acquired with the help of multi-rotor unmanned aerial vehicle can be used to calculate the insect pest proportion in monitored sample field. It can provide forest conservation experts with evidence for assessing the insect pest damage. To conduct forest monitoring work and calculate the proportion of pest infested area in monitored sample field with more preciseness and fast turnaround, in this paper, we aimed to solve poor time response circle and limited monitoring range problems that exist in current forestry information monitoring method. Firstly, in this paper, we built both hardware and software systems of multi-rotor unmanned aerial vehicle. Aerial vehicle equipped with image collecting devices was used to monitor in forestry pest insect infested area and collect data in the Liaoning testing forest. In order to obtain proper resolution images, aerial vehicle took off the center of the chosen monitoring area vertically to collect photo resources. By considering needed resolution requirements on image segmentation comprehensively, the height of about 50 m was chosen for image acquisition. On the analytical basis of monitoring images, an image segmentation method based on composite gradient watershed algorithm was proposed. This method introduced global histogram equalization to eliminate the influence of dark texture and adopted the morphological hybrid open-closing reconstruction filter to complete the denoising work of the image samples, eliminate the image interference to the segmentation effect, and suppress the over-segmentation phenomenon in image segmentation process. The gray-scale image was obtained by gray-scale transformation of the pre-processed image. The non-correlation regions (road and bare ground) were extracted by calculating the composite gradient of each pixel point in the gray image. Interference to the segmentation result may arise in segmenting process due to the similar color of non-correlation region and pest insect infested area. In this paper, the mentioned region was removed from the original image, which greatly avoided the interference of the non-related region to the pest area and ensured the accuracy of the result. Finally, the watershed algorithm was applied to realize the segmentation and extraction of insect pest area in images. In order to verify the effectiveness of the proposed method, the traditional watershed algorithm and K-means clustering algorithm were used for comparing experiment methods in the segmentation of eight images with different levels of insect pest. With the help of mage segmentation device, the accurate pest insect infested area was labeled manually, and it was taken as reference value in pest insect proportion calculating step. The experiment result showed that the segmentation effect was much more similar to the manual operation result. Specifically, the relative error rate decreased by 6.56% and 3.17% and the relative limit measuring accuracy was improved by 7.19% and 2.41% in this proposed method when traditional watershed algorithm was compared with K-means cluster algorithm. Our result showed that multi-rotor unmanned aerial vehicle was helpful in real time and effective monitoring of forest pest insect. The algorithm proposed in this paper was able to accurately segment and extract pest insect area in monitoring images and the proportion of pest area in whole sample fields was acquired, thus providing valid data support for forest pest monitoring and preventing work in the future. [ABSTRACT FROM AUTHOR]

Cotton canopy coverage is an important index for monitoring cotton growing in field. It is easy but not accurate to calculate, because it is difficult to accurately segment the cotton canopy in the complex environment image of cotton field. This paper presents an adaptive threshold segmentation approach of cotton canopy image based on logistic algorithm in order to improve the segmentation precision and robustness for cotton canopy image. Firstly, the cotton canopy image is transformed into HSV (hue, saturation, value) color space. This color space is designed by human color description. In this color space, the color feature of the pixel can be expressed by 3 independent components i.e. H, S and V. In this paper, the logistic regression algorithm is used to compute threshold used in image segmentation. The logistic regression algorithm is often used in 2 kinds of classification problem, so our method need an artificially defined variable. This variable and a single color feature variable can form a dataset as the input of logical regression algorithm to calculate the segmentation threshold. In our paper, the proposed artificially defined variable is set to a specific value that is 1, the effect of which is to reduce the impact for computed segmentation threshold. The cotton canopy image’s pixel is divided into 2 classes: target and background. The H channel feature of 2 classes can be extracted in HSV color space, and the green ratio (G/(R+G+B)) of 2 classes can be extracted in RGB (red, green, blue) color space. Those features’ thresholds are computed by logistic regression algorithm. H channel thresholds are used to achieve the first segmentation. Secondly, the first segmentation result is divided to highlight pixels and low pixels. The highlight pixels mainly include light canopy and light soil, and the low pixels mainly include shadow canopy and shadow soil. However, it is difficult to segment cotton canopy in the low pixels. In order to solve this problem, extra-green (Exg) color feature is used as segmentation feature to get cotton canopy in the low pixels. Thirdly, the highlight pixels in the first segmentation result and the low pixels segmented by Exg threshold are segmented by green ratio threshold. This segmentation is called the second segmentation. At last, the segmentation result of cotton canopy is acquired by morphology repair operation, and it ensures the integrity of the canopy region and the independent noise removal. In order to verify the effect of the method proposed in this paper, 320 test images were captured from the cotton producing areas in Xinjiang, China from April to July 2016. The acquisition was often on sunny day, aiming at obtaining images under different lighting conditions, different positions in cotton field, and different cotton growth periods. These images were collected by the Canon EOS5D digital camera with 6 912×3 416 pixels, and zoomed into 1 728×1 152 pixels to improve segmentation effect. This algorithm programming development environment is Python 2.7, and OpenCV 2.4.9. The experimental results show that the average relative object area error (RAE) by our method is only 5.46%, the Exg feature OTSU method 11.78%, the four-component segmentation method 24.11%, and the saliency segmentation method 36.92%. The overall average matching rate by our method is 93.07%, the Exg feature OTSU method 76.43%, the four-component segmentation method 71.67%, and the saliency segmentation method 66.92%. The average processing time of this paper proposed method was 4.63 s, which was much more time-consuming than the super-green characteristic OTSU method (3.84 s) and the four-component segmentation method (2.56 s), but this time less than that of the segmentation method (6.25 s). Therefore the proposed method in our paper has better performance than other methods in cotton canopy segmentation task, and is effective to segment the cotton canopy in the complicated background and different cotton growth periods. The proposed method can provide certain basis for implementation of cotton growth condition automatic monitoring. [ABSTRACT FROM AUTHOR]

In the planting environment, it is difficult to detect the ripe mulberry targets automatically by machine vision because of the small shape, irregular distribution of fruits, non-uniformly illuminating condition and complex background interference. This paper presents a visual segmentation approach of ripe mulberries based on frequency-tuned (FT) saliency map and the pulse coupled neural network (PCNN) model in order to improve the segmentation precision and robustness for ripe mulberries. Firstly, the captured mulberry image is transformed into Lab color space. This color space is designed by the perception of the human eye in the natural color. Consequently, the color feature of the target can be expressed by 3 independent components i.e. L, a, and b. Next, the FT saliency map of mulberries can be computed by a multiscale edge detector. And it typically distinguishes differences between the region of ripe mulberries and the complex background in Lab color space. This saliency is illustrated as the difference matrix which includes the Euclidean distance of each pixel between the Lab vectors in the Gaussian filtered image and the average Lab vectors in the original image. Thereafter, the hue feature is extracted from the HSI (hue, saturation, intensity) color space of the captured mulberry image, and then equalized by contrast limited adaptive histogram equalization (CLAHE), which can enhance the color contrast of the mulberry image without being affected by illumination changes. And the feature image of mulberries is generated by fusing the FT saliency and the hue feature, as an input of the improved multilevel threshold PCNN model. Furthermore, the region of ripe mulberries can be detected by this PCNN model, which is a bio-inspired neural network and derived from synchronous dynamics of neuronal activity in a mammal visual cortex. This model is able to cause the adjacent neurons with similar inputs to pulse synchronously, and be appropriate to segment the small target from the complex background. Finally, the segmentation result of the ripe mulberry is acquired by morphology repair operation, and it ensures the integrity of the target region and the independent noise removal. In order to verify the effect of the method proposed in this paper, there were 200 test images captured from the mulberry ecological park of the Tianfu Town, Chongqing City, China in May, 2016. The acquisition time was often on sunny day and cloudy day, aiming at obtaining images under different lighting conditions, such as balanced illumination, imbalanced illumination and backlight. These images were collected by the Canon EOS70D digital camera with 3 648×5 472 pixels, and zoomed into 720×1 280 pixels to apply to the remote intelligent monitoring system of mulberry growth. The algorithm programming development platform is Visual C++ 2015. The experimental results point out that the average misclassification error (ME) by our method is only 1.87%, infinitely superior to the OTSU method with FT saliency (17.73%), the K-means method (10.69%), the improved multilevel threshold PCNN method based on Itti saliency map (7.34%), and the PCNN method based on graph-based visual saliency (GBVS) image (5.83%). The average segmentation time of a test image based on our method is 2.562 3 s. Consequently, our approach is effective to segment the ripe mulberries in the complicated background and different lighting conditions. [ABSTRACT FROM AUTHOR]

Based on the data measured by instruments, the plant three-dimensional reconstruction is an important part of the plant digital research. Through the establishment of the plant three-dimensional model, not only can the growth rule of the plants in real environment be researched by measuring the crop parameters quickly, but also the spatial structure of local visual scene of plant can be explored by analyzing plant leaf distribution feature. In order to construct three-dimensional structure of situ strawberry plants precisely for further study in plant spatial structure, the paper took strawberry plants of elevated cultivation environment as the research object and proposed a three-dimensional strawberry canopy morphology reconstruction algorithm based on multi-source image contour segmentation. We divided the main algorithm into 3 parts: multi-source image pre-processing, coarse segmentation of intensity image and model fitting. To make full use of the advantages of color images in color segmentation, color image and intensity image in different resolution are registered and merged. The paper used the improved multi-source information fusion algorithm of strawberry plants based on feature. By using feature-based multi-source information fusion algorithm, feature information of each source image is extracted and analyzed. Then the invariable feature points are selected. Later by checking the similarity of the feature points and adding appropriate parameter constraints, the registration information is obtained. Multi-source image mapping relationship is established by applying registration information. Then by fusing the preprocessed images, the information complement of color image and intensity map is realized and finally the intensity image to be split is obtained by the image preprocessing. Calculating local center of the vector field of intensity image to be segmented means calculating vector field for each pixel. Then the largest local pixels are picked out. Later the vector direction of each pixel is divided into 3 categories by a symbolic function. Clustering potential scattering point set in an array and the local control point are determined by applying a given threshold. Finally by applying the active contour model of the parameters and the central control point to the segmented intensity image we get a coarse segmentation image of the blade. A method of model reconstruction based on surface fitting was proposed for further processing. Intensity image’s segmentation contour is regard as the edge contour, and the contour interior point cloud is extracted. The method of region marking is used to mark the used point cloud which belongs to original depth point cloud data, and by checking the number of unmarked point clouds we can know whether the extraction is completed. The plane fitting selection mechanism based on point cloud is designed to compare the minimum mean square deviation of the surface and the plane after fitting, and the optimal fitting model is selected. All the optimal models are displayed in a coordinate system, and the points are colored one by one. Finally, the reconstruction and display of the three-dimensional model of strawberry are finished. To verify the effectiveness of the algorithm, the paper took the distance difference between average single-leaf length and leaf distance as an evaluation index. Experimental results showed that the correction rate of number of blades was 85.6%, that of single-leaf model was 88.4% and that of distance difference was 82.4%. The results can be applied to the spatial position measurement of in situ strawberry plants. The research provides a new method for the construction of plant spatial structure in local vision scenes of agricultural robots. [ABSTRACT FROM AUTHOR]

The amount of wheat ears per unit area is an important parameter for assessing wheat yield and wheat planting density. Generally, phenotype parameters should be obtained by manual count technique which is time-consuming and needs great effort. Aiming at the traditional image segmentation method based on color feature, texture feature and Haar feature, in this paper, we proposed an improved K-means algorithm for estimating the number of wheat ears. This method established a direct mapping relationship from the low-level features of the image to the number of wheat ears in the image through the color feature so that the target did not need to be segmented or detected. This type of method was more suitable for complex lighting and dense wheat ears, and had higher computational efficiency than counting methods based on image segmentation. This method made full use of the color feature of wheat ear image, took the area feature of local region extracted from local region as the basis of wheat ear judgment, and used the number of sub-regions in clustering region as the estimation value of wheat ear number, thus avoiding the task of target detection and location, and greatly improving the accuracy of Wheat ear counting. According to classification results of K-means method, the wheat ear image was divided into three regions. The green area pixel value representing the wheat eat region was set to 255, the pixel value of other areas was set to 0, so the binary image of wheat ears was obtained. Most of the binarized areas in the image were wheat ears, and a small part of the binarized area that was too small or too large was caused by the leaves. Following the steps were used to count wheat ears: 1) extracting the connected regions of the binary image and labeling them; 2) calculating the area of each connected region, the area was represented by the number of pixels in the region; 3) using the following method to filter out the connected region where the area was too small or too large; 4) finally, the number of wheat ears was obtained by counting the retained binary regions. This method established a direct classification relation from the image color feature of the lower layer to the image containing wheat ears so that the image did not need to be segmented or detected. Based on color feature clustering, this method could estimate the number of wheat ears in local area of space, and it was more scalable without training. The average prediction precision of the total wheatear number in a wheatear image for 12 wheatear images was 94.69%. And the Statistical error of the total wheatear number was between 2.17% and 7.41%, the average statistical error was 5.31%. The statistical experiment results showed that the accuracy of this algorithm to wheat ear counting was better than the traditional method based on image color feature and texture feature. [ABSTRACT FROM AUTHOR]

In order to segment cotton from background accurately and quickly under the unstructured natural environment, an improved scene adaptive optimization segmentation algorithm was proposed. Firstly, the traditional Otsu algorithm was adopted for cotton segmentation by analyzing the rule of gray histogram distribution of images acquired. During each segmentation of cotton image, the background pixel was set to zero and the gray histogram distribution of the target pixel was analyzed after segmented. The segmentation stopped when the histogram distribution appeared to be a steep unimodal distribution and the final segmentation threshold was the optimized threshold. There were still some background pixels in the segmentation results in the first step. Therefore, secondly, a threshold selection rule was utilized to segment cotton from background based on the statistical analysis of the RGB (red, green, blue) values. Cotton pixels were then extracted after morphology processing. Thirdly, cotton pixels with the RGB values were labeled as positive samples, and then background pixels with the same number of cotton were labeled as negative samples. The feature vectors which the positive and the negative samples shared in common were removed to improve the accuracy. A large-scale training samples were obtained automatically by using previous steps. In the fourth step, the samples were used to train the ELM (extreme learning machine) classification model for cotton segmentation. In the fifth step, the RGB values of all pixels in the test images were normalized and passed to the classification model and the output value will be 1 if the pixels belong to the cotton, or 0 if the pixels belong to the background. Another morphological procedure was employed to remove the noise segments of the output results. In the final step, the centroids of the connected regions which had a roundness of greater than 0.3 were extracted. If the pixel distance between 2 centroids was less than 40 pixels, a new centroid position was returned as a two-point center, so that some incomplete parts separated by the leaf were connected. Then the unsupervised sampling and segmentation of cotton under the unstructured light environment were achieved and the coordinates of cotton were derived. The algorithm proposed in this paper was verified by cotton segmentation experiment. The original datasets were collected on both sunny and cloudy days in the cotton breeding station located in Sanya, Hainan, China. The training datasets were composed of labeled samples generated from 20 images, 10 from sunny day and 10 from cloudy day. Another 60 images, 30 from sunny day and 30 from cloudy day, constituted the testing datasets. For each image in testing datasets, the accuracy is determined by the comparison of numbers between identified cotton and actual cotton. By applying the proposed method, the cotton was identified after segmentation. The average processing time is 0.58 s and the average recognition rates for the images on sunny and cloudy day were 94.57% and 97.98%, respectively. A comparison analysis was conducted between the proposed method and popular methods under Windows 10 64bits with an Intel Core i7-7400 CPU 3.00 GHz, 12 GB RAM, and MATLAB R2016a. For the segmentation of images on sunny day, accuracies by applying the proposed method, SVM (support vector machine) and BP (back propagation) with only RGB features as training samples were 94.57%, 77% and 83.35%, respectively. And the segmentation time was 0.58, 177.31 and 6.52 s, respectively. Accuracies by applying the proposed method, SVM and BP with Tamura texture and RGB features as training samples were 75.68%, 80.54% and 86.17%, respectively. And the segmentation time was 467.87, 678.35 and 551.01 s, respectively. The segmentation results of images on cloudy day were much like that on sunny days except slightly higher in accuracy. The results indicate that the proposed method exceeds popular methods in segmentation speed and accuracy. Moreover, the proposed method avoids texture feature extraction for each pixel, which guarantees better real-time performance. [ABSTRACT FROM AUTHOR]

The phenotypic traits of corn ear are important quantitative data in maize breeding and variety identification. In tradition, breeding workers are employed to deal with lots of corn ears by means of manual measurement and visual count, however this process is seriously labor-consuming and time-costing, and the measured traits are prone to be subjective and incomplete. In recent years, some semi-automatic systems based on machine vision and image analysis have been developed and applied to the maize variety test, however fully automated test system is still a challenge task owing to the strict high-throughput and high-precision requirements in large-scale maize breeding. To balance efficiency and accuracy of variety test for corn ears, in this paper, a high-throughput phenotypic measurement method and system based on panoramic surface image is proposed. Firstly, a novel mechanic system is proposed, which automatically conveys corn ears above a chain-roller structure, while the rolling corn ears are continuously imaged by a fixed industrial camera that is perpendicular to the moving plane of corn ear. In only several seconds, dozens of side images in which corn ears are in different positions can be collected to generate the image dataset of single corn ear. By analyzing the movement state of corn ear, a transformation model which describes the relationship among ear roll, camera imaging and surface position is then built to bridge the image sequence and the panoramic surface image of corn ear. Corn ears in the image sequence are respectively segmented and the center axes are dynamically determined by figuring out the shape and bounding box. This model always extracts the most appropriate sub regions of corn ear from image sequence, and then stitches them to the calculated positions on the panoramic surface image. As a result, the panoramic image of corn ear demonstrates the three-dimensional surface information in a two-dimensional image, and thus provides more intuitive and complete way for phenotyping calculation of corn ear. The valid surface region of corn ear in the panoramic image is further determined by the boundary detection technique that is performed by evaluating the perimeters of corn ear in the image sequence. Robust kernel segmentation based on hierarchical threshold method is also utilized to extract all candidate kernels which satisfy area and shape constraint, and some more restrictive filters based on machine learning methods, such as SVM (support vector machine), can also be taken to evaluate the validation of kernels. The segmented kernels in the panoramic image are used to calculate the total kernels, number of ear rows and kernels per row. The experimental results show that the proposed method and system can achieve optimized efficiency and accuracy balance. High-throughput convey mechanism improves the efficiency of image acquisition to 15 ears per minute. Compared with the methods based on single and multiple images, the variety test method based on panoramic surface image can make full use of the entire surface information of corn ear and reveal its individual phenotypic traits. The computation accuracies of ear length, ear diameter, number of ear rows, kernels per row and total kernels are up to 99%, 91.84%, 97.15%, 98.89% and 95.37% respectively. [ABSTRACT FROM AUTHOR]

Wheat is an economic crop correlated with national lifeblood, and its yield has a direct impact on people’s living standard and economic development, while the occurrence of disease is an important cause of crop yield decline. There are many kinds of crop diseases. Timely detection of disease types and corresponding prevention and control are urgent requirements to reduce the risk of crop yield decline. The disease segmentation is the priority among priorities of disease detection, and segmentation of lesion information is a prerequisite for disease identification, discrimination of disease degree, and pesticide application decision. The picture of wheat taken under natural conditions is greatly affected by the environment. The main obstacle of image segmentation is to find interesting parts in complex background. At present, the RGB (red, green, blue) sub region component segmentation method is usually used for image segmentation, and then the results are obtained by using some merging methods, but there is a large amount of computation in the segmentation of sub region components. For the wheat lesion segmentation, there exist the problems of noise and lesion edge being not clear. The research on wheat lesion image segmentation algorithm shows that the general image segmentation method has poor adaptability and compatibility, and other methods of mixing is difficult to achieve the desired results. Fuzzy edge detection with strong adaptability is the first algorithm to solve such problems. The traditional fuzzy edge detection method is first-order differentiating the preprocessed images, and edge detection is realized by edge discontinuity. Aiming at the disadvantages of the traditional algorithm such as high error rate, easy to lose the weak edge information, an improved image threshold segmentation algorithm based on fuzzy edge detection is proposed in this paper. In the aspect of image preprocessing, after analyzing the shortcomings of the traditional fuzzy edge detection, 2 improvements have been made to the algorithm. Gradient inverse weighted average filtering method is modified for the removal of noise and lesion of wheat, and numerical hierarchical improvement is made to multi-level fuzzy algorithm to enhance the edge information of the lesion. In the threshold segmentation algorithm, parameters directly influence the efficiency of image segmentation, so the level of detail segmentation on wheat spot shape can rely on the regulation of 2 aspects: One is the threshold, and the threshold value is influenced by relative pixel gray difference control; the other is the data involved in the calculation, and the data are related to the time of calculation. Reducing the participation data is the main method to improve the efficiency of the segmentation algorithm. An improved threshold segmentation method for maximum inter-class variance ratio is proposed. Based on the enhancement of image edge, threshold segmentation is applied to improve the threshold selection method. We use improved new formula to classify 2 kinds of variances and improve the overall performance of threshold segmentation from 2 aspects. The traditional threshold segmentation algorithm is improved, which is used to extract wheat spot shape feature from the wheat spot image. Compared with the traditional threshold segmentation algorithm, the improved algorithm based on fuzzy edge enhancement and threshold segmentation achieves an average accurate segmentation rate of 98.76%. The improved algorithm highlights the lesion edge information, and has the advantages of high segmentation efficiency and low noise. The noise ratio is reduced by 8.36 percentage points, and the time consuming is reduced by 0.331 s, which provides a reference for the improvement of image segmentation method. In the process of image segmentation, the improved algorithm is used to segment the wheat disease images, and the average multiple segmentation results are used as the parameters of final segmentation result. From the comparison results of 3 wheat lesion segmentation pictures, it can be seen that the improved algorithm is more meticulous to reflect the morphological characteristics of wheat disease, while retaining the edge information of wheat disease. [ABSTRACT FROM AUTHOR]

The behaviors of a lactating sow reflect welfare and health that affect piglet survival and growth during lactation. Computer vision has been widely used to perceive the behavior of animals for precision husbandry, which is useful to increase the productivity and reduce the disease rate. Effective and accurate segmentation of individual lactating sow is a vital step to record and analyze the lactating sow behavior automatically. However, under real pigsty conditions, it is a challenge to segment lactating sow from the background due to occlusion, uneven color on sow body surface, variations of sow size and pose, varying illumination and complex floor status. In this paper, we proposed an algorithm for lactating sow image segmentation based on fully convolutional networks (FCN). To design FCN for accurate segmentation, VGG16 was chosen as a basic network where the fully connected lays were converted to convolutional layers, and the FCN-8s skip structure was designed by combining semantic information from a deep, coarse layer with appearance information from a shallow, fine layer. We called this network FCN-8s-VGG16. The steps of our work were as follows: First, top view images were taken from 28 pens of pigs under a real pigsty circumstance and a total of 4 334 images were obtained, of which 3811 training images were selected from images of 7 pens and 523 test images were selected from images of the other 21 pens. And, all the images in training set and test set were manually labeled. Second, adaptive histogram equalization was used to improve contrast in training images. Then, the pre-processed training set was fed into FCN-8s-VGG16 to develop an optimum FCN model by the fine-tuning of the network parameters using Caffe deep learning framework on an NVIDIA GTX 980 GPU (graphics processing unit). After that, test set was put into the trained model to obtain the segmentation results. Then, to fill holes within objects and remove small objects, a post-processing was performed by using a disk structure of mathematical morphology and calculating the areas of connected regions. Finally, we compared our FCN-8s-VGG16 network architecture with different network architectures including a different skip architecture (FCN-16s based) and 2 different basic networks (CaffeNet based and AlexNet based). Besides, comparisons with other methods were also conducted, including the state-of-the-art simultaneous detection and segmentation (SDS), Graph-based and Level-set algorithm. The results on the test set showed that the algorithm achieved a complete segmentation of lactating sow by minimizing the effects of uneven color, light variations, occlusions, adhesion between sow and piglets and complex floor status, with an average accuracy of segmentation of 99.3% and a mean regional coincidence degree of 95.2% at an average speed of 0.22 second per image. However, it is hard to completely segment the sow’s head when sow’s head is downwards to floor, or close to the wall or adheres to piglets. The comparison with different network architectures showed that the mean regional coincidence degree of our proposed network architecture was higher than that of the others, and on GPU, the segmentation speeds of our FCN-8s-VGG16, FCN-16s based, CaffeNet based and AlexNet based were 0.22, 0.21, 0.09, and 0.09 second per image, respectively, which had good real-time performance. The comparison with other methods showed that our FCN-8s-VGG16 model outperformed others, which improved the mean regional coincidence degree of SDS, Graph-based and Level-set by 9.99, 31.96 and 26.44 percentage point, respectively. All of the experimental results suggest that the proposed method demonstrates a higher generalization and robustness, and provides an effective access to accurate and fast segmentation of lactating sow image under a pigsty circumstance. [ABSTRACT FROM AUTHOR]

Observing animal's individual and social behaviors is the most effective way to assess animal welfare and healthy. Automated trajectory tracking based on head/tail locations is supposed to be extremely helpful for the realization of pig behavior recognition, especially for group housed pigs in the commercial pig facility. The methods of trajectory tracking for group housed pigs based on head/tail location were described in this paper. The video of group housed nurseries was taken in a commercial pig breeding farm of Hubei Jinlin Original Breeding Swine Co. Ltd. on January 12th, 2016. A high resolution camera (Woshida CL03) was used to record 15 min video. Afterwards, image frames were extracted from the original video in a one-second time interval. Image frames were processed in a computer (configured with IntelCore i7-4790 CPU (central processing unit), 3.6 GHz, 8 G memory) with MATLAB software platform. The image processing for each image frame included 4 steps: background removal, pigs division, head/tail identification and trajectory tracking modification. The background removal was based on the RGB (red, green, blue) color space, from which a vector of RGB mean values of the pig's body was calculated. If the Euclidean distance between the RGB values of one pixel and the RGB mean values vector was less than a small threshold of 100, the pixel was involved in a pig body area and set as 1. Otherwise, it was outside any pig body area and set as 0. When all pixels of the image frame were scanned and calculated by this method, a binary image was acquired. The white area referred to pig's body area, while the black area referred to the background. After that, the morphology erosion and expansion were utilized before the watershed segmentation algorithm to improve the dividing effect for the pigs with adhesion. Pigs division was implemented on the binary images with the improved watershed segmentation algorithm. To discriminate each pig in each image frame, a video tracking and marking method was necessary to be implemented in the video. After being manually marked with the identify number in the first frame, each pig had a unique number and was labelled automatically throughout the video. Abstracting image frames from the video with a very short time interval (1 s), the distance of 2 centroids of the identical pig between 2 continuous image frames would be sufficiently small. Therefore, the video tracking was to find the pig with the closest distance in the next image frame and mark it with the same identify number of the current pig until all the pigs were marked. After each pig was marked throughout the video, using the head/tail location as the coordinates of the pig, the trajectory of each pig in herd could be tracked by the trajectory calculation. Extracting the outline of each pig in frames, the head and the tail outlines were divided from the whole outline, after a sixth of whole outline distance was moved along the outline in 2 opposite directions from the 2 intersection points of the outline and short axis of the minimum bounding rectangle. After the head/tail outline curve was gained from each pig outline, 2 recognition algorithms, the analogous Hough clustering recognition algorithm and the roundness recognition algorithm, were employed to identify the head/tail of each pig. Thus the location of the pig's head/tail could be spotted by locating the centroid of the heat/tail curve. Then the trajectory tracking of the pigs was calculated based on the location of head/tail, and corrected by the motion trends of pigs. Experiment showed that the background was successfully removed from each image frame using the Euclidean distance of RGB values between the pixels and the mean value vector. The improved watershed segmentation algorithm has been verified as an effective tool to divide the pigs with adhesion. The identify number of each pig was tracked from the first frame to the end. The average recognition rate of analogous Hough clustering algorithm was 71.79% for the identification of pig's heads/tails, while the roundness algorithm was 79.67%, which was less sensitive to the distortion of head outline curve. If not including the pigs outside the camera range, the recognition rates would be up to 75% and 85.7% respectively. The roundness algorithm shows an obvious advantage in comparison. The modified trajectory of each pig shows a high agreement with the manually labelled trajectory. More understanding for pigs' behaviors can be acquired from the trajectory of head/tail locations. This trajectory tracking method provides a good reference for further research of behavior recognition. [ABSTRACT FROM AUTHOR]

The processing technology of the machine picking seed cotton (MPSC) should be influenced significantly by the kinds and the contents of impurities. But during the MPSC processes in China, there is a lot of the mixed level and the mixed ginned, as well as the lack of online detection technology. In view of the present situations, a method for the identification of impurities in MPSC image using genetic algorithm (GA) to optimize the SVM parameters has been presented in this paper. First, in order to label three categories of the cotton fiber, the light-colored impurities and the dark impurities, the feature vectors were constructed for the image segmentation by the color characteristics of hue, saturation and intensity of pixel neighborhood, and the texture features of the average brightness, the average contrast, the smoothness, the third moment, the consistency and the entropy. GA of optimal retention strategy was used to optimize the penalty parameter and kernel function parameter to establish the SVM classifier of image segmentation, and then morphological operation such as hole filling, opening operation, closing operation was adopted. The specific operation was the use of open operation on light-colored impurities to eliminate the influence of small area noise; while the use of closed operation on dark impurities to smooth the target boundary on the basis of retaining small areas of dust. Then, for the impurity recognition process, shape features including area, perimeter, eccentricity, rectangle degree and shape factor were added to feature vectors besides color feature and texture feature of marked region; five categories of cottonseed, cotton leaf, bell shell, stiff valve and dust miscellaneous were marked; and the SVM classifier for impurity recognition by GA was established. At last, automatic segmentation and impurity recognition for MPSC image were realized using these two SVM classifiers. For the experiment MPSC was taken as the test material. The color planar array CCD camera of technical grade (MV-EM510C/M, Microvision, Inc.) with the industrial lens (M0824-MPW2) was used for the shooting system which included the LED diffuse light source of a 4-segment strip (AFT-WL21244-22W) and the light source controller (AFT-ALP2430-02). When shooting, the camera and the light source were placed in the darkroom, and the MPSC used for the test was pressed against the transparent glass plate on the other side of the darkroom. The data training and the picture test were conducted using MATLAB R2014b and libsvm-3.21 toolbox, and 60 pictures were equally used for both training and testing. The SVM multi-classifier was established during segmentation and recognition operation instead of simple binary classifier. This method utilized SVM small sample and high dimensionality learning ability, and the segmentation and recognition accuracy were further increased. Experimental results show that, comparing with the segmentation results from fuzzy C - means clustering algorithm and traditional SVM algorithm, the classification accuracy was improved significantly by using neighborhood spatial information. When the contrast of the edge of the target area in the image was weak, the suggested method could effectively avoid the situation of taking the shadow of the target as the edge which had shown great adaptability in applications. In this study, compared with the traditional SVM recognition algorithm, GA was used to automatically optimize SVM penalty parameter and kernel function parameter. SVM classifiers for image segmentation and impurity recognition were obtained through the suggested method overcame the blindness of parameter selection and the shortcomings of BP neural network generalization ability. Since the data distribution shapes of the homogeneous image in the color space are usually of similarity, and the characteristics of MPSC impurities texture are of stability, the optimal SVM parameters are common to the same kind of ima [ABSTRACT FROM AUTHOR]

Plant phenotype is a key link in understanding gene function and environmental effects, and with the demand of further research of plant function genomics, the traditional phenotypic observation has become the main bottleneck. High-throughput plant phenome analysis technology such as image processing is an effective way to solve this problem. In this paper, a maize plant drought stress phenotype testing method was proposed based on the time-series images. Firstly, an acquisition system of time-series images was developed. System hardware included the mvc3000 high speed industrial camera (the maximum resolution 1600×1200 pixels, 24 FPS (frames per second)), the Pentax len (8.5 mm, f/1.5), the storage card (32G), the holder and the background. Software development environment included Win7 operating system, Microsoft Visual Studio 2010 Professional, and OpenCV2.4.3. Maize plant time-series images were acquired from 08:00 to 17:00 every 10 mins, and various natural light conditions strongly affected the profile of the crop images taken outdoors. This often made it rather difficult for vegetation parts segmented from the background parts in an image. We adopted the K-means and SVM (support vector machine) algorithm to create a classifier used in rough segmentation, and the classifier was composed of 24 dimension feature vectors including color features, texture features, and gradient features. The result refinement was implemented with a non-local mean (NLM) filter, and actually the anisotropic structural-aware filter was adopted to implement the NLM function, in which the pixels were reinforced with each other in the same structure and even if they were not the first-order neighboring pixels. Secondly, the plant organs such as stem and leaf were segmented in binary image with a scan line method, which used the plant morphology characteristics such as alternate phyllotaxis, curved blade, and leaf margin in folds as prior knowledge. The method denoted the pixels according to the number of neighborhoods. Specifically, each row of pixels in the scan line was marked as a candidate region, and the candidate regions were labeled as the original organ or new organ according to the number of neighborhoods. When all of the candidate regions were labeled, a cluster of regions which presented rectangular shape were defined as plant stem. A labeled cluster of leaf would be merged to the other cluster if they shared neighborhood regions up to 70%. Finally, the ratio SOP1/SOP2 was used to characterize the drought stress response. SOP is the intersection point of stem and leaf, SOP is a point in stem line, LEN is the length of leaf tip to stem line, SOP1 is a point in leaf curve with length 1/4LEN, SOP2 is a point in leaf curve with length 2/3 LEN. SOP1/SOP2 represented the degree of leaf bending. It was comparable even between the different leaf orders, and had strong robustness to environmental factor disturbance. The field tests showed that the SOP1/SOP2 of drought-resistant genetically-modified maize presented an ascending trend before midday and a descending trend after then, which could be caused by transpiration leading to leaf water loss. The change trend of environmental factors such as air relative humidity against SOP1/SOP2 showed that under drought stress condition, genetically-modified maize was more sensitive to moisture content of the atmosphere, and the test on conventional maize plant showed that the SOP1/SOP2 lacked the regularity with the time. [ABSTRACT FROM AUTHOR]

The Kinect sensor scanning images for agricultural plants are vulnerable to field light conditions and background noise, etc. In addition, the amount of data of the color image and depth image affect the efficiency and accuracy of the plant area, which leads to the difficulty of meeting the requirement for the Kinect sensor in agricultural plant detection. For the above problems, considering the influence of the light conditions and complex background information in agricultural environment on the quality of the plant detection and the depth data acquisition, in this paper, we proposed a plant depth detection image restoration method based on K-means and K-nearest neighbor. We also developed a novel method of image restoration to reduce the impact of background information to improve the accuracy of the color image segmentation, and to enhance the accuracy of depth data. Firstly, a RGB threshold segmentation algorithm was applied to original RGB-formatted plant color images to extract plant target areas from backgrounds. Three components R, G, and B were respectively separated from RGB color space, and the difference between G and R or B was primary extract of the plant area information. Meanwhile, for the color characteristic of the environment, a K-means clustering segmentation algorithm was performed on the extracted plant target areas to remove background noise and enhance target contours. Secondly, to fix the errors of the depth data and meet the requirements of the agricultural plant detection operations, the color image and depth image were registered to restore the suspicious pixels depth data based on K-nearest neighbor algorithm. Then, a K-Nearest Neighbor algorithm was presented to recovery the black hole pixels for depth images. Finally, we acquired the depth data of target plant from the detected images. Compared with conventional RGB threshold segmentation method and K-means algorithm method, the proposed method can be used to solve the problem of the color image noise. The experiment results showed that, the segmentation error can be reduced by 12.12% with RGB threshold segmentation method, and 41.48% with K-means algorithm method. The average segmentation error can be up to 12.33% by using RGB threshold segmentation first and then the K-means algorithm. Furthermore, the proposed method can be used to restore the depth data, and can significantly reduce the effect of the backgrounds. Thus it had a good improvement to the edge sharpness of the depth data, and the accuracy of the empty point depth data of single frame. The result of this study can be a reference for agricultural plant detection and 3D reconstruction, precision of target spraying. [ABSTRACT FROM AUTHOR]

Restricted by imaging equipment and external natural environment, apple image produces lots of noise in the process of collection and transmission, which is one of the important factors that affect the accuracy and efficiency of image recognition. In order to reduce the effect of the noise on the target identification of apple harvesting robot, the segmentation method for apple image with noise is studied, which is not affected by noise. Firstly, by constructing similarity matrix, gray value, local spatial information and non-local spatial information of each pixel are utilized to construct a three-dimensional feature dataset. And then, the space compactness function is introduced to compute the similarity between each feature point and its nearest neighbors. Obviously, the similarity matrix is sparse matrix. Secondly, the outliers of similarity matrix are tuned by splitting the outlier matrix and representing it linearly with the other remaining column vector. Finally, tuned similarity matrix is decomposed by Laplacian vector, and eigenvector matrix is constructed and then normalized; the next step is that row vector of the matrix is clustered by k-means algorithm. The clustering result is obtained for three-dimensional feature dataset, and the image segmentation result is also obtained. The experiments of 2 apple images are carried out to validate the optimization algorithm proposed in the paper. The segmentation accuracy of the optimization method for a single apple under the influence of different noise is over 99%. The segmentation accuracy is over 98% for overlapping apple. The segmentation accuracy rate is 99.014% on average for 30 apple images, which is under the influence of Gaussian noise with the variance of 0.05 and salt and pepper noise with the probability of 0.01. The results of optimization method are compared with the results of the original spectral clustering algorithm and the spectral clustering algorithm based on space feature. The advantage of the optimization method is achieving de-noising effect. Also, the role of tuning the similar matrix’s outliers is to achieve clustering optimization. In the setting conditions of this experiment, the segmentation accurate rate can be improved by 5%-6% compared to the spectral clustering algorithm based on space feature, and by 9%-25% compared to the original spectral clustering algorithm. At the end, the running time is analyzed and compared for the algorithms, and the experiments of 30 images, which contain 3 types of images i.e. 128×128, 256×256 and 512×512 pixels and each type has 10 images, are carried out to validate the algorithm’s efficiency. From the result of experiments, we know the optimization algorithm’s running time is less than the original spectral clustering algorithm and is close to the spectral clustering algorithm based on space feature on the premise of achieving better segmentation accurate rate. Through the analysis and comparison, the conclusions obtained from the study are as follows: first, the optimization algorithm has the robustness for the noise; second, the optimization algorithm reduces the wrong rate of the boundary region’s pixels; third, the optimization algorithm improves the segmentation accuracy and efficiency. The results provide a reference for fast target recognition of apple harvesting robot. [ABSTRACT FROM AUTHOR]

Field crop image segmentation has drawn considerable attention in many aspects of agriculture, such as identification of physiological stage, disease, insect, and vegetation cover estimation. This research was conducted in order to achieve environmentally adaptive segmentation of field rapeseed plant and background. A digital camera which was mounted on a tripod that was around 1.5 m high, Canon EOS Digital Rebel XS, was utilized to take pictures of the field rapeseed plants. For the sake of continuous monitoring of rapeseed plant, the camera acquired the images 2 times per day. The fact that color distribution of a single-colored object in the hue-saturation (HS) plane is not invariant with brightness changes has been testified in several researches. Statistical results also showed that at a specified intensity, the histogram shape of hue was similar to the Gaussian distribution. Accordingly, the single Gaussian model was used to characterize the distribution of hue at certain intensity. Fifteen images under different illumination conditions, which changed from sunny days, cloudy days, to rainy days, were selected to establish the HI_LUT (hue intensity looking-up table). First, all the background was removed, and only the green pixels which represented rapeseed plants were kept. The green pixels in RGB (red, green, blue) color space were transformed into HSI (hue, saturation, intensity) color space. The expectation and variance values of hues were computed at certain intensity. As for one given pixel, if the distance between the hue value of that pixel and the expected hue was smaller than a certain threshold, the pixel was segmented as green crop. However, how to select an appropriate threshold value was a key problem that shall be solved, for different threshold value may give rise to different results. This paper selected 45 field rapeseed plant images under different illuminations (sunny days, cloudy days, and rainy days) and different physiological stages (seedling, three-leaf stage, and four-leaf stage) as the samples to discuss the relationship between the selection of the threshold value and the segmentation result. Different threshold values were tested in order to find an appropriate threshold. Results showed that the best segmentation results were achieved and the integrity of the shape characteristic of rapeseed plant target was kept when the threshold value ranged from 2.4 to 2.6. However, if the threshold value equaled to 1.0, some green pixels were segmented as background. While the threshold was set as 4.0, non-green pixels were misclassified as rapeseed plants. In order to demonstrate the performance of the Gaussian HI algorithm, 4 established algorithms, namely, CIVE (color index of vegetation extraction), EXG-EXR (excess green - excess red), EXG (excess green) and VEG (vegetation) were implemented to make comparison with the Gaussian HI algorithm. Meanwhile, the ME (misclassification error) and RAE (relative objective area error) values were both calculated. Several conclusions could be drawn from the experimental results. 1) Good segmentation results could be achieved with the Gaussian HI algorithm with a few image samples. 2) The ME value of Gaussian HI algorithm reached 1.8%, while that of the other 4 algorithms were 2.7%, 3.8%, 3.1% and 4.2%, respectively. The RAE value was less than 3.6%, and that of the other 4 algorithms were 12.8%, 34.0%, 8.5% and 25.8%, respectively. 3) The standard deviation of the ME was 0.7%, and that of the RAE was 4.5%, which demonstrated that the algorithm showed better stability. The above test results verify the algorithm can segment the field rapeseed plant image effectively and guarantee the completeness of the crop shape, which can provide reliable database for the physiological stage identification of rapeseed plant. [ABSTRACT FROM AUTHOR]