Search

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]

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]

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]

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]