Research On Image Segmentation And Recognition Algorithm Of Crop Diseases Based On Convolutional Neural Network

Crop diseases have a great impact on the yield and quality of crops.Traditional artificial methods are not effective in recognition.Convolutional neural networks can be used to identify quickly and efficiently to meet the needs of precision agriculture.At present,most recognition methods are to directly input the collected disease images into the convolutional neural network for recognition,but the images collected under actual natural conditions are affected by complex background,noise and other factors,and the recognition accuracy is low,and the robustness is poor.For this reason,this paper proposes a new identification method.The first step is to segment the disease image collected under actual natural conditions to obtain the disease image with the background removed;the second step is to use the convolutional neural network to recognize the segmented image.The main research contents of this paper are as follows:(1)The first step is to segment the crop disease image.In crop disease image segmentation,traditional convolutional neural networks have the problem of low segmentation accuracy.For this reason,this thesis adds the conditional random field to the Seg Net network to establish an improved Seg Net network.This model combines the information of an image in a certain area with the context of the image content,so that the image details are displayed more comprehensively and the image processing is more accurate.Compared with traditional methods and deep learning methods on the test set,the results show that the improved Seg Net network performs best in terms of both the recognition effect and the average processing time of a single image.At the same time,the test results on the test set subsets under three different complex background conditions show that the improved Seg Net network is the most robust among the above-mentioned traditional and deep learning segmentation methods.In summary,the method in this thesis can quickly and accurately segment crop disease images collected under actual natural conditions,and meet the requirements of real-time and accuracy.(2)The second step is disease image recognition.In this thesis,after the first segmentation,the segmented image is recognized.The segmented image is still affected by factors such as illumination and noise.To this end,this thesis improves the VGG network by adding high-order residuals and parameter sharing feedback sub-networks,and establishes an improved VGG network to recognize crop diseases.The apparent characteristic expression of crop diseases is provided by the high-order residual subnetwork,which makes the disease recognition more accurate.The background noise in the deep feature of the diseased image is weakened by the parameter sharing feedback subnetwork,which makes the improved VGG network have stronger robustness.In the ANS-CD2688 dataset,the improved VGG network is compared with SVM,Alex Net,Res Net50,and VGG16 to verify its higher recognition accuracy.The results show that the improved VGG has the highest recognition accuracy.In the Plant Village subset and the ANS-CD2688 dataset,the improved VGG network in this thesis is experimentally analyzed from the aspect of robustness.The results show that the improved VGG network has the strongest robustness.In summary,the method in this thesis can improve the recognition accuracy and robustness of the segmented disease image.(3)A set of crop disease identification software is designed.The software contains two modules of segmentation and identification.Software users can choose the modules according to their needs.In summary,this article has performed highly accurate and robust recognition of crop disease images collected under actual natural conditions,which is of great significance in the prevention and control of crop diseases.