Research On The Model Of Soybean Disease Grading Based On Convolutional Neural Network

The issue of food production concerns the fundamental interests of the people and the stability of society.Soybean is one of the important food crops in my country,and the disease will lead to lower yields and economic losses.Because the characteristics of diseases are relatively similar,only relying on experience to deal with them may miss the best control period.At the same time,the degree of disease is different,and the prevention and control measures will be different.Therefore,taking measures in time according to the disease level of soybean plants is of great significance for effective disease control,avoiding drug waste,reducing environmental pollution,and reducing economic losses.At present,the research on disease identification of major food crops such as corn and rice has been very substantial at home and abroad.In comparison,there are few researches on the identification of crop disease levels,especially in the automatic identification of soybean disease degree,the accuracy rate needs to be improved.In view of this,this research takes the leaf image samples of three main soybean diseases(grey spot,mosaic disease and root rot)as the research object,and proposes a soybean disease grading model based on convolutional neural network.The specific research includes the following: aspects:(1)The preprocessing of soybean disease images by AISA image segmentation algorithm based on Grab Cut was discussed.The hyper-green factor(2G-BR)in the hyper-green algorithm is used to mark the pixels within a specific range of the image that are smaller than a given threshold as the background.The image obtained after segmentation by Grab Cut is matched with the original image,and the interior of the restored leaf is misjudged as the pixels of the background get a more accurate segmentation image.The algorithm replaces the manual labeling process in Grab Cut,effectively retains the characteristics of soybean leaves and diseased areas,reduces the influence of background factors,and improves the performance of convolutional neural networks.(2)The soybean disease recognition model based on convolutional neural network is studied.Train 7 traditional convolutional neural network models(VGG16,VGG19,Res Net50,InceptionV3,Xception,Mobile Net,Google Net),analyze and judge the recognition effects of different models,and select the three models with higher accuracy(Inception-V3,Xception,Mobile Net)to build a weighted deep voting model through genetic algorithm.The model achieved 99.31% recognition accuracy for gray leaf spot,96.67% accuracy for mosaic disease,and 97.33% accuracy for root rot,realizing the recognition of soybean leaf diseases.(3)The soybean disease grading model based on CNN-LSTM is studied.The images of the same diseased soybean leaf were continuously collected,and a time series data set was constructed according to the order of collection time.The spatial features of the disease were extracted through the convolutional neural network and then put into the LSTM network,and the time features were further extracted according to the context of the time series.The attention mechanism is introduced to optimize the process of spatial feature extraction,and three hierarchical models of different diseases are constructed.Experiments show that the accuracy of the grading model for gray spot disease reaches 94.9%,the accuracy of the grading model for mosaic disease reaches 96.7%,and the accuracy of the grading model for root rot reaches 93.9%,which realizes the automatic grading of soybean diseases.(4)Developed a soybean disease grading system.Deploy the above model into the We Chat applet for soybean disease grading to realize automatic grading on the mobile phone.At the same time,the applet proposes reasonable prevention measures according to the grading results,and provides reference for researchers and users.The development of this system can test the performance of the model trained in this paper,which is convenient for users to use,reduces the cost of use,and has certain practical value.In summary,this research provides a new idea for the application of convolutional neural network to soybean disease grading,and the CNN-LSTM model is applied to soybean disease grading research for the first time,providing a reliable technical support for soybean and other crop disease grading,which laid a theoretical and experimental foundation for the follow-up research on soybean disease early warning and disease trend prediction.