Research On Disease Degree Recognition Algorithm Based On Crop Leaf Images

Agriculture is the foundation of the society.The problem of crop diseases restricts the development of crop planting.How to control the disease efficiently and accurately is particularly important.At present,chemical methods such as insecticides,or using radiation and physical methods to isolate the barrier are generally used to prevent and control crop diseases.In this paper,based on the method of computer image processing and machine vision,the algorithm of disease degree recognition of crop leaf image is studied based on Convolution NeuralNetwork.The main research contents are as follows:(1)Aiming at the data set of crop leaf disease image established in this paper,five Convolutional NeuralNetwork models,AlexNet,VGGNet,ResNet,Xception and Densenet,are selected to identify the disease degree of crop leaf image,and compared with the traditional image processing algorithm of HOG+ SVM.50269 images of crop leaves including 10 species,such as tomato and grape,were collected through field photographing and Internet collection.According disease or not,the disease type and the disease degree,they were classified as healthy,general and serious and labeled manually.Then,the data sets were preprocessed.Based on the traditional method of HOG + SVM and the selected five convolutional neural network models of AlexNet,VGGNet,ResNet,Xception and DenseNet,the algorithm of disease degree recognition of crop leaf image is realized.After the experimental results are compared,DenseNet model has the highest recognition accuracy of 78.67%.(2)A SMOTE + Adaboost algorithm is proposed to improve the recognition accuracy of the model for unbalanced samples.In order to improve the recognition accuracy of the model for the small number of samples,a new method based on SMOTE + Adaboost algorithm is proposed,which improves the recognition accuracy of Densenet model by nearly 8%.(3)A network structure of DenseNet-Xception fusion model is proposed to recognize the disease degree of crop leaf image.Firstly,the network model is optimized by migration learning.Secondly,based on the model fusion theory and Boosting algorithm,a network model structure of DenseNet and Xception fusion is proposed.The final recognition accuracy is 97.26%.(4)A software system for disease degree recognition of crop leaf image is designed and developed.Based on the research results of disease degree recognition of crop leaf image,according to the system demand analysis,a software system of disease degree recognition of crop leaf image is designed and developed by using Visual Studio and MATLAB software.