Research On Crop Disease Image Classification Algorithm Based On Convolution Neural Network

When deep learning algorithm is widely used in the field of image,scientific and technological workers begin to use convolutional neural network to classify crop diseases.However,many network models can not perform well in crop disease data sets because the differences between crop disease images are not obvious;In addition,the deeper the network layers are constructed,the more network parameters are,and the more difficult it is to train.Therefore,this paper,from the above two problems respectively,On the one hand,the convolution neural network model is improved to improve the accuracy of crop disease image classification;On the other hand,the structure of the lightweight model is studied,and a lightweight network model suitable for crop disease image classification is designed.The specific work is as follows:(1)To address the problems of small differences in crop disease images,low recognition accuracy and complex model training of traditional machine learning methods on crop disease datasets,this paper proposes a crop disease recognition algorithm based on common feature learning and data augmentation.Firstly,for the problem of unbalanced dataset,the dataset is expanded with Mixup data augmentation algorithm to enrich the number of samples;then,for the feature extraction module,the channel attention module is embedded in the deep residual network to focus on learning the crop leaf disease features and ignore the interference of background information on the model;after extracting the image features,the feature map is fed into the common feature learning module to improve the linear correlation between images and enhance the generalization performance and robustness of the model.The experimental results show that the model based on common feature learning and mixup data enhancement can extract the semantic and detailed information of categories,which can effectively improve the accuracy of crop disease image recognition.(2)To address the problems of redundant weight parameters of crop disease classification algorithm,complex model,difficult training and large amount of calculation,a lightweight crop disease classification network model based on asymmetric ghost transform is proposed in this paper.Firstly,in order to avoid the problem of too many parameters caused by convolution operation,the first layer of the network model uses the extended convolution method to extract the features of receptive fields of different sizes of images and fuse them;Then,the proposed asymmetric ghost transform module is used to enrich the square convolution kernel skeleton in order to enhance its robustness.The experimental results show that the lightweight crop disease classification network based on asymmetric ghost transform can not only accurately classify the crop disease image data,but also effectively reduce the amount of parameters of the network model.The use of convolution neural network to classify crop diseases not only improves the efficiency of disease classification,but also can detect and prevent diseases in the early stage of crop growth,avoid food and economic losses,and lay a foundation for the intelligent development of agriculture.