Tomato Disease Recognition Based On Lightweight Convolution Neural Network

Plant diseases and insect pests are one of the main factors threatening plant growth.Accurate identification and classification of plant diseases and insect pests is the basis of effective control of plant diseases and insect pests.The rapid classification of tomato diseases and pests by using the deep learning network model is the basis for assisting researchers and farmers to implement precise spraying quickly and targeted,improve the utilization rate of pesticides and establish smart agriculture.There are some similarities between different tomato diseases and insect pests.Traditional crop identification and classification methods can not reflect the pathological changes of tomato,and their practicability is poor.Conventional neural network is not effective in classifying these diseases and insect pests.To solve the above problems,this paper adopts the convolutional neural network model for experiments.This paper takes the tomato leaf diseases and insect pests data set of AI Challenger crop diseases and insect pests data set as the research object,and constructs the data set of tomato early blight,late blight,leaf mildew,powdery mildew,spot blight,red spider injury,Mosaic virus disease,yellow leaf curve virus disease and healthy leaves.A new convolutional neural network Mobile Net-AD was established to solve the problems such as long training time,high training cost and weak generalization ability of tomato leaf pests classification task with traditional neural network model.Firstly,the neural network models of Inception V3,Xception,Squeeze Net,Mobile Net V1 and Mobile Net V3-small are analyzed in terms of model selection.It was evaluated from the aspects of Accuracy,running time,F1-score,precision,Recall and Specificity.The lightweight convolutional neural network Mobile Net V3-small was selected as the basic network model for tomato leaf pest classification after considering the performance indexes of the neural network comprehensively.Secondly,in view of the insufficient number of samples and uneven distribution of samples in the AI Challenger data of tomato leaf pests and diseases,an improvement scheme was proposed.In this paper,data enhancement and transfer learning were applied to improve the classification effect of the model,and a lightweight convolutional neural network combining the above two methods was proposed.In addition,the effects of changes in different optimizers and loss functions on the model classification were discussed.In this paper,the Auxiliary loss function and the ASGD optimizer were combined to optimize the model,so as to improve the accuracy of model classification,accelerate the rate of convergence,and enhance the robustness of the model.Finally,in order to solve the problem that the accuracy of the existing classification task of tomato leaf diseases and insect pests neural network needs to be improved,a classification model of tomato leaf diseases and insect pests with mixed attention mechanism was proposed,and the data set was enhanced to increase the model generalization.Introducing Mixed Coordinate Attention(CA)and A Simple,Parameter-Free Attention Module(Sim Am)to improve the attention mechanism Chapter 4 Lightweight convolutional neural network Mobile Net-AD,Re-calibrate the features of channel and space,and discuss the effect of adding attention modules CA and Sim Am separately on model performance.The attention mechanism was visualized by heat map,and the experimental results showed that the accuracy of the proposed Mobile Net-CS in tomato pest classification task could reach 98.40%,which could improve the accuracy of tomato leaf pest classification to a certain extent.So it can meet the practical application demand of online diagnosis and control of tomato pest.