Research On Tomato Leaf Disease Identification Based On Deep Residual Network

As a traditional crop with a long history in my country,tomato has not only become an important part of people’s daily diet,but also plays an irreplaceable role in my country’s agricultural strategic resources.However,with the increase of tomato leaf diseases year by year,it has seriously restricted the tomato industry of vigorous development.Manual identification of tomato leaf diseases requires a lot of experience and time,and requires high identification personnel.With the rise of knowledge technologies such as image processing,machine learning,and deep learning,automatic detection of crop leaf diseases has become a reality.Based on deep learning theory,this paper uses convolutional neural network to classify and identify 10 kinds of tomato leaf images in Plant Village,including 9 kinds of leaf diseases and1 kind of healthy leaves.The specific research contents are as follows:In order to realize automatic identification of tomato leaf diseases,different convolutional neural networks were used to train on the tomato leaf disease image dataset.The experimental results show that the use of convolutional neural network model to process image data is simple,and can obtain higher accuracy of leaf disease identification.In order to solve the problem that the existing tomato leaf disease identification methods are few and the identification accuracy needs to be improved,a network model InceptionLeaky Re LU-CBAM-ResNeXt50(IL-CBAM-ResNeXt50).By improving the residual structure of the traditional residual network Res Net50,it can adapt to the size of different tomato leaf disease spots and extract more disease details.At the same time,the Leaky Re LU activation function is used to solve the neuron death phenomenon of the Re LU activation function in the negative semi-axis.The attention mechanism CBAM module is added,so that the network can pay more attention to the diseased part of the tomato leaf image,enhance the valuable feature channels,and suppress the useless feature channels,thereby improving the model’s recognition accuracy of tomato leaf diseases.In order to solve the problem that the size of the convolutional neural network model is large and it is inconvenient to deploy on mobile terminals for real-time identification of tomato leaf diseases,a lightweight residual network-based network model for tomato leaf disease identification,Leaky Re LU-CBAM-ghost-Res Net34(L-CBAM-ghost-Res Net34).The traditional residual network Res Net34 is combined with the ghost module,and the convolution operation and linear operation in the ghost module are used to compress the volume of the network model.Then,by replacing the Re LU activation function and adding the attention mechanism CBAM module,it increases the extreme value.The accuracy of the model is improved with a small amount of parameters,computation,and model volume.From the comprehensive experimental results,the IL-CBAM-ResNeXt50 model proposed in this paper has an accuracy rate of 99.4% in tomato leaf disease identification.Compared with the traditional residual network Res Net50,the accuracy rate is increased by 0.9%,and it has better tomato leaf disease identification performance.At the same time,the proposed LCBAM-ghost-Res Net34 model has an accuracy rate of 98.8% in tomato leaf disease identification,and the model size is only 2.84 MB.Compared with the traditional residual network Res Net34,the accuracy rate is increased by 0.5%,and the amount of model parameters is reduced by 96%,the size of the model is reduced by 96%,and the number of floating-point operations is reduced by 86%,which is conducive to the deployment of the model on mobile terminals.