Study On Recognition Method Of Maize Leaf Common Diseases Based On Convolutional Neural Network

Corn is an important food crop in our country.Its planting area and wide,high output provide strong support for our fast economic development.Due to farming techniques of farmers,diversity of pathogenic bacteria and weather,maize leaf diseases become more and more serious,which seriously affects maize yield.Traditional recognition methods rely on artificial naked eye observation,but to accurately identify leaf disease types,expert groups and continuous crop monitoring are needed to support,and cannot meet the real-time requirements.With the rapid development of machine learning,the classification of maize leaf diseases can be realized by using machine learning technology.However,the process of treating maize leaf disease with this method is complicated,the training time is long,and the classification accuracy still has room to improve.Therefore,accurate and timely identification of diseases can reduce unnecessary economic losses and promote the increase of maize production.This paper innovates on the basis of convolutional neural network technology and studies the classification of common diseases of maize leaves.The specific research content is as follows:(1)The data set of common maize leaf diseases was established.A part of the original image data set was obtained from Plant Villag and a total of 2147 images were collected in Changyi District,Jilin City,Jilin Province in August 2022.After image collection,a total of3716 data sets of common diseases of maize leaves were reconstructed through further screening and data enhancement preprocessing methods such as noise enhancement and noise reduction.(2)Construct convolutional neural network to identify common diseases of maize leaves.The convolutional network models Res Net50,Alex Net,VGGNet16 and Goog Le Net were selected as the recognition models for common diseases of maize leaves in view of the cumbersome work and low recognition rate of manual detection methods,as well as the complex processing,long training time,large number of weight parameters and weak generalization ability of machine learning methods.The recognition rate reached 93.65%,89.72%,92.51% and 91.04% respectively,which improved the recognition rate compared with traditional machine learning methods.(3)In order to improve the feature extraction ability of convolutional network,the method of adding attention mechanism and optimizing model network structure was adopted to improve the recognition network of common maize leaf diseases.Firstly,the CBAM module is deeply integrated with the network to make the network pay more attention to the target area and improve the disease classification performance of the model.Then,the structure of the model Res Net50-CBAM is improved,and the 7×7 convolution nuclei are replaced by three 3×3 small convolution nuclei to obtain the C-Res Net50-CBAM model.The recognition rate of its model is97.84%.The results showed that the accuracy of the model after optimizing and integrating the attention mechanism was improved compared with the accuracy of direct training in the ablation experiment,the accuracy of only optimizing the model,and the accuracy of only adding the attention mechanism.The attention of the improved network model C-Res Net50-CBAM was able to focus more on the characteristics of maize leaf disease.Thus,the target characteristics of maize leaf diseases can be extracted and the common diseases can be accurately identified.This method can provide technical support for image processing,classification and recognition.