Research On Maize Leaf Pest Identification Technology Based On Deep Learning

Our country is a big agricultural country,and corn crops are one of the main food crops in my country.To ensure food security,the participation of emerging technologies is urgently needed.In this context,more and more researchers are conducting research on corn crop diseases and insect pests,and most of the research work adopts machine learning algorithm models,such as SVM.After deep learning technology has gradually become the focus of research,many researchers have tried to apply deep learning models to the problem of corn pests and diseases,from the initial relatively simple VGG16 model to the more complex Res Net model,which has greatly promoted the development of smart agriculture.However,the systems developed by these researchers are mostly limited to laboratories,which have disadvantages such as high promotion cost,difficulty for farmers to operate,and low recognition accuracy.For diseases and pests of corn crops,the paper proposes an optimized Mobile Net model for identifying corn leaf diseases and pests on mobile devices..The specific work is as follows:(1)Data collection and preprocessing.A total of 1406 images of maize healthy leaves,large leaf spots,small leaf spots,and maize rust leaves were collected as a dataset.At the same time,image preprocessing is performed on the dataset.(2)Optimization and training of Mobile Net network.The fully connected layer of the model is replaced with a 1×1 convolutional layer to obtain an optimized Mobile Net model(IMN).The IMN model is trained by the training strategy of transfer learning,and a model that can accurately identify four types of maize leaves is obtained.(3)Comparison test between the IMN model and the existing model.In order to explore the difference between the method proposed in this study and the traditional method,this study builds the SVM and VGG16 models as the comparison models.According to the comparison test results,it can be seen that the IMN model adopts the training strategy of transfer learning and the learning rate is set to 0.001.After epochs,it is close to convergence,and the accuracy rate reaches 98.25%.When the VGG16 model adopts the transfer learning training strategy and the learning rate is set to 0.001,after three epochs,the model accuracy rate is 75.25%,and the model does not converge.The SVM model uses RBF as the kernel function and adopts a multi-classification strategy.The comparison experiment between IMN and SVM also shows that the accuracy of the IMN model is higher than that of the SVM model.(4)Explore the effects of different training methods and learning rates on the model.In this study,two training strategies of transfer learning and zero learning are used for the same model.From the experimental results,it can be seen that the convergence speed of the model trained by the transfer learning strategy is much higher than that of the model trained by the new learning strategy,and the accuracy,The generalization ability is also better than the latter.By adopting the same learning method for the same model and setting different learning rates,it can be seen that the convergence rate of the model trained by the transfer learning strategy is much higher than that of the model trained by the new learning strategy.(5)Construction of an APP for corn leaf disease and insect pest identification.Convert the model to a format that can be transplanted to the mobile terminal on the PC side,develop a corn leaf disease and insect pest identification APP on the Android platform,and transplant the model to the mobile terminal.The APP has two functions of local prediction and real-time prediction.In this study,the classic Mobile Net model is improved to obtain an IMN model,and the IMN model is trained by means of transfer learning.Finally,the trained IMN model is deployed on the mobile terminal,and a good recognition effect is achieved.