Research On Tomato Leaf Pest And Disease Identification Based On Deep Learning

Tomato is an important crop in China,and its output will be affected by dozens of common tomato diseases such as leaf mold,mosaic disease,and leaf blight.If the disease can be accurately diagnosed and identified in the early stages of tomato disease,and timely measures can be taken to treat the diseased plants,the loss of tomato yield can be reduced to a certain extent.The focus of this process is how to quickly and accurately diagnose tomato diseases and pests Diagnosis.The traditional tomato disease detection and diagnosis is based on professional technicians,using the knowledge and experience of technicians to diagnose tomato diseases and insect pests.However,the traditional detection methods have certain limitations in terms of the number of technicians,the speed of detection,the area of prevention and control,and the accuracy of detection,and cannot fully meet the needs of agricultural modernization and informationization.In order to overcome the above limitations and realize agricultural informatization,this paper builds a deep convolutional neural network based on the theory of digital image recognition in deep learning,and uses the network model to classify the tomato leaves.The research results show that the new diagnostic plan The ten common diseases and insect pests of tomato have high recognition accuracy.This paper first constructs a residual neural network model with 18 convolutional layers,2 residual blocks,1 maximum pooling layer,1 flattened fully connected layer,and 1 classification layer.The constructed model is obtained 86.9%test accuracy rate.The classification of pests and diseases based on tomato leaves is a fine-grained classification task.Therefore,the original residual neural network model was improved to construct a bilinear residual neural network model.The new network model obtained 93.125%accuracy on the test data set rate.Then use VGG16,Inception-V3 and other traditional neural network models to compare with the network model constructed in this paper.The test results show that the bilinear residual network model has higher recognition accuracy.Finally,the grape leaf,corn,potato and other crop leaf data sets were used to train and identify the bilinear residual network model.On these data sets,the model obtained a higher recognition accuracy than the tomato leaf data set,indicating that the model not only has Excellent fine-grained classification and recognition capabilities and reliable portability.