Tomato Leaf Disease Recognition Based On Attention Mechanism And Bilinear Pooling

Tomato is an important cash crop in China,and its yield is easy to be limited by leaf blight,leaf mildew and late blight.In the early stage of tomato leaf disease,accurate identification of disease types can take timely targeted measures to reduce the loss of yield.The traditional identification technology is cumbersome,and has certain deficiencies in the control area and detection speed,which can’t meet the needs of modern agriculture.In view of the above contradictions,this paper improved the existing convolutional neural network based on the knowledge of deep learning and image processing,and used the improved model to conduct classification experiments on the tomato leaf disease data set.The final results showed that the network in this paper had a high recognition rate of the 16 diseases.In this paper,the In-SE-Bilinear Model is constructed,and the Goog Le Net network model is used as the basis.Due to the good performance of its multi-scale convolution kernel(Inception block),the feature information of different scales can be obtained.However,due to the large number of layers in its network,the number of layers needs to be reduced.At the same time,in the process of feature extraction,the attention mechanism module is cascelled behind each Inception module to improve the sensitivity of the model to the disease area.Finally,since the data set is a fine-grained image with small inter-specific differences and large intra-class differences,the connection of a single-scale convolution kernel and Bilinear Pool before the full connection layer can reduce the network dimension,and enable the network to reduce the interference of image background information and enhance the accuracy of the model.The accuracy of the improved model is 97.06%,which is significantly lower than that of the conventional convolutional neural network.In order to further measure the performance of the network,the precision,recall rate and average interaction ratio(MIo U)calculated by using the confusion matrix obtained from the test set are 92.52%,92.44% and 92.55%,respectively.The experimental results show that this method can effectively improve the classification and recognition ability of the model.At the same time,in order to study the transplanting ability of the network,the network was used again to classify the leaf diseases of corn,apple and potato.Finally,it was concluded that the network had good transplanting characteristics.