Image Recognition Of Plant Diseases Based On Fine-grained

China is a large agricultural country,and the health of crops is an important factor affecting the yield of the country’s economic crops.When growers are faced with crop diseases,the type of disease is commonly judged manually,which is prone to misjudgment thus causing economic losses.In order to improve crop yields,it is valuable to find a method that can reliably identify plant diseases.The observation of plant leaves can effectively distinguish plant species and disease types.Plant leaf disease images are fine-grained images,and traditional image classification methods require manual annotation,which has the problems of high labor intensity and low recognition accuracy.With the development of deep learning,convolutional neural network has better recognition accuracy for fine-grained images.In order to improve the recognition accuracy of finegrained images of plant diseases,this paper selects convolutional neural networks as the basis,and uses three methods: transfer learning,adding mixed attention mechanism,and stacking model to recognize fine-grained images of 58 types of diseases for 10 crops: apple,grape,strawberry,citrus,tomato,cherry,peach,corn,pepper,and potato.The research of this paper is as follows:(1)Study on the pretreatment of plant disease image.due to the inconsistent image size in the obtained plant disease fine-grained image data set,there are some differences in the light and shade,degree and shooting time of the image.the experimental data set is enhanced by using size normalization,random rotation,color jitter,gaussian noise,and the experimental data set is established for subsequent model training.(2)A fine-grained image recognition model for plant diseases based on DenNet transfer learning is proposed.The pre-training parameters DenseNet121 convolution neural network on the ImageNet are transferred to the plant disease image data set,the Pytorch is used as the deep learning framework,the established data set is used as the input of the DenseNet121 network,and the network training parameters are optimized by using dynamic learning rate.The average recognition accuracy of fine-grained images of plant diseases is 88.24% DenseNet121-P,which is based on migration learning;(3)In this paper,a fine-grained image recognition model for plant diseases based on mixed attention mechanism is proposed.Based on the deep residual network,channel attention module and spatial attention module are added to improve the feature acquisition ability of convolution neural network to fine-grained images of plant diseases.A plant disease recognition model with mixed attention mechanism was obtained by using plant disease data set to train the deep residual network with mixed attention.The recognition accuracy of the model for fine-grained images of plant diseases ResNet50-P,92.08.(4)This paper presents a fine-grained image recognition model based on stacking model.Using the Stacking stacking model method,the trained model DenseNet121-P、ResNet50-P、ResNet34、ResNet50、DenseNet121 is stacked,and the stacked model is trained with the fine-grained image data set of plant disease.The fine-grained image classification recognition model based on stacking model is obtained S-Plant,and the average recognition accuracy on the fine-grained image data set of plant disease is 92.97.Compared with the traditional neural network,the effectiveness of this method for fine-grained image recognition of plant diseases is proved.