Crop Disease Identification With Bimodal Feature Integration And Hgnn

At present,crop diseases have become one of the major challenges in our country.With the continuous improvement of agricultural production capacity,the production environment is becoming more and more complex,and diseases are encountered more frequently in the process of production.Extremely serious crop diseases will not only cause huge economic property losses,but also exacerbate the demand for agricultural products.With the development of computer technology,deep learning has made great breakthroughs in image processing and recognition.At present,many experts and scholars use deep learning to automatically learn relevant features from data to help agricultural workers identify and diagnose crop diseases,discover symptoms of crop diseases in time and take preventive measures.Based on the above,this paper studies a multi-crop disease identification method based on deep learning,proposesd Bimodal FINet and MF-Bimodal FINet networks.The main contents are:(1)Crop disease recognition method based on bimodal feature fusion based on hypergra ph neural network.Aiming at the problems of existing crop disease recognition models,such as low recognition accuracy and single crop recognition caused by only using single modal features,a new crop disease recognition method Bimodal FINet based on dual modal feature fusion of hypergraph was proposed.Bimodal FINet is composed of text modal branch,image modal branch and hypergraph neural network.The recurrent neural network(Text RNN)and the improved convolutional neural network(Res Next50-CA)are used to construct the dual-modal feature fusion network of double-branch parallel structure of image text.The semantic feature information and feature representation rich in spatial location information are extracted to obtain two modal features.The feature fusion method is used to realize the complementary and fusion of each branch feature information.More abundant bimodal characteristic information of disease was obtained.Finally,hypergraph neural network is used to encode the dual-mode characteristic information after fusion to obtain the correlation and data representation between the data,thus improving the accuracy of the model.The experimental results show that the Bimodal FINet network has achieved 94.83% accuracy on the data set,and the classification effect is better than other models.(2)Crop disease recognition method based on dynamic hypergraph neural network.In view of the problems that the existing crop disease recognition model based on deep learning will lose important feature information in the process of convolution and the fusion effect between different modes is not good,the original model Bimodal FINet was improved and the feature extraction and fusion module(MF module)and dynamic hypergraph neural network were embedded.Firstly,the feature extraction and fusion module(MF module)was embedded on the basis of Res Next50,and four kinds of convolution check of different sizes were used to extract the feature of the disease,so as to capture more details and improve the incomplete feature extraction of the image branch network.Secondly,coordinate attention is embedded on the basis of Res Next,which can effectively solve the problem of loss of disease characteristic information.Finally,the feature information of the two branches is fused and sent to the dynamic hypergraph neural network.The network firstly realizes the dynamic update of the hypergraph structure of each layer through the dynamic hypergraph construction layer,extracts the local and global correlation,and then uses the dynamic graph convolution layer to collect the feature information between the vertex and the hyperedge to capture the deep relationship and information between the disease data.Improve the effect of the fusion of two modes,so as to get better classification effect.The experimental results showed that the recognition accuracy of MF-Bimodal FIDNet network was up to 96.95%,which was superior to other models and could provide a theoretical basis for crop disease recognition.