Research And Application Of Transformer In Crop Fine-Grained Disease Identification

Agriculture is the source of human food and clothing,the foundation of survival,and the basic industry for the construction and development of China’s national economy.Crop diseases are one of the main reasons for the reduction of agricultural production.With the development of agricultural informatization,intelligent identification of crop diseases through artificial intelligence methods is a development trend.Especially in recent years,with the development of deep learning,the identification of coarse-grained crop diseases,that is,the judgment of the disease category of crops,has made great achievements,and in most cases it has exceeded human experience judgment.However,the fine-grained disease identification of crops,that is,not only judging the disease category of crops,but also judging the severity of the same disease of crops,still needs further research.The Transformer model is currently a research hotspot in the field of computer vision.This paper uses the deep learning method to study the application of Transformer in the field of crop fine-grained disease recognition.In this paper,based on the idea of combining convolutional neural network and Transformer,a fusion model of Res Net and Vi T(Vision Transformer)-Residual Vision Transformer(Res Vi T)is constructed.After experimental comparison,the recognition accuracy of this model is higher than that of the fusion models Res Net and Vi T,and also higher than some lightweight models,while maintaining a very low amount of parameters and a low amount of calculation.Then the self-attention function of the Res Vi T model was improved,and a coefficient α was added to the self-attention function.Through experiments on the tomato data set,it was determined that the recognition effect was the best when the coefficient α was 2.5.Next,the model pre-trained on the tomato data set will be transferred to the apple data set and grape data set,and a good recognition effect has been achieved,which proves that the proposed Res Vi T model has good generalization and self-efficacy.Effectiveness of Attention Function Improvement.Finally,in order to put the theoretical results of the research into practical use,this paper also developed a We Chat applet for crop fine-grained disease identification.For many crops,in addition to the core disease identification function,the applet can also provide the prevention and control measures of the disease.Users can also browse the knowledge of crop diseases.After logging in,they can view the identified historical records in the user center.