Research And Implementation Of Plant Disease Recognition Model Based On Deep Learning

Automatic identification of plant diseases is very important for agricultural development in China.Fast and efficient identification methods can greatly reduce the economic losses caused by plant diseases to agricultural practitioners.Aiming at the problem of plant disease image recognition,the use of deep learning network can learn in an end-to-end way,so as to obtain better recognition effect and higher recognition efficiency.Based on the deep learning method,this thesis introduces the dynamic domain adaptive method to solve the problem of small sample data in the task of plant disease image recognition,and introduces the fine-grained recognition method based on transformer to solve the problem of hard recognition between different diseases of similar plants.This thesis mainly does the following three aspects:Firstly,in reality,the data provided by plant disease recognition task are mostly unlabeled data or some labeled data,which brings difficulties to the learning of deep network.With the help of transfer learning method,the knowledge learned from the source domain data can be transferred to the learning of the target task,so as to alleviate the problem of insufficient labeled data in the target task.In order to reduce the difference of image data between source domain and target domain in the process of migration,a plant disease recognition method based on deep dynamic joint adaptive network is proposed in this thesis.When training the network,this method first uses the dynamic joint adaptation between domains in the multi-layer network structure to analyze the relative importance of edge distribution and conditional distribution,and completes the targeted data distribution adaptation,then uses the entropy minimization principle to make the learning target classifier pass through the low-density region of the target domain.Experiments show that this method improves the recognition accuracy of the target domain image in the task of plant disease recognition.Secondly,the classification task of different diseases of similar plants belongs to fine-grained recognition task,which has the characteristics of high intra-class difference and low inter-class difference,which is more challenging than the general recognition task.In order to better obtain the fine-grained representation features of images,a fine-grained recognition method of plant diseases based on bilinear transformer interactive network is proposed in this thesis.In this method,firstly,the transformer network structure is used to extract the features,and the features that can better represent the key information of the image are selected according to the network weight during the feature extraction.Secondly,the selected features are self interacted to enhance the representation ability of the features.Then,in order to better judge the differences between images,the bilinear structure is used to complete the comparison and interaction based on similar images,and finally the enhanced features are fused,the fused features are input into the classifier to complete the classification.Experiments show that this method improves the accuracy of fine-grained recognition of plant diseases and birds.Finally,aiming at the task of plant disease identification,this thesis designs and implements the plant disease identification prototype system by integrating the above two plant disease identification methods.The system can upload local images to the web and make accurate identification.At the same time,it also has the functions of similar diseases display and user feedback.