Research On Intelligent Identification Of Citrus Fine-grained Diseases Based On Convolutional Neural Network

As one of the important economic crops in my country,citrus is easily affected by the climate,geographical environment and pathogens of its planting place during its planting process,resulting in the emergence of various diseases and insect pests.Therefore,it is one of the effective methods for disease control to identify diseases through changes in citrus leaves.At the same time,due to the characteristics of citrus diseases with small differences between classes and large differences within classes,visual interpretation in the field is prone to misidentification.In addition,the orchard is located in a remote location,and experts cannot be present to provide technical support in time when a disease occurs,which easily delays the best treatment time for fruit trees.Currently with the rapid development of deep learning,its application in crop disease recognition has become a research hotspot.Compared with traditional machine learning,the recognition accuracy of deep learning is usually higher,and it can quickly and accurately identify diseases.Therefore this thesis uses the convolutional neural network to realize the citrus multi-type disease recognition and fine-grained disease detection.The research content includes:（1）Research on the recognition model of citrus various kinds of diseases.Due to the lack of public data sets of citrus diseases,this paper obtained a variety of citrus diseases by manual data collection,at the same time used data augmentation and transfer learning to solve the problem that small sample data is prone to over-fitting during training.Aiming at the problem that the feature difference between some citrus diseases is not easy to identify accurately,this thesis constructs a feature extraction network based on Res Net-50,and replaces the pooling layer with an improved pyramid pooling module to improve the feature extraction ability of the model.Then the improved model is compared with VGG-16 and Mobile Net.The results show that the proposed classification method has an accuracy rate of 99.29% on the test set,which can accurately recognize various citrus diseases.（2）Research on the recognition model of citrus fine-grained pests and diseases.Aiming at the problems of small and different shapes of citrus disease spots,which lead to feature extraction difficultly and low segmentation accuracy,this thesis proposes a dual-branch citrus fine-grained disease recognition network based on attention mechanism.By constructing a dual-branch network to improve the model’s ability to extract detailed information on pests and diseases,and introducing attention mechanisms to enhance the features that are more useful for recognizing citrus diseases,so as to achieve accurate segmentation of citrus disease spots.Aiming at the lack of objective criteria for judging the severity of citrus diseases,this paper developed a criteria for determining the severity of citrus diseases by calculating the ratio of the area of lesions to the leaf area or by calculating the number of lesions,so as to achieve relatively objective and accurate fine-grained disease recognition.Finally this thesis uses the improved model to compare with 6 existing mainstream semantic segmentation models,and the final m Iou of the improved model on the test set reaches84.35%,which is 3.66% higher than the original model and significantly higher than other models.（3）Development of a citrus fine-grained disease recognition system.Aiming at the needs of citrus growers and orchard managers for rapid recognition of citrus fine-grained pests and diseases,this thesis designs and develops a mini app for citrus fine-grained disease recognition based on We Chat developer tools and Pycharm.Users can quickly obtain the results of the disease types and severity of citrus by uploading pictures,which provides technical support for accurate disease diagnosis and rapid treatment.