| In recent years,the area of fruit tree plantations has been expanding,and fruit trees are vulnerable to different kinds of diseases in their growth process due to the influence of environment,climate,planting land and other factors.In the early growth stage of fruit trees,the disease symptoms are generally manifested in the leaves of fruit trees.If not treated,it will eventually affect the fruit,resulting in the decline of apple quality and yield.In order to prevent such problems,this paper takes apple leaves and disease spots in the natural environment as the research object,and carries out the research on apple leaf disease spot detection and classification system based on deep learning network and image processing technology,so as to realize the segmentation,disease degree classification and disease spot detection of early apple leaves:Collect the images of apple leaves in the natural environment,including healthy leaves,spotted defoliation leaves,scab leaves and rust leaves,and preprocess and enhance the collected images to ensure the universality of the collected images.1.Collect the images of apple leaves in the natural environment,including healthy leaves,spot defoliation leaves,scab leaves and rust leaves,and preprocess and enhance the collected images to ensure the universality of the collected images.2.Extract the disease spot area according to the diseased leaves,and classify the disease grade according to the proportion of relative area.In order to solve the problems of low accuracy of leaf lesion segmentation and poor training effect of network model in natural environment,an improved UNET network model is proposed.Firstly,the residual module resnet-34 is added to the front end of feature extraction to solve the problem of information loss caused by down sampling during feature extraction,and significantly improve the feature extraction effect of leaf disease spots;Secondly,by adding the edge extraction module and fusion module,on the one hand,it can help the model better extract the contour features and improve the segmentation accuracy of the model.On the other hand,it can fuse the feature information of the lesion segmentation module and edge extraction module to realize the output of semantic segmentation categories.3.In order to solve the problems of low accuracy and poor robustness of leaf spot detection in natural environment,an improved yolov3 apple leaf spot detection model is proposed.Firstly,in the feature extraction stage,the attention mechanism of feature pyramid is introduced to integrate the attention information into the extraction process,which significantly improves the detection effect of small lesions and easily neglected leaf diseases;Secondly,the structure of the introduced attention module is optimized to improve the detection accuracy of the model on the basis of considering the detection speed.4.Based on the above research results,the apple leaf disease classification and detection system is designed and implemented by using qtdesigner and python language.When the user inputs the diseased leaves and selects different networks,the system will carry out diseased spot segmentation,classification and detection,and output the operation results and disease prevention suggestions in the result output area. |
