Research And Application Of Apple Leaf Disease Identification Based On Improved YOLOv5s

Apples are a fruit of great nutritional value and are very popular,and their market demand is increasing.However,the presence of apple leaf diseases ser Io Usly affects the yield and quality of apples,as well as causing significant production and economic losses to farmers.In the past,disease detection was mostly manual and required a lot of time,effort and cost,and errors in judgement could occur,which ser Io Usly affected the development of apples.To avoid misdiagnosis,fast and accurate disease identification is crucial for the long-term development of the apple industry.Therefore,this thesis builds a disease identification model based on deep learning to identify six common apple leaf diseases and achieve accurate judgement of disease types,thereby improving apple production and quality.The details of the research are as follows:(1)Research on apple leaf disease recognition based on classical target detection model.To address the problem of insufficient amount of sample data in the dataset,data expansion was performed on six types of apple leaf disease images using random cropping,greyscale,90-degree rotation,180-degree rotation and Gaussian blurring,and the Label Img annotation tool was used to annotate the diseases on the leaves.Then five classical target detection models,Faster R-CNN,YOLOv5 s,YOLOv5m,YOLOv5 l and YOLOv5 x,were selected to detect the diseases.Comparing the detection results,it was found that the YOLOv5 s model had better comprehensive performance in detecting the targets,so it was chosen as the improved basic model.(2)A study on apple leaf disease recognition based on the improved YOLOv5 s model.The use of Ghost module instead of ordinary convolution in the YOLOv5 s model and the improvement of the backbone network can reduce the computation of the model;the introduction of CA attention mechanism in the Back Bone part to improve the attention of the model for important regions and enhance the attention of the model;the introduction of BiFPN feature fusion module in the Neck part to improve the extraction and processing of features by the model capability;the EIoU loss function is adopted to improve the regression accuracy of the bounding box.Based on the above improvement points,a new model,namely the YOLOv5s-GCBE model,was proposed.The experimental results show that the accuracy,recall and m AP@0.5 of the model reach 94.5%,92.1% and 95.8%respectively,which perform better in target detection and can improve the accuracy and confidence of the target detection frame in the input image,and also have better performance in multi-target detection.(3)Implementation of an apple leaf disease detection platform.To address the lack of practical apple disease diagnostic tools in production scenarios,a detection platform for apple leaf diseases was developed in conjunction with the constructed YOLOv5s-GCBE detection model.The platform is able to detect six common apple leaf diseases and provide a series of disease control measures,which facilitates fruit farmers to make accurate judgments on leaf disease types and effectively reduces economic losses.