基于EBP-YOLOv8的葡萄叶病害检测与识别方法研究OA北大核心CSTPCD

In order to improve the accuracy of grape leaf disease detection in real environments,suitable for real-time video monitoring,UAVs and other embedded AI application scenarios,the YOLOv8 target detection model was improved in terms of model structure,lightweight and so on,and constructed EBP-YOLOv8.Firstly,BiFPN structure is introduced into the neck network to strengthen the fusion between the feature layers of the model and improve the detection ability of small targets.Secondly,C2_P is used to replace the C2f structure in the neck network to realise the lightweight of the model without reducing the accuracy of the model.Then,the EMA attention mechanism is integrated into the feature extraction network to improve the attention of the region of interest and the model to identify complex background and similar disease spots;and finally,the CIOU loss function is replaced by the ECIOU loss function to improve the detection performance of the model and make the model converge better.EBP-YOLOv8 compared with YOLOv8n,Faster-RCNN,RetinaNet,YOLOv8n,YOLOv8s,YOLOv7,YOLOv7-Tiny,YOLOv4-Tiny,the mAP improved by 3.2%,13.87%,3.49%,3.2%,1.3%,5%,4.7%and 8.8%respectively,and the model size is only 5.3MB.The improved algorithm effectively improves the detection accuracy while ensuring the real-time performance of the algorithm,which can provide an effective reference for the development of real-time edge system for vine leaf disease detection.