改进YOLOv8网络的小目标识别算法研究OA

To address the issues of high missed detection rate,frequent false positives,insufficient de-tection speed,and weak feature extraction capability encountered when using the traditional YOLOv8 net-work for small object detection in complex factory environments,this paper proposes an improved method combining multi-path features and attention mechanisms.Specifically,to address the weak feature repre-sentation of small objects and their susceptibility to background interference,a CA attention mechanism is introduced to enhance focus on key regions and suppress background noise.To improve multi-scale fea-ture fusion and retain more detailed information,a bidirectional weighted feature pyramid network(BiF-PN)is constructed.At the same time,to balance detection speed and model efficiency,a dual-branch input structure is used,where the image input is divided into a GhostNet path and a backbone network path for lightweight feature extraction.Additionally,feature representation and classification accuracy are further optimized through Convolutional(Conv)operations.Finally,the proposed improved model is applied to target recognition tasks in factory environments.The results show that the improved YOLOv8 network a-chieves increases of 6.7 percentage points,11.0 percentage points,6.2 percentage points,and 9.1 percent-age points in recall,precision,mAP50:95,and mAP50,respectively.