电力信息与通信技术2025，Vol.23Issue(7)：30-37,8.DOI:10.16543/j.2095-641x.electric.power.ict.2025.07.04
基于YOLO-Efficient的高空作业安全带检测
Aerial Work Safety Belt Detection Based on YOLO-Efficient
摘要
Abstract
To enhance the efficiency and accuracy of safety belt detection in aerial work scenarios,this study introduces an improved model,YOLO-Efficient,which is based on the YOLOv8 model.First,the model incorporates a lightweight detection head called GNHead,designed with the Group Normalization algorithm.It uses GNConv instead of the traditional 3x3 convolution to enhance localization and classification performance.By employing shared convolution techniques,the model significantly reduces the overall number of parameters.Additionally,a Scale layer is added for feature scaling,improving the model's adaptability to targets of varying scales.Second,in the Backbone part,the model replaces the traditional YOLOv8 network's Conv module and C2f module with the generalized dynamic convolution from KernelWarehouse,known as KWConv.This approach offers greater flexibility and functionality while using fewer parameters,achieved by skillfully sharing and mixing predefined parts between layers.This reduces computational demands and significantly increases detection speed,making the model well-suited for real-time surveillance applications.Finally,the model replaces the original CIoU loss function with the WIoUv2 loss function,focusing more on medium and high-quality anchor frames.This change reduces boundary regression loss and notably improves detection accuracy.The effectiveness of these optimizations in the YOLO-Efficient model is demonstrated through its application in power plant overhead work safety monitoring,confirming its fast detection capabilities and high accuracy.关键词
高空作业安全带/YOLOv8/GNHead/KWConv/WIoUv2Key words
Aerial Seat Belt/YOLOv8/GNHead/KWConv/WIoUv2分类
信息技术与安全科学引用本文复制引用
刘威,卢妍洁,高焜,李梓轩,琚贇,张之刚..基于YOLO-Efficient的高空作业安全带检测[J].电力信息与通信技术,2025,23(7):30-37,8.基金项目
国家自然科学基金项目(62373149). （62373149）
