融合注意力机制和Bi-YOLO的变电站异物检测研究  

Research on foreign object detection in substations by integrating attention mechanism and Bi⁃YOLO

作　　者：马显龙 曹占国 段雨廷 于虹 周帅 MA Xianlong;CAO Zhanguo;DUAN Yuting;YU Hong;ZHOU Shuai(Electric Power Research Institute of Yunnan Power Co.,Ltd.,Kunming 650217,China)

机构地区：[1]云南电网有限责任公司电力科学研究院,云南昆明650217

出　　处：《电子设计工程》2025年第1期186-189,195,共5页Electronic Design Engineering

基　　金：2021年云南电网有限责任公司科技项目(YNKJXM20210022)。

摘　　要：由于变电站内设备悬挂异物的现象频发,为预防站内人员巡检不及时导致电力事故发生,在YOLOv5的基础上,提出融合注意力机制和Bi-YOLO的变电站异物检测研究。采用加权双向特征金字塔替代原有的特征金字塔网络,将坐标注意力模块嵌入C3结构,添加混合注意力模块,以提高变电站异物的特征提取能力和检测效率。实验结果显示,相较于YOLOv5算法,提出的检测算法多类别平均检测精度提高3.3%,mAP值达91.3%,满足实时性要求。Due to the frequent occurrence of foreign objects hanging on equipment in substations,in order to prevent power accidents caused by untimely inspection by station personnel,a research on substation foreign object detection integrating attention mechanism and Bi⁃YOLO is proposed based on YOLOv5.The weighted bidirectional feature pyramid is used to replace the original feature pyramid network,the coordinate attention module is embedded in C3 structure,and the mixed attention module is added to improve the feature extraction ability and detection efficiency of foreign bodies in substation.The test results show that compared with the YOLOv5 algorithm,the multi⁃class average detection accuracy of the proposed detection algorithm is increased by 3.3%,and the mAP value is up to 91.3%,meeting the real⁃time requirements.

关 键 词：YOLOv5 注意力机制 CBAM 目标检测 Bi-FPN 

分 类 号：TN49[电子电信—微电子学与固体电子学]