面向电网人员智能调度的实景巡检算法

Real scene inspection algorithm for personnel intelligent scheduling in power grid

为提高实景巡检中绝缘子缺陷的检测精度,该文提出一种基于双路注意力通道的YOLOv5(Double squeeze and excitation-you only look once version 5,DSE-YOLOv5)算法。该算法使用K-means聚类算法优化先验框参数,提高先验框与目标的匹配度;并对YOLOv5的骨干网和特征融合网络进行改进,加强对小目标检测能力;最后使用距离交并比损失的非极大值抑制(Non-max suppression using distance intersection over union,DIoU-NMS)进行检测。试验结果表明,与传统YOLOv5相比,所提DSE-YOLOv5在对绝缘子缺陷的检测上精准率、召回率、平均精度均值都有提升,满足了电网维修人员在智能调度时对识别精准度的需求。

To improve the detection accuracy of insulator defects in real scence inspection,a double squeeze and excitation-you only look once version 5(DSE-YOLOv5)algorithm is proposed.In the proposed algorithm,the K-means clustering algorithm is first used to optimize the prior frame parameters to improve the matching degree between the prior frame and the target.Then,the backbone network and feature fusion network of YOLOv5 are improved to strengthen the ability of small target detection.Finally,the non-max suppression using distance intersection over union(DIoU-NMS)is used for detection.Experimental results show that,compared with traditional YOLOv5,the proposed DSE-YOLOv5 has improved the performance in terms of the accuracy,recall rate and mean average precision(mAP)of insulator defect detection,which meets the demand for identification accuracy for intelligent dispatching of power grid maintenance personnel.