基于GFPN-YOLOv5的配网架空线路线夹护套检测方法

Detection Method of Clamp Sheath of Overhead Line in Distribution Network Based on GFPN-YOLOv5

配网架空线路未安装线夹护套是典型的施工缺陷。针对配网架空线路线夹护套存在多尺度变化与小目标检测困难的问题,在YOLOv5的基础框架上,提出了一种基于注意力机制与图特征金字塔网络(Graph Feature Pyramid Networks,GFPN)相结合的线夹护套目标检测方法。为了提高线夹护套图像中各部件边界信息的检测能力,通过超像素分层构建了GFPN结构,并设计了三种图神经网络(Graph Neural Network,GNN)。采用通道注意力与空间注意力机制进行消息传播,设计了GNN与特征金字塔网络(Feature Pyramid Networks,FPN)的特征映射规则。建立了螺栓型耐张线夹护套数据集,实验结果表明GFPN-YOLOv5模型对线夹护套的检测准确度提高了5.6%,同时在线夹护套样本多尺度变化和小目标场景下具有较好的泛化能力。

The lack of clamp sheath on the overhead line of the distribution network is a typical construction defect.Aiming at the problems of multi-scale change and difficulty in small target detection of overhead line route jacket in distribution network,a line clamp sheath target detection method based on the combination of attention mechanism and Graph Feature Pyramid Networks(GFPN)was proposed on the basic framework of YOLOv5.In order to improve the detection ability of the boundary information of each part in the wire clip sheath image,the GFPN structure was constructed by superpixel layering,and three graph neural networks(GNN)were designed.The channel attention and spatial attention mechanisms were used for message propagation,and the feature mapping rules of GNN and Feature Pyramid Networks(FPN)were designed.The experimental results show that the GFPN-YOLOv5 model improves the detection accuracy of wire clamp sheath by 5.6%,and has good generalization ability under multi-scale changes and small target scenarios of online clamp sheath samples.