基于改进的YOLOv3输电线路异物检测  

Foreign object detection method for transmission lines based on improved YOLOv3

作　　者：熊鸿军[1] 熊旺 XIONG Hongjun;XIONG Wang(School of Business,Shanghai Dianji University,Shanghai 201306,China)

机构地区：[1]上海电机学院商学院,上海201306

出　　处：《上海电机学院学报》2023年第6期350-355,366,共7页Journal of Shanghai Dianji University

摘　　要：输电线路存在异物将会对电网的安全、稳定运行造成很大的影响。因此,对于输电线路异物检测十分重要。提出了一种基于改进的YOLOv3算法来识别输电线路的异物。将主干网络换为更有效的CSPDarkNet-53,该网络引入了性能更优异的残差模块;同时在特征金字塔层中增加了下采样通道,加强了特征融合;并在主干网络与特征金字塔连接处增加了注意力机制,以便有针对性地提取特征。实验结果表明:改进后的模型准确率达到85.02%。与原YOLOv3输电线路异物检测相比,准确率提高了7.21%,改进后的模型能准确有效地辨识出异物。Foreign items on transmission lines will have a great impact on the safe and stable operation of the power grid.Therefore,it is important for the detection of foreign objects on transmission lines.An improved YOLOv3 algorithm is proposed to identify foreign objects on transmission lines.First,a more effective CSPDarkNet-53 is utilized as the networks backbone,which introduces a residual module with better performance.Second,lower sampling channels are added to the pyramid layer to enhance feature fusion.Finally,an attention mechanism is added to extract targeted features at the connection between the backbone network and the feature pyramid.Experimental results show that the accuracy of the improved model reaches 85.02%.Compared with the original YOLOv3-based foreign object detection on the transmission line,the accuracy rate has increased by 7.21%,and the improved model can accurately and effectively identify foreign objects.

关 键 词：异物检测 特征金字塔层 残差模块 改进的YOLOv3 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]