基于边缘计算和改进YOLOv5s算法的输电线路故障实时检测方法

Real-Time Detection Method for Transmission Line Faults Applying Edge Computing and Improved YOLOv5s Algorithm

随着输电线路无人机巡检工作的常态化,暴露出故障图像检测实时性、模糊目标检测精准性难以满足实际工作需求的问题。文章提出一种基于边缘计算和改进YOLOv5s算法的输电线路故障实时检测方法。以YOLOv5s为基础检测模型,基于Ghost轻量化模块重构模型获取数据特征的卷积操作过程,提高了模型的检测速度;采用基于KL散度分布的损失函数作为目标框定位损失函数,提升了模型对模糊图像检测的精度。将改进的YOLOv5s算法部署于华为Atlas 200 DK边缘模块中,对绝缘子自爆、防震锤脱落、鸟巢3类故障进行检测,其平均精度均值可达84.75%,检测速度为34 frame/s。结果表明,改进的算法在保证检测实时性的同时,能够提升对模糊故障目标图像的检测精度,满足无人机搭载边缘设备的输电线路巡检需求。

With the normalization of unmanned aerial vehicle(UAV) inspection of transmission lines, the problems of real-time detection of fault images and accuracy of blurred target detection are difficult to meet the actual work requirements. This paper proposes a real-time detection method for transmission line faults, which is based on edge computing and improved YOLOv5 s algorithm. This method is based on YOLOv5 s model, and the model is reconstructed on the basis of Ghost lightweight module to realize the convolution operation process of obtaining data features, which improves the detection speed of the model. The loss function based on KL(Kullback-Leibler) divergence distribution is used as the target box localization loss function to improve the accuracy of blurred image detection. The improved YOLOv5 s algorithm is deployed in the Huawei Atlas 200 DK edge module to detect three types of faults: insulator self-explosion, shock hammer falling-off, and bird’s nest. The mAP can reach 84.75%, and the detection speed is 34 frame/s. The results show that the improved algorithm can improve the detection accuracy of blurred fault target images while ensuring the real-time detection, and meet the inspection requirements of transmission lines equipped with edge devices by UAV.