基于深度学习的架空线路关键部件典型缺陷识别研究

Research on Typical Defects Identification of Key Components of Overhead Lines Based on Deep Learning

中文关键词:  YOLOv3  目标检测  RepVGG  SIoU  架空线路巡检

英文关键词:YOLOv3  target detection  RepVGG  SIoU  overhead line inspection

中文摘要:

      为了精确识别定位架空线路中关键部件的缺陷情况，提出了一种改进的YOLOv3检测方法，首先采用RepVGG模块替换骨干网络Darknet-53中的残差单元，加快推理速度；其次通过改进损失函数，引入SIoU使模型训练更快，精确率更高；最后通过改进检测头，采用不同的分支进行计算，提升检测效果。实验结果表明，改进方法与YOLOv3相比，精确率提高了3.4%，召回率提高了2.6%；性能相比于SSD、Faster R-CNN网络模型也具有一定优越性。

英文摘要:

      In order to accurately identify and locate the defects of key components in overhead lines,an improved YOLOv3 detection method is proposed.Firstly,the RepVGG module is used to replace the residual unit in the backbone network Darknet-53 to speed up inference.Secondly,by improving the Loss function,SIoU is introduced to make the model training faster and improve the accuracy.Finally,by improving the detection head and using different branches for calculation,the detection effect is improved.The experimental results show that compared with YOLOv3,the improved method improves the precision rate by 3.4% and the recall rate by 2.6%,it also has certain advantages compared with SSD and Faster R-CNN network models in performance.