改进YOLOv3的输电线路异物检测方法

Improved YOLOv3 foreign body detection method in transmission line

输电线路因所处环境复杂,极易附着异物,若不及时发现和清理将会对输电线路安全运行造成严重影响。针对输电线路图像巡检中的异物检测精度不高的问题,提出改进YOLOv3的输电线路异物检测方法(YOLOv3-RepVGG)。该方法基于YOLOv3目标检测网络并对其改进,首先采用RepVGG模块替换骨干网络Darknet-53的残差单元,同时加倍模块数量来提高网络对图像特征的提取能力;其次通过增加网络的多尺度检测框提升检测精度,采用CIOU损失函数来一步优化网络模型。实验结果表明,提出的YOLOv3-RepVGG方法与YOLOv3相比,输电线路异物检测m AP提高了9.8%,其中精确率提高19.5%,召回率提高1.2%;与目标检测SSD,Faster R-CNN网络相比,YOLOv3-RepVGG在性能上也具有一定优越性。

Due to the complex environment of the transmission line,foreign matter is easily attached to it.If it is not discovered and cleaned up in time,it will have a serious impact on the safe operation of the transmission line.Aiming at the problem of low accuracy of foreign body detection in transmission line image inspection,an improved YOLOv3 transmission line foreign body detection method(YOLOv3-RepVGG)is proposed.This method is based on the YOLOv3 target detection network and improves it.First,the RepVGG module is used to replace the residual unit of the backbone network Darknet-53,and the number of modules is doubled to improve the network’s ability to extract image features;secondly,by increasing the network’s multi-scale detection frame Improve the detection accuracy and use the CIOU loss function to optimize the network model in one step.Experimental results show that compared with YOLOv3,the proposed YOLOv3-RepVGG method improves the m AP of foreign body detection in transmission lines by9.8%,in which the accuracy rate is increased by 19.5%,and the recall rate is increased by 1.2%;compared with the target detection SSD,Faster R-CNN network,YOLOv3-RepVGG also has certain advantages in performance.