基于卷积块注意模型的YOLOv3输电线路故障检测方法

Fault Detection of YOLOv3 Transmission Line Based on Convolutional Block Attention Model

针对航拍巡检图像中待检测目标易受复杂背景和部分遮挡影响而造成传统算法难以准确检测的问题,提出一种基于卷积块注意模型的YOLOv3输电线路故障检测方法。首先,在YOLOv3算法框架中融合卷积块注意模型来提升图像中故障目标区域的显著度;然后,通过引入高斯函数对非极大值抑制方法进行改进,降低存在部分遮挡目标的漏检率;其次,采用Focal Loss改进损失函数来提高检测网络的检测精度;最后,利用某供电局近3年无人机巡检视频制作训练样本和测试样本,并将提出的算法与4种经典目标检测算法进行比较。实验结果表明,相比于4种对比算法,该文算法能够在保证较高检测精度的同时具有较好的实时性。算法的平均检测精度可达94.6%,分辨率为1280´720的图像检测速度为40帧/s。

The targets to be detected in aerial inspection images are easily affected by complex background and partial occlusion,which makes it difficult for traditional algorithms to detect accurately.Aiming at this problem,a fault detection method of YOLOv3 transmission line based on convolutional block attention model is proposed.Firstly,in the YOLOv3 algorithm framework,the convolutional block attention module is fused to improve the saliency of the fault target area in aerial inspection images.Secondly,the non-maximum suppression method is improved by introducing Gaussian function to reduce the missing rate of the partially occluded targets.Thirdly,the loss function is adopted to improve the detection accuracy of the detection network.Finally,the training samples and test samples are prepared by using the aerial inspection video clips of a power supply bureau in the past three years,and the algorithm proposed in this paper is compared with four classical target detection algorithms.The experimental results show that compared with the four classical algorithms,the proposed algorithm can guarantee higher detection accuracy and better real-time performance.The average detection accuracy of this algorithm can reach to 94.6%,and the image detection speed of 1280×720 is 40 frames per second.