基于CBAM注意力机制的输电线路语义分割算法  

Semantic segmentation algorithm of transmission lines based on Cbam attention mechanism

作　　者：刘娜 LIU Na(State Grid Economic and Technological Research Institute Co.,Ltd.,Yichang,Hubei 443002)

机构地区：[1]国网经济技术研究院有限公司,湖北宜昌443002

出　　处：《长江信息通信》2023年第9期60-62,共3页Changjiang Information & Communications

摘　　要：我国工业化进程加速,对电力的需求逐年上升,但输电线路建设多集中在高山和荒野之间,使得人工巡检的成本极高,而且效率低下。因此,近年来出现了各种检测系统,用于对输电线路进行故障排查。这些检测系统需要首先对复杂场景下的输电线路进行分割,因为分割准确性极大地影响后续的检测和排障任务。大多数分割算法都基于U型网络,这种网络对通道特征和空间特征进行平等处理,没有关注对分割重要的区域的优先权,从而影响了图像语义分割的性能。文章采用U-Net网络并在其基础上引入了CBAM注意力机制对空间特征以及通道特征进行校准,提升语义分割网络性能,在输电线路分割现实场景数据集中该算法取得了良好的效果。With the acceleration of industrialization in China,the demand for electricity has been increasing year by year.However,the construction of transmission lines is often concentrated in high mountains and wilderness,making manual inspection extremely costly and inefficient.Therefore,in recent years,various detection systems have emerged to troubleshoot transmission lines.These detection systems need to first segment the transmission lines in complex scenes,as the segmentation accuracy greatly affects subsequent detection and troubleshooting tasks.Most segmentation algorithms are based on U-shaped networks,which treat channel and spatial features equally without giving priority to the regions important for segmentation,thus affecting the performance of image semantic segmentation.In this paper,we adopt the U-Net network and introduce the CBAM attention mechanism to calibrate the spatial and channel features,improving the performance of the semantic segmentation network.This algorithm achieved good results in the real scene dataset of transmission line segmentation.

关 键 词：语义分割 U-Net架构 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]