广东电力2025，Vol.38Issue(3)：37-45,9.DOI:10.3969/j.issn.1007-290X.2025.03.005
基于双重注意力机制的电力工件位姿估计算法
Power Workpiece Pose Estimation Algorithm Based on Double Attention Mechanism
摘要
Abstract
To address the lack of deep learning datasets in the field of live-line working and the low accuracy of power equipment pose estimation,a custom dataset comprising three common types of power equipment under occluded scenario was constructed s and a dual-attention mechanism-based pose estimation algorithm was proposed.Firstly,to enable the model to adaptively capture long-range dependencies,a dual-attention mechanism was employed to aggregate global feature information from images.Secondly,to enhance the feature fusion efficiency of the attention mechanism,a residual cascading module based on dual attention,which integrates local and global features through residual connections was introduced.Finally,by incorporating a novel constraint to strengthen the back propagation process and refining the loss function,the information loss during feature dimension recovery was mitigated.Experiments on the custom power equipment dataset demonstrated that the proposed algorithm extracts richer semantic information,achieving an accuracy of 81.74％based on the average 3D model point distance,significantly improving prediction precision.This method enhances the performance of power equipment pose estimation and provides reliable technical support for live-line working operations.关键词
图像处理/深度学习/位姿估计/视觉定位/电力工件Key words
image processing/deep learning/pose estimation/visual localization/power workpiece分类
动力与电气工程引用本文复制引用
黄明民,曲睿,李勇,黄瀚樟..基于双重注意力机制的电力工件位姿估计算法[J].广东电力,2025,38(3):37-45,9.基金项目
国家自然科学基金项目(61720106009、61773359) （61720106009、61773359）
广西自然科学基金项目(2022GXNSFBA035661) （2022GXNSFBA035661）
