基于RGB-D边缘信息融合的接地设备图像语义分割

Semantic Segmentation of Grounding Equipment Image Based on RGB-D Edge Information Fusion

摘要: 传统的接地设备安装质量检测方法主要由工作人员采用手工测量和经验判断的方式来检查，工作量大，检测效率低且存在经验误差，无法准确地测量出安装螺栓的参数信息。针对这一问题，本文提出了一种基于RGB-D边缘信息融合的接地设备图像语义分割方法。首先将获取到的接地设备相连螺栓的可见光和深度图像进行预处理；然后引入注意力辅助模块帮助提取可见光图像特征和深度图像特征，并进行特征融合；通过金字塔池化模块防止过拟合，最后利用边缘信息辅助监督，输出最优的预测图像语义分割结果。实验结果表明，本文方法能够实现对接地设备相连螺栓图像的精确分割，并将其应用于接地设备的安装质量检测之中。

Abstract: The traditional method of inspecting the installation quality of grounding equipment is mainly performed by manual measurement and empirical judgment. The workload is large, the inspec-tion efficiency is low and there are empirical errors. It is difficult to get accurate parameters of the installation bolts. To address this problem, this paper proposes a semantic segmentation method for grounding equipment images based on RGB-D edge information fusion. Firstly, the visible and depth images of the connected bolts of the grounding equipment are preprocessed. Then, an attention assistance module is introduced to help extract visible image features and depth image features, and perform feature fusion. The pyramid pooling module is used to pre-vent overfitting, and the edge information is used to assist supervision, and the optimal predic-tive image semantic segmentation results are output. Experimental results show that this me-thod can achieve accurate segmentation of grounding equipment and apply it to the inspection of the quality of the grounding equipment installation.