间隔棒安装机器人识别安装间隙的分割算法  

Semantic segmentation algorithm for spacer bar installation robot to recognize installation gap

作　　者：周雨涵 王旭红[1] 邹德华[2] 吴荣骐 Zhou Yuhan;Wang Xuhong;Zou Dehua;Wu Rongqi(School of Electrical and Information Engineering,Changsha University of Science&Technology,Changsha 410114,China;Hunan Electric Power Company Ultra High Voltage Transmission Company,Changsha 410100,China)

机构地区：[1]长沙理工大学电气与信息工程学院,长沙410114 [2]湖南省电力公司超高压输电公司,长沙410100

出　　处：《现代计算机》2024年第18期22-27,82,共7页Modern Computer

摘　　要：针对间隔棒安装机器人作业过程中,存在间隔棒安装间隙识别精度不高,影响间隔棒定位、分割速度慢、不利于在移动端嵌入式设备上进行部署等问题,提出了一种基于改进的DeepLabV3+轻量级间隔棒安装机器人安装间隙识别算法。使用轻量级MobileNetV3网络减少参数量和计算量;对空间金字塔池化模块(ASPP)进行先降维再升维,在保证模型分割精度的同时减少模型的参数量;引入ECA注意力机制以得到更精确的目标边界;结合Focal Loss和Dice Loss损失函数提高模型对间隔棒安装间隙的分割精确度。实验结果表明,改进DeepLabV3+算法平均交并比(MIoU)和平均像素精度(MPA)分别提升了8.25和1.85个百分点,预测速度提升了22.47%,将改进算法训练的模型部署到机器人上进行实验,识别出的间隔棒安装间隙与现场实验过程相符,验证了改进算法的有效性。Aiming at the problems of spacer bar mounting robot’s operation process,such as the spacer bar mounting gap recognition accuracy is not high enough to affect the spacer bar localization,and the slow segmentation speed is not favorable for the deployment on mobile embedded devices,a lightweight spacer bar mounting robot mounting gap recognition algorithm based on the improved DeepLabV3+is proposed.The lightweight MobileNetV3 network is used to reduce the number of parameters and computation;the ASPP is first downscaled and then upscaled to reduce the number of parameters of the model while guaranteeing the segmentation accuracy of the model;and the ECA attention mechanism is introduced to obtain more accurate target boundaries;the combined Loss function of Focal Loss and Dice Loss was used to enhance the segmentation effect of the installation gap of spacer rod.The experimental results show that the MIoU and MPA of the improved DeepLabV3+algorithm are improved by 4.25 and 1.85 percentage point,respectively,and the prediction speed is improved by 22.47%.The model trained by the improved algorithm is deployed to the robot for experiments,and the spacer bar mounting gaps identified are in line with the field experimental process,which verifies the effectiveness of the improved algorithm.

关 键 词：间隔棒安装机器人 安装间隙 改进DeepLabV3+ 轻量化 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP242[自动化与计算机技术—计算机科学与技术] TM75[电气工程—电力系统及自动化]