摘要
基于现有巡检机器人在颜色空间对呼吸器的缺陷检测,提出了一种特征匹配与深度学习相结合的呼吸器缺陷检测方法。该方法首先制作呼吸器脚本图片,利用SURF特征匹配算法裁剪出呼吸器定位大图;然后引入通道注意力机制,改进SSD目标检测算法,经过标注与训练,得到呼吸器缺陷检测和分类。经过对比实验,在硅胶变色状态检测精确度达到98.5%,在外观破损检测精确度达到98.0%。实际场景中,该方法单张图片检测时间0.46 s,基本满足实际要求。
Aiming at the problems that the existing inspection robots detect the defects of transformer respirators in the color space,there is a single type of detection defects and low detection accuracy,a method of respirator defect detection combining feature matching and deep learning is proposed.The accuracy of defect detection is improved,and the detection of various types of defects is realized.In this method,the script picture of the respirator is first made,and the large image of the respirator is cut out by using the SURF feature matching algorithm.Then,the channel attention mechanism is introduced to improve the SSD target detection algorithm.After labeling and training,the respirator defect detection and classification are obtained.Bycomparative experiments,the detection accuracy of the discoloration state of the silica gel reaches 98.5%,and the detection accuracy of the appearance damage reaches98.0%.In the actual scene,the method takes 0.46 s to detect a single image,which basically meets the actual requirements and provides a theoretical basis for the inspection robot to detect transformer respirators.
作者
项新建
潘磊
吴海腾
XIANG Xinjian;PAN Lei;WU Haiteng(School of Electrical Engineering and Automation,Zhejiang University of Science and Technology,Hangzhou 310023,China)
出处
《实验室研究与探索》
CAS
北大核心
2022年第6期57-61,97,共6页
Research and Exploration In Laboratory
基金
浙江省重点研发计划项目(2018C01085)。
关键词
巡检机器人
呼吸器缺陷检测
特征匹配
通道注意力机制
inspection robot
respiratordefect detection
feature matching
channel attention mechanism
作者简介
项新建(1964-),男,浙江永康人,硕士,教授,研究方向为人工智能、机器人、物联网理论与技术。Tel.:15868153622,E-mail:188002@zust.edu.cn。
