摘要
针对传统基于线结构光的视觉测量系统存在光条纹分割精度低的问题,提出了一种改进U-Net的光条纹分割算法。改进算法使用VGG16的卷积池化层代替U-Net编码块中的卷积池化层,在U-Net编-解码层间的跳连接中引入坐标注意力机制,在U-Net编码块末端接入金字塔池化模块,采用Dice函数和交叉熵函数的组合作为网络的损失函数,解决了光条纹占比失衡问题。基于线结构光测量原理,设计了工件尺寸测量系统。实验结果表明:改进U-Net算法的平均像素准确度(mpa)为95.61%,平均交并比(mIoU)为89.73%,均高于其他对比算法;工件测量尺寸的绝对误差小于0.1 mm,相对误差小于1%,重复精度小于0.2%,满足工件的检测要求。
To improve the accuracy of light stripe segmentation in the traditional vision measurement system based on line-structured light, an improved light stripe segmentation algorithm based on U-Net is proposed. The proposed algorithm uses the convolution pooling layer of VGG16 instead of that in the U-Net coding block, introduces the coordinate attention mechanism in the hop connection between U-Net coding and decoding layers, and connects the pyramid pooling module at the end of U-Net coding block. Additionally, it uses a combination of Dice function and cross entropy function as the loss function of the network, so as to solve the problem of imbalance of light stripe proportion. Based on the principle of line-structured light measurement, a workpiece size measurement system is designed. Experimental results show that the mean pixel accuracy(mpa) of the improved U-Net algorithm is 95. 61%and mean intersection over union(mIoU) is 89. 73%, which are higher than other comparison algorithms. The absolute error of workpiece measurement size is less than 0. 1 mm, the relative error is less than 1%, and the repetition accuracy is less than 0. 2%, meeting the inspection requirements of the workpiece.
作者
闫文伟
陈帅
穆宝岩
高亮
Yan Wenwei;Chen Shuai;Mu Baoyan;Gao Liang(Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016,Liaoning,China;Institutes for Robotics and Intelligent Manufacturing,Chinese Academy of Sciences,Shenyang 110169,Liaoning,China;University of Chinese Academy of Sciences,Beijing 100049,China;Key Laboratory on Intelligent Detection and Equipment Technology of Liaoning Province,Shenyang 110179,Liaoning,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2022年第12期323-335,共13页
Laser & Optoelectronics Progress
基金
中国科学院战略性先导科技专项(C类)(XDC04000000)
国家自然科学基金面上项目(62073312)
辽宁省重点研发计划(2020JH2/10100023)
中国航发自主创新专项资金项目(ZZCX-2018-035)
王宽诚教育基金会
辽宁省“兴辽英才计划”项目(XLYC2002055)。
关键词
线结构光
光条纹分割
深度学习
特征点提取
非接触测量
line-structured light
light stripe segmentation
deep learning
feature point extraction
non-contact measurement
