摘 要:为解决当前轴承质检人工成本高,效率低的问题,在对现有轴承缺陷检测算法方案进行研究后,提出了一种基于语义分割的实时在线轴承缺陷识别系统。在六面体黑室中配合高精度电动转台采集高质量轴承照片,利用小波 SURE 自适应阈值去噪法对图像进行预处理。选用语义分割的方法进行缺陷检测,在传统 FCN 算法的基础上,提出采用 PSPNet 分割算法,该算法可以融合更多的全局信息以解决误分割、漏分割的问题。并对模型进行不断评估,优化进程。经实验测得识别准确率高达97.75%,符合工业质检需求。
关键词:缺陷检测;语义分割;六面体黑室;小波阈值去噪;PSPNet
DOI:10.19850/j.cnki.2096-4706.2022.03.039
基金项目:2021 国家级大学生创新训练计划项目(202111070008);校级教学研究重点项目(2020JYZD-04)
中图分类号:TP273+.4 文献标识码:A 文章编号:2096-4706(2022)03-0145-04
Research on Bearing Defect Detection System Based on Semantic Segmentation
YIN Zhenhan, REN Yafei, ZHANG Baochi, ZHANG Shuaibing, SHI Yibin
(Luoyang Institute of Science and Technology, Luoyang 471023, China)
Abstract: In order to solve the problems of high labor cost and low efficiency of bearing quality inspection, after studying the existing bearing defect detection algorithm scheme, a real-time online bearing defect recognition system based on semantic segmentation is proposed. In the hexahedral black chamber, high quality bearing photos are collected with high-precision electric turntable, and the images are preprocessed by wavelet SURE adaptive threshold denoising method. The semantic segmentation method is selected for defect detection. Based on the traditional FCN algorithm, the PSPNet segmentation algorithm is proposed, which can fuse more global information to solve the problem of false segmentation and missing segmentation. The model is continuously evaluated to optimize the process. The experimental results show that the recognition accuracy is as high as 97.75%, which meets the needs of industrial quality inspection.
Keywords: defect detection; semantic segmentation; hexahedral black chamber; wavelet threshold denoising; PSPNet
参考文献:
[1] 邓耀力 . 基于机器视觉的轮毂轴承缺陷检测方法研究 [J].中国设备工程,2019(13):106-108.
[2] 左羽,陶倩,吴恋,等 . 基于卷积神经网络的植物图像分类方法研究 [J]. 物联网技术,2020,10(3):72-75.
[3] 王嫣然,陈清亮,吴俊君 . 面向复杂环境的图像语义分割方法综述 [J]. 计算机科学,2019,46(9):36-46.
[4] 梁格颖,王文琪,汪文,等 . 基于空洞卷积的语义图像分割算法研究 [J]. 信息通信,2019(6):33-36.
[5] 杜柳青,余永维 . 磁瓦表面缺陷机器视觉检测与识别方法[J]. 图学学报,2014,35(4):590-594.
[6] 李超,孙俊 . 基于机器视觉方法的焊缝缺陷检测及分类算法 [J]. 计算机工程与应用,2018,54(6):264-270.
[7] KRIZHEVSKY A,SUTSKEVER L,HINTON G E. ImageNet classification with deep convolutional neural networks [J].Communications of the ACM,2017,60(6):84-90.
作者简介:尹振汉(2000—),男,汉族,河南信阳人,本科在读,研究方向:机器视觉。