摘 要：为了提升工业零件缺陷检测的精度和速度，在 Mask R-CNN 的基础上，引入了引导锚框的 anchor 生成方案提升检测精度；在此基础上对 Mask R-CNN 网络框架进行改进，去掉 Mask 分支，实现检测速度的优化。采用的数据集是DAGM 工业缺陷数据集，并与先前的代表方法进行对比实验。实验表明，改进后的算法在检测精度方面对比原始算法提升了约1.94%，且速度也提升了 1.2 frame/s，提升了工业零件缺陷检测的速度和精度。
关键词：工业缺陷检测；Mask R-CNN；引导锚框
DOI:10.19850/j.cnki.2096-4706.2022.03.037
基金项目：安徽省自然科学基金面上项目（1908085MF189）
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2022）03-0137-04
Industrial Parts Defect Detection Technology Based on Improved Mask R-CNN Algorithm
SHANG Jie, WU Guanmao
(College of Computer Science and Engineering, Anhui University of Science & Technology, Huainan 232001, China)
Abstract: In order to improve the accuracy and speed of industrial parts defect detection, on the basis of Mask R-CNN, the anchor generation scheme of the guide anchor frame is introduced to improve the detection accuracy. on this basis, the Mask R-CNN network framework is improved and Mask branch is removed to realize the optimization of detection speed. The data set used is the DAGM industrial defect data set, and a comparison experiment with the previous representative method is carried out. Experiments show that the improved algorithm has improved by 1.94% in detection accuracy compared with the original algorithm, and the speed has also improved by 1.2 frame/s, the speed and accuracy of industrial parts defect detection are improved.
Keywords: industrial defect detection; Mask R-CNN; guide anchor frame
参考文献：
[1] GIRSHICK R，DONAHUE J，DARRELL T，et al.Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation [C]//2014 IEEE Conference on Computer Vision and Pattern Recoginition.Columbus：IEEE，2014：580-587.
[2] GIRSHICK R.Fast R-CNN [C]//2015 IEEE International Conference on Computer Vision（ICCV）.Santiago：IEEE，2015：1440-1448.
[3] REN S，HE K，GIRSHICK R，et al.Faster R-CNN：Towards Real-Time Object Detection with Region Proposal Networks [J].IEEE Trans Pattern Anal Mach Intell，2017，39（6）：1137-1149.
[4] REDMON J，DIVVALA S，GIRSHICK R，et al.You Only Look Once：Unified， Real-Time Object Detection [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）.Las Vegas：IEEE，2016：779-788.
[5] BOCHKOVSKIY A，WANG C Y，LIAO H Y M.YOLOv4：Optimal Speed and Accuracy of Object Detection [J/OL].arXiv：2004.10934 [cs.CV].[2022-01-02].https://arxiv.org/abs/2004.10934.
[6] HE K， GKIOXARI G，DOLLAR P， et al. Mask R-CNN [C]//2017 IEEE International Conference on Computer Vision（ICCV）.Venice：IEEE，2017：2980-2988.
[7] WANG J，CHEN K，YANG S，et al.Region Proposal by Guided Anchoring [C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition（CVPR）.Long Beach：IEEE，2019：2960-2969.
[8] ZOPH B，CUBUK E D，GHIASI G，et al.Learning Data Augmentation Strategies for Object Detection [J/OL].arXiv：1906.11172 [cs.CV].[2022-01-03].https://arxiv.org/abs/1906.11172.
[9] YU X，GONG Y，JIANG N，et al.Scale Match for Tiny Person Detection [C]//2020 IEEE Winter Conference on Applications of Computer Vision（WACV）.Snowmass Village：IEEE，2020：1246-1254.
作者简介：尚洁（1996—），女，汉族，江苏连云港人，硕士在读，研究方向：目标检测等；通讯作者：吴观茂（1965—），男，汉族，安徽淮南人，副教授，博士，研究方向：计算机深度学习等。