电子测量与仪器学报

马志程,李丹,张宝龙.基于改进 Mask R-CNN 的光学元件划痕缺陷检测研究[J].电子测量与仪器学报,2023,37(4):231-239

基于改进 Mask R-CNN 的光学元件划痕缺陷检测研究

Research on scratch defect detection of optical elements based on improved Mask R-CNN

DOI：

中文关键词: 缺陷检测 Mask R-CNN 注意力机制梯度均衡化的 Focal Loss

英文关键词:defect detection Mask R-CNN attention mechanism GHM-Focal Loss

基金项目:

作者	单位
马志程	1.天津科技大学电子信息与自动化学院
李丹	1.天津科技大学电子信息与自动化学院
张宝龙	1.天津科技大学电子信息与自动化学院

Author	Institution
Ma Zhicheng	1.School of Electronics and Automation, Tianjin University of Science and Technology
Li Dan	1.School of Electronics and Automation, Tianjin University of Science and Technology
Zhang Baolong	1.School of Electronics and Automation, Tianjin University of Science and Technology

摘要点击次数: 675

全文下载次数: 864

中文摘要:

光学元件缺陷会直接影响整个光学系统的性能,在光学元件缺陷检测中,划痕缺陷无疑是检测的难点,划痕缺陷存在着尺寸小,长宽比却比较大,易受杂质影响的问题,本文将深度学习算法应用到光学元件缺陷检测,并根据划痕缺陷的特点,对 Mask R-CNN 网络模型进行了改进,使算法对划痕缺陷也有了更好的检测效果。首先,将原有的 ResNet 更换为本文提出的 CSPRepResNet,并添加 ESE 注意力机制,提高了特征提取的能力并减少了计算量;其次,利用 K-means 算法重新聚类 anchor boxes 的长宽比例;再次,将目标检测的损失函数由 Cross Entropy 改为梯度均衡化的 Focal Loss,解决了正负样本不平衡问题的同时,更有利于对困难样本的检测,还可以消除离群点的影响。总体来说,检测的 mAP@ . 5 由原来的 52. 1%提高到 57. 3%,提高了 5. 2%,且推理速度几乎不变,可见,改进后 Mask R-CNN 对光学元件划痕缺陷有更好的检测效果。

英文摘要:

Optical element defects will directly affect the performance of the entire optical system. In the detection of optical element defects, scratch defects are undoubtedly the difficulty of detection. The scratch defects have the problems of small size, large aspect ratio, and easy to be affected by impurities. In this paper, depth learning algorithm is applied to optical element defect detection, and according to the characteristics of scratch defects, the Mask R-CNN network model is improved. The algorithm also has a better detection effect on scratch defects. First, the original ResNet is replaced by CSPRepResNet proposed in this paper, and ESE attention mechanism is added to improve the ability of feature extraction and reduce the amount of computation. Secondly, K-means algorithm is used to recluster the length width ratio of anchor boxes. Thirdly, the loss function of target detection is changed from Cross Entropy to gradient balanced Focal Loss, which solves the problem of imbalance between positive and negative samples, is more conducive to the detection of difficult samples, and can also eliminate the influence of outliers. In general, the tested mAP@ . 5 The original 52. 1% is increased to 57. 3%, an increase of 5. 2%, and the reasoning speed is almost unchanged. It can be seen that the improved Mask R-CNN has a better detection effect on optical element scratch defects.

查看全文查看/发表评论下载PDF阅读器