融合引导锚框算法的Faster-RCNN缺陷检测  

Faster-RCNN Part Defect Detection Based on Guided Anchoring Algorithm

作　　者：郭兰申[1] 李杨[1] 黄凤荣 钱法 GUO Lan-shen;LI Yang;HUANG Feng-rong;QIAN Fa(School of Mechanical Engineering Hebei University of Technology,Tianjin 300131,China)

机构地区：[1]河北工业大学机械工程学院,天津300131

出　　处：《机械设计与制造》2022年第4期160-164,共5页Machinery Design & Manufacture

基　　金：基于冗余的高精度MEMS惯性导航系统技术研究(61973333)。

摘　　要：针对传统零件表面缺陷检测方法准确性差效率低,无法满足智能制造需求的问题,提出基于Faster-RCNN深度学习算法的缺陷检测方法。在Faster-RCNN基本算法的基础上,引入引导锚框算法生成anchor方案,解决算法中anchor方案对本次检测的缺陷目标缺乏针对性、产生大量的冗余区域建议窗口的问题,以提高检测的准确性和效率;通过对比非极大值抑制中不同的IOU阈值对检测结果的影响,确定最优的IOU阈值,并设计零件缺陷样本采集方案,建立三种零件缺陷数据集,在此基础上对方法的有效性进行试验验证。实验结果表明,该方法能够大幅度提高零件表面缺陷检测的准确性和效率,各缺陷检测结果的平均精度可达97.7%以上,平均检测速度达到4.3 fps,满足了智能制造的急迫需求。To cope with the low accuracy and inefficiency problems in the detection of parts surface defects by traditional surface defects methods,a defect detection method based on Faster-RCNN deep learning algorithm is proposed in this work.Based on the basic algorithm of Faster-RCNN,the Guided Anchoring algorithm is introduced to generate the anchor scheme,which solves the problem that the anchor scheme in the algorithm lacks specificity for the defect target of this detection and generates a large number of redundant area suggestion windows.The problem is to improve the accuracy and efficiency of the detection;to compare the influence of different IOU thresholds on the detection results in the non-maximum suppression,determine the optimal IOU threshold,and design the part defect sample collection scheme to establish three parts defects.The data set,on the basis of which the effectiveness of the method is tested and verified.The experimental results show that the proposed method can greatly improve the accuracy and efficiency of surface defect detection.The average precision of each defect detection result can reach over 97.7%,and the detection speed can reach as high as 4.3 fps,which meets the urgent needs of intelligent manufacturing.

关 键 词：表面缺陷检测 卷积神经网络 深度学习 Faster-RCNN算法 引导锚框算法 

分 类 号：TH16[机械工程—机械制造及自动化] TP181[自动化与计算机技术—控制理论与控制工程]