作者：彭伟康1, 陈爱军1, 吴东明2, 朱利森2
作者单位：1. 中国计量大学计量测试工程学院，浙江 杭州 310018;
2. 东精集团有限公司，浙江 金华 321015
关键词：水准泡;缺陷检测;Faster R-CNN;特征金字塔;RFP
摘要：
目前水平尺制造行业采用人工方法对水准泡进行出厂检测，其准确率低、速度慢，该文提出一种基于改进Faster R-CNN的水准泡缺陷检测方法。采用ResNet101作为特征提取网络来避免网络退化，同时融合递归特征金字塔（recursive feature pyramid，RFP）得到多尺度的特征图输出，通过主干网络再训练的方式使输出特征图更好地适应模型检测任务。然后针对水准泡数据集样本目标来设计区域生成网络锚框，将得到的多尺度特征图输入区域生成网络进行候选区域提取。最后经过ROI Pooling层后得到水准泡缺陷检测结果。在包含
Defect detection method of level bubble based on improved Faster R-CNN
PENG Weikang1, CHEN Aijun1, WU Dongming2, ZHU Lisen2
1. College of Metrology and Measurement Engineering, China Jiliang University, Hangzhou 310018, China;
2. Dongjing Group Co., Ltd., Jinhua 321015, China
Abstract: Aiming at the problems of low precision and slow speed caused by manual inspection of leveling bubble in the current level ruler manufacturing industry, a defect detection method based on improved Faster R-CNN is proposed in this paper. ResNet101 is used as the feature extraction network to avoid network degradation. At the same time, the Recursive Feature Pyramid is fused to obtain multi-scale feature map output, and the output feature map is better adapted to the model detection task through the retraining of the backbone network. Then, the anchor frame of region generation network is designed for the sample target of level bubble data set, and the obtained multi-scale feature map is input into the region generation network to extract candidate regions. Finally, the level bubble defect detection results are obtained after the ROI Pooling layer. Experiments were performed on a data set containing
level bubble images. The experimental results show that the improved Faster R-CNN model fused with RFP can effectively improve the detection accuracy of the model, and the mean Average Precision on the test set reaches 96.7%.
Keywords: level bubble;defect detection;Faster R-CNN;feature pyramid;RFP
2021, 47(7):6-12 收稿日期: 2020-09-18;收到修改稿日期: 2020-11-14
基金项目:
作者简介: 彭伟康(1996-),男,浙江嘉兴市人,硕士研究生,专业方向为图像处理和模式识别研究
参考文献
[1] 魏煜, 朱善安. 最优阈值变换和轮廓跟踪在轮廓检测中的应用[J]. 计算机工程与应用, 2004(24): 88-90
[2] 盛伟, 王庆国, 朱善安. 基于Canny边缘检测和加权最小二乘法的气泡水平仪实时检测方法[J]. 机电工程, 2016, 33(10): 1182-1187
[3] 喻婷, 朱善安. 基于数学形态学的气泡水平尺自动检测系统研究[J]. 机电工程, 2017, 34(3): 213-218
[4] 王庆, 姚俊, 谭文禄, 等. 基于Faster R-CNN的排水管道缺陷检测研究[J]. 软件导刊, 2019, 18(10): 40-44
[5] 徐守坤, 王雅如, 顾玉宛, 等. 基于改进Faster RCNN的安全帽佩戴检测研究[J]. 计算机应用研究, 2020, 37(3): 901-905
[6] 王超洋, 樊绍胜, 刘铮, 等. 基于改进FasterRCNN的配网架空线路异常状态检测[J]. 电力学报, 2019, 34(4): 322-329
[7] GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition(CVPR), 2014: 580-587.
[8] 李爱莲, 刘浩楠, 郭志斌, 等. 改进ResNet101网络下渣出钢状态识别研究[J]. 中国测试, 2020, 46(11): 116-119,125
[9] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6): 1137-1149
[10] REDMON J, FARHADI A. YOLOv3: An Incremental Improvement[EB/OL]. [2018-04-08]. https://arxiv.org/pdf/1804.02767.pdf.
[11] LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot multibox detector[C]//European Conference on Computer Vision, 2016: 21-37.
[12] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 770-778.
[13] QIAO S Y, CHEN L C, ALAN Y. DetectoRS: Detecting objects with recursive feature pyramid and switchable atrous convolution[EB/OL]. [2020-06-03]. https://arxiv.org/pdf/2006.02334.pdf.
[14] LIN T, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017: 2117-2125.
