基于Faster R-CNN的圆柱形金属工件表面缺陷检测
|
摘要
为有效进行圆柱形金属工件表面缺陷检测,设计一套利用深度学习框架——Faster R-CNN的表面缺陷检测系统。该系统算法利用Resnet网络进行特征提取,采用RPN方法提取缺陷的候选目标矩形区域,再利用Fast R-CNN在候选区域基础上进行缺陷检测。此外,为克服金属表面反光并获得高质量图片,设计一套合适的图像采集系统。实验表明,该检测系统能有效克服光滑金属表面的强反射,从而获取高质量图片;同时利用基于Faster R-CNN框架的方法进行缺陷检测,较好地解决了圆柱形金属表面缺陷检测能力弱的问题,在置信度阈值为0.9时,其查全率为95.0%,查准率为96.0%,检测速度为65ms/幅。
A surface defect detection system is proposed in this paper for the quality inspection of cylindrical workpieces. The surface defect detection algorithm is based on Faster R-CNN. Firstly,the Resnet network is used to extract the feature map. Then the RPN is adapted to extract candidate rectangle area. Finally,the defects are detected by Fast R-CNN. What's more,to overcome the reflection of metal surface and get high quality pictures for detecting,an image acquisition system is designed. Experiments show that the system has an advantage in overcoming the reflection of metal and getting high quality pictures of cylindrical metal workpieces. Meanwhile the detection algorithm based on Faster R-CNN can effectively detect defects. When the confidence threshold is 0.9,the recall rate is 95.0% and the precision rate is 96.0%,and the speed of detection is 65 ms per picture.
A surface defect detection system is proposed in this paper for the quality inspection of cylindrical workpieces. The surface defect detection algorithm is based on Faster R-CNN. Firstly,the Resnet network is used to extract the feature map. Then the RPN is adapted to extract candidate rectangle area. Finally,the defects are detected by Fast R-CNN. What's more,to overcome the reflection of metal surface and get high quality pictures for detecting,an image acquisition system is designed. Experiments show that the system has an advantage in overcoming the reflection of metal and getting high quality pictures of cylindrical metal workpieces. Meanwhile the detection algorithm based on Faster R-CNN can effectively detect defects. When the confidence threshold is 0.9,the recall rate is 95.0% and the precision rate is 96.0%,and the speed of detection is 65 ms per picture.
引文
[1] SU J H,LIU S L. Measure system of surface flaw and morphology analysis of cylindrical high precision parts[J]. Laser&Optoelectronics Progress,2014,51(4):041202.
[2]罗菁,董婷婷,宋丹,等.表面缺陷检测综述[J].计算机科学与探索,2014(9):1041-1048.
[3]汤勃,孔建益,伍世虔.机器视觉表面缺陷检测综述[J].中国图象图形学报,2017(12):1640-1663.
[4]韩斌.计算机视觉检测中的若干问题研究及应用[D].无锡:江南大学,2009.
[5]孙宝华,韩跃平,徐青,等.基于SURF算法的产品表面缺陷检测研究[J].红外技术,2014,36(6):503-507.
[6]张学武,丁燕琼,闫萍.一种基于红外成像的强反射金属表面缺陷视觉检测方法[J].光学学报,2011,31(3):112-120.
[7] LIU H,GUO S,CHEN Y F,et al. Shearography using wave-defect interactions for crack detection in metallic structures[C]. Washington D. C.:Proceedings of the 2017 IEEE International Ultrasonics Symposium(IUS),2017.
[8]李明,刘悦.基于支持向量机的金属表面缺陷检测算法[J].世界有色金属,2017,65(2):120-121.
[9] GIRSHICK R. Fast R-CNN[C]. Proceedings of the 15th IEEE International Conference on Computer Vision,2015:1-9.
[10] REN S,HE K,GIRSHICK R,et al. Faster R-CNN:towards realtime object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(6):1137-1149.
[11]史凯静,鲍泓,徐冰心,等.基于Faster RCNN的智能车道路前方车辆检测[J].计算机工程,2018,44(7):36-41.
[12]李伟山,卫晨,王琳.改进的Faster RCNN煤矿井下行人检测算法[J].计算机工程与应用,1-16.[2018-06-21]. http://kns.cnki.net/kcms/d LI Weishan.
[13]戴陈卡,李毅.基于Faster RCNN以及多部件结合的机场场面静态飞机检测[J].计算机应用,2017,37(S2):85-88.
[14] SHI Y,LI Y,WEI X,et al. A faster-RCNN based chemical fiber paper tube defect detection method[J]. 2017,173-7.
[15] YI L,LI G,JIANG M. An end-to-end steel strip surface defects recognition system based on convolutional neural networks[J]. Steel Research International,2017,88(2):1600068.
[16] WEIMER D,SCHOLZ-REITER B,SHPITALNI M. Design of deep convolutional neural network architectures for automated feature extraction in industrial inspection[J]. CIRP Annals,2016,65(1):417-20.
[17] HE K,ZHANG X,REN S,et al. Deep residual learning for image recognition[C]. Las Vegas:The 29th IEEE Conference on Computer Vision and Pattern Recognition,2016.
[18] UIJLINGS J R R,VAN DE SANDE K E A,GEVERS T,et al.Selective search for object recognition[J]. International Journal of Computer Vision,2013,104(2):154-171.
[19]栗琳.基于视觉的高反射球面缺陷快速检测关键技术研究[D].天津:天津大学,2013.
[20]张静,叶玉堂,谢煜,等.金属圆柱工件缺陷的光电检测[J].光学精密工程,2014,50(7):1871-1876.
[21]闵永智,岳彪,马宏锋.基于图像灰度梯度特征的钢轨表面缺陷检测[J].仪器仪表学报,2018,39(4):220-229.
[2]罗菁,董婷婷,宋丹,等.表面缺陷检测综述[J].计算机科学与探索,2014(9):1041-1048.
[3]汤勃,孔建益,伍世虔.机器视觉表面缺陷检测综述[J].中国图象图形学报,2017(12):1640-1663.
[4]韩斌.计算机视觉检测中的若干问题研究及应用[D].无锡:江南大学,2009.
[5]孙宝华,韩跃平,徐青,等.基于SURF算法的产品表面缺陷检测研究[J].红外技术,2014,36(6):503-507.
[6]张学武,丁燕琼,闫萍.一种基于红外成像的强反射金属表面缺陷视觉检测方法[J].光学学报,2011,31(3):112-120.
[7] LIU H,GUO S,CHEN Y F,et al. Shearography using wave-defect interactions for crack detection in metallic structures[C]. Washington D. C.:Proceedings of the 2017 IEEE International Ultrasonics Symposium(IUS),2017.
[8]李明,刘悦.基于支持向量机的金属表面缺陷检测算法[J].世界有色金属,2017,65(2):120-121.
[9] GIRSHICK R. Fast R-CNN[C]. Proceedings of the 15th IEEE International Conference on Computer Vision,2015:1-9.
[10] REN S,HE K,GIRSHICK R,et al. Faster R-CNN:towards realtime object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(6):1137-1149.
[11]史凯静,鲍泓,徐冰心,等.基于Faster RCNN的智能车道路前方车辆检测[J].计算机工程,2018,44(7):36-41.
[12]李伟山,卫晨,王琳.改进的Faster RCNN煤矿井下行人检测算法[J].计算机工程与应用,1-16.[2018-06-21]. http://kns.cnki.net/kcms/d LI Weishan.
[13]戴陈卡,李毅.基于Faster RCNN以及多部件结合的机场场面静态飞机检测[J].计算机应用,2017,37(S2):85-88.
[14] SHI Y,LI Y,WEI X,et al. A faster-RCNN based chemical fiber paper tube defect detection method[J]. 2017,173-7.
[15] YI L,LI G,JIANG M. An end-to-end steel strip surface defects recognition system based on convolutional neural networks[J]. Steel Research International,2017,88(2):1600068.
[16] WEIMER D,SCHOLZ-REITER B,SHPITALNI M. Design of deep convolutional neural network architectures for automated feature extraction in industrial inspection[J]. CIRP Annals,2016,65(1):417-20.
[17] HE K,ZHANG X,REN S,et al. Deep residual learning for image recognition[C]. Las Vegas:The 29th IEEE Conference on Computer Vision and Pattern Recognition,2016.
[18] UIJLINGS J R R,VAN DE SANDE K E A,GEVERS T,et al.Selective search for object recognition[J]. International Journal of Computer Vision,2013,104(2):154-171.
[19]栗琳.基于视觉的高反射球面缺陷快速检测关键技术研究[D].天津:天津大学,2013.
[20]张静,叶玉堂,谢煜,等.金属圆柱工件缺陷的光电检测[J].光学精密工程,2014,50(7):1871-1876.
[21]闵永智,岳彪,马宏锋.基于图像灰度梯度特征的钢轨表面缺陷检测[J].仪器仪表学报,2018,39(4):220-229.