摘 要：花色布匹在生产的过程中，其相比于单色布匹的生产需要引入更多的加工工序，比如印花、后整理等工序，经常导致花色布匹产生更多的瑕疵类别。为了实现花色布匹瑕疵的智能识别与检测，文章给出一种基于Cascade R-CNN 改进的花色布匹瑕疵智能识别与检测方法，实验结果表明，相比同类算法，文章提出的方法在花色布匹瑕疵数据集上识别的准确率提升了2.39%，mAP 评估指标提高了3.83% 的显著效果。
关键词：花色布匹瑕疵；目标识别；缺陷检测；深度学习；卷积神经网络
中图分类号：TP391.41；TS107 文献标识码：A 文章编号：2096-4706（2020）23-0020-05
Improved Intelligent Recognition Method of Pattern and Color Fabric Defects Based on Cascade R-CNN
LU Guijia
（Guangdong University of Technology，Guangzhou 510006，China）
Abstract：In the process of patterned and color fabric production，it needs to introduce more processing procedures than the production of monochrome fabric，such as printing and finishing，etc，which often leads to more defect categories in the pattern and color fabric. In order to realize the intelligent recognition and detection of the defects of the pattern and color fabric，this paper presents an improved method of intelligent recognition and detection of the pattern and color fabric defects based on Cascade R-CNN. The experimental results show that compared with similar algorithms，the method proposed in this paper improves the accuracy of recognition on the pattern and color fabric defect data set by 2.39%，and the mAP evaluation index is improved by 3.83%.
Keywords：pattern and color fabric defect；target recognition；defect detection；deep learning；convolutional neural network
参考文献：
[1] HENRY Y.T. NGAN A，Grantham K.H，etc. Automatedfabric defect detection—A review [J].Image and Vision Computing，2011，29 （7）：442-458.
[2] CHAN C H，PANG G K H. Fabric defect detection by Fourieranalysis [J].IEEE Transactions on Industry Applications，2000，36（5）：1267.
[3] 吴志洋，卓勇，李军，等. 基于卷积神经网络的单色布匹瑕疵快速检测算法 [J]. 计算机辅助设计与图形学学报，2018，30（12）：2262-2270.
[4]K R I Z H E V S K Y A ， S U T S K E V E R I ， H I N T O N G .ImageNet classification with deep convolutional neural networks [J].Communications of the ACM，2017，60（6）：84-90.
[5] 车翔玖，刘华罗，邵庆彬. 基于Fast RCNN 改进的布匹瑕疵识别算法 [J]. 吉林大学学报 （工学版），2019，49（6）：2038-2044.
[6] GIRSHICK R. Fast R-CNN [C]//2015 IEEE InternationalConference on Computer Vision (ICCV).Santiago：IEEE，2015:1440-1448.
[7] CAI Z W，VASCONCELOS N. Cascade R-CNN：DelvingInto High Quality Object Detection [C]//2018 IEEE/CVF Conference onComputer Vision and Pattern Recognition：IEEE，2018.
[8] LIN T Y，DOLLAR P，HE K，et al. Feature PyramidNetworks for Object Detection [C]//2017 IEEE Conference on ComputerVision and Pattern Recognition （CVPR）：IEEE，2017.
[9] NEUBECK A，GOOL L J V. Efficient Non-MaximumSuppression [C]//International Conference on Pattern Recognition：IEEE，2006.
[10] SHORTEN C，KHOSHGOFTAAR T M. A survey on ImageData Augmentation for Deep Learning [J].Journal of Big Data，2019，6（1）：1-48.
[11]K I S A N TAL M，WOJNA Z，MURAW S K J，e t a l .Augmentation for small object detection [C]//9th International Conferenceon Advances in Computing and Information Technolog：Aircc PublishingCorporation，2019.
[12] DENG J，DONG W，SOCHER R，et al. ImageNet：A largescalehierarchical image database [C]//IEEE Conference on ComputerVision & Pattern Recognition.Miami：IEEE，2009：248-255.
作者简介：陆贵家（1996—），男，壮族，广东广州人，硕士研究生在读，主要研究方向：深度学习，计算机视觉，目标检测。