孙旋;高小淋;曹高帅;

• 1:桂林理工大学机械与控制工程学院
• 2:郑州电力职业技术学院电力工程系

摘要(Abstract)：

针对织物疵点面积小且长宽比跨度大的问题，提出一种基于改进Faster R-CNN的多种织物疵点检测算法。以Faster R-CNN检测算法为基础，选取优化后的ResNet50作为Faster R-CNN的主干网络，在保持ResNet50深度不变的情况下，拓宽残差结构宽度，通过调整网络部分层结构并优化网络参数，使网络提取更多特征信息并减少网络计算量。针对织物疵点检测精度低的问题，在Faster R-CNN中引用FPN网络进行多尺度预测，并将改进的K-means聚类算法生成的预测框代替原Faster R-CNN中人工设计的预测框，增强网络聚焦“小目标”疵点的特征能力，进一步提高疵点检测精度。实验结果表明：相较于原Faster R-CNN,基于改进的Faster R-CNN在平均精度上提高了6.6%,且对于“小目标”与“细长型”疵点，识别率分别高达95%与97%,在织物疵点检测中具有较好的应用价值。

关键词(KeyWords)： 织物疵点;疵点检测;Faster R-CNN;优化ResNet50;改进的K-means

Abstract:

Keywords:

基金项目(Foundation): 国家自然科学基金项目(52065016)

作者(Authors): 孙旋;高小淋;曹高帅;

DOI: 10.19333/j.mfkj.20220305708

参考文献(References)：

• [1] 李飞.基于低秩分解的织物疵点检测方法研究[D].天津：天津工业大学，2021.LI Fei.Research on fabric defect detection method based on low-rank decomposition [D].Tianjin:Tiangong University,2021.
• [2] 许玉格，钟铭，吴宗泽，等.基于深度学习的纹理布匹瑕疵检测方法[J].自动化学报，2020.DOI:10.16383/j.aas.c200148.XU Yuge,ZHONG Ming,WU Zongze,et al.Detection method of textured cloth defects based on deep learning [J].Journal of Automation,2020.DOI:10.16383/j.aas.c200148.
• [3] 刘洋.基于深度学习的工业钢材瑕疵检测研究[D].合肥：安徽建筑大学，2021.LIU Yang.Research on industrial steel defect detection based on deep learning [D].Hefei:Anhui Jianzhu University,2021.
• [4] 孙禹锋.基于深度学习的经编布匹瑕疵检测软件设计与实现[D].上海：东华大学，2021.SUN Yufeng.Design and implementation of defect detection software for warp knitted fabrics based on deep learning [D].Shanghai:Donghua University,2021.
• [5] 段春梅，张涛川，李大成，等.基于残差网络和迁移学习的瓶坯表面质量检测研究[J].机械工程与自动化，2021(2):138-139,143.DUAN Chunmei,ZHANG Taochuan,LI Dacheng,et al.Research on surface quality detection of preform based on residual network and transfer learning [J].Mechanical Engineering and Automation,2021(2):138-139,143.
• [6] 张丽瑶，王志鹏，徐功平.基于SSD的织物瑕疵检测的研究[J].电子设计工程，2020,28(6):40-44.ZHANG Liyao,WANG Zhipeng,XU Gongping.Research on fabric defect detection based on SSD [J].Electronic Design Engineering,2020,28(6):40-44.
• [7] 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.CHE Xiangjiu,LIU Hualuo,SHAO Qingbin.Improved cloth defect recognition algorithm based on Fast RCNN [J].Journal of Jilin University (Engineering Edition),2019,49(6):2038-2044.
• [8] 何永贵.基于聚合通道特征与局部视觉显著性的布匹瑕疵识别方法研究[D].镇江：江苏大学，2020.HE Yonggui.Research on cloth defect recognition method based on aggregated channel features and local visual saliency [D].Zhenjiang:Jiangsu University,2020.
• [9] 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.WU Zhiyang,ZHUO Yong,LI Jun,et al.Fast detection algorithm for monochrome cloth defects based on convolutional neural network [J].Journal of Computer Aided Design and Graphics,2018,30(12):2262-2270.
• [10] 王生伟.基于图像处理的服装织物质量检测系统[D].南京：东南大学，2019.WANG Shengwei.Garment fabric quality detection system based on image processing [D].Nanjing:Southeast University,2019.
• [11] 陈宇俊.基于深度卷积神经网络的织物瑕疵检测方法研究[D].广州：广东工业大学，2018.CHEN Yujun.Research on fabric defect detection method based on deep convolutional neural network [D].Guangzhou:Guangdong University of Technology,2018.
• [12] 郜仲元，余灵婕，章玉铭，等.基于改进Pix2PixGAN的织物疵点检测算法[J].毛纺科技，2022,50(3):9-13.GAO Zhongyuan,YU Lingjie,ZHANG Yuming,et al.Fabric defect detection algorithm based on improved Pix2PixGAN [J].Wool Textile Journal,2022,50(3):9-13.
• [13] 李宇，刘孔玲，黄湳菥.基于深度学习的布匹疵点检测方法[J].毛纺科技，2021,49(4):98-103.LI Yu,LIU Kongling,HUANG Nanyu.Fabric defect detection method based on deep learning [J].Wool Textile Journal,2021,49(4):98-103.