基于引导滤波和局部异质性度量的生丝疵点检测  

Raw silk defect detection based on guided filtering and local heterogeneity measurement

作　　者：卢鸯 杨言语 韦世界 韩高锋 曾凡高 李子印[2] LU Yang;YANG Yanyu;WEI Shijie;HAN Gaofeng;ZENG Fangao;LI Ziyin(Zhejiang Light Industrial Products Quality Inspection Institute,Hangzhou,310018,China;China Jiliang University,Hangzhou,310018,China)

机构地区：[1]浙江省轻工业品质量检验研究院,浙江杭州310018 [2]中国计量大学,浙江杭州310018

出　　处：《棉纺织技术》2025年第3期44-49,共6页Cotton Textile Technology

基　　金：国家市场监督管理总局科技计划项目(2022MK048);浙江省市场监督管理局青年科技项目(QN2023446);浙江省基础公益研究计划项目(LGN20F50001)。

摘　　要：针对生丝疵点图像背景与疵点差别极小的特点,提出一种基于引导滤波和局部异质性度量的生丝疵点检测方法。首先,使用引导滤波来抑制生丝疵点背景保留疵点边缘。其次,通过局部异质性度量来进一步抑制背景、增强疵点。根据局部结构信息占比率来度量疵点与背景的结构异质性,最大限度抑制背景;根据疵点与背景的灰度差异特征作为基础,构造局部灰度差异性度量以突出疵点。最后采用子块自适应阈值分割进行疵点分割。将生丝疵点图像分为亮背景图像数据集PQL-1以及暗背景图像数据集PQL-2各300张进行疵点检测试验,结果表明:该研究算法在亮、暗背景下检测率分别为86.9%及85.7%,虚警率分别为15.9%、12.3%,具有较好的场景鲁棒性。In view of the fact that the difference between the background and defects of raw silk defect images was extremely small,a raw silk defect detection method based on guided filtering and local heterogeneity measurement was proposed.Firstly,guided filter was used to suppress the raw silk defects and preserve the edges of the background.Secondly,local heterogeneity was used to further suppress the background and enhance the defects.According to the proportion of local structure information,the structural heterogeneity of the defect and background was measured,and the background was suppressed as much as possible.The local gray difference measure was constructed to highlight the defects.Finally,sub-block adaptive threshold segmentation was used for defect segmentation.The raw silk defect images were divided into 300 pieces of bright background image data set PQL-1 and 300 pieces of dark background image data set PQL-2.The results show that the detection rate of the algorithm was 86.9% and 85.7% in bright and dark background,the false alarm rate was 15.9% and 12.3%,respectively,which has better scene robustness.

关 键 词：生丝 疵点检测 引导滤波 局部异质性度量 子块自适应分割 

分 类 号：TS101.8[轻工技术与工程—纺织工程]