基于机器学习的织物疵点检测  

Detection of Fabric Defects Based on Machine Learning

作　　者：王帅 岳鹏飞 董晗睿[1,2] 侯爽 余灵婕 WANG Shuai;YUE Peng-fei;DONG Han-rui;HOU Shuang;YU Ling-jie(School of Textile Science and Engineering,Xi'an Polytechnie University,Xil an 710048,China;Key Laboratory of Functional Textile Material and Product,Ministry of Education,Xi'an Polytechnic University,Xi'an 710048,China)

机构地区：[1]西安工程大学纺织科学与工程学院,陕西西安710048 [2]西安工程大学功能性纺织材料及制品教育部重点实验室,陕西西安710048

出　　处：《纺织科技进展》2020年第10期25-30,共6页Progress in Textile Science & Technology

基　　金：国家级大学生创新创业训练计划资助项目(201910709012);陕西省自然科学基础研究计划项目(2019JQ-182)。

摘　　要：针对织物疵点自动检测的问题,选用破洞、稀纬、断经、结头、棉球、沾色、带纱等7种织物疵点类型,运用决策树、随机森林、支持向量机算法来对疵点数据训练,建立分类模型。选择了不同纹理和疵点类型共计4000幅织物图像作为训练和测试对象,交叉验证结果表明,随机森林模型效果最佳,模型准确率达97.92%。Focusing on the problem of automatic detection of fabric defects,a method of fabric defect detection was proposed.Using three machine learning algorithms including decision trees,random forests and support vector machine,the defect recognition model could detect seven kinds of defects such as broken holes,sparse wefts,warp breaks,knots,cotton balls,stains,and belt yarns.More than 4 000 fabric images with different textures and defects were used as training data to optimize the model,and the cross-validation results showed that the random forest performed best with the accuracy rate of 97.92% among the adopted three algorithms.

关 键 词：疵点检测 机器学习 决策树 随机森林 支持向量机 

分 类 号：TS391.4[轻工技术与工程] TS101.9