基于分类算法的高实时性织物疵点检测  

Research on high real-time fabric defect detection based on classification algorithm

作　　者：杨晓波 白直灿 YANG Xiaobo;BAI Zhican(Zhejiang Shuren University,Hangzhou,Zhejiang 310015,China)

机构地区：[1]浙江树人学院,浙江杭州310015

出　　处：《毛纺科技》2023年第8期112-117,共6页Wool Textile Journal

基　　金：浙江省自然科学基金项目(Y1110023)。

摘　　要：为了进一步提高织物疵点的检测效率,提出一种基于机器学习的高实时性织物疵点检测算法。首先分析了织物疵点检测算法流程,通过学习和分类2个阶段对疵点进行检测;然后利用高斯函数对织物纹理图像进行建模,并提取4类纹理特征刻画织物纹理并突出纹理中包含的疵点,采用分类算法获得分类器完成疵点判定;最后通过对比实例验证文章所提算法的可行性。研究结果表明:本文所提算法的平均准确率为97%,比传统算法高10%;采用分类算法可以节约训练时间,在织物幅宽相近的情况下,检测速度可达41 m/min,可以满足工业环境中在线实时检测要求。In order to improve the efficiency of fabric defect detection,a high real-time fabric defect detection algorithm based on machine learning was proposed.Firstly,the algorithm flow of fabric defects detection was analyzed,and the defects were detected through two stages:learning and classification.Then,Gaussian function was used to model fabric texture image,and 4 types of texture features were extracted to characterize fabric texture and highlight defects contained in texture,and a classifier was obtained to complete the defect judgment.Finally,the feasibility of the proposed algorithm was verified by comparing examples.From the defect detection accuracy rate,the average accuracy of the proposed algorithm was 97%,which was higher than the traditional detection algorithm.The results show that the classification algorithm can save training time,and the correct rate of defects detection is 10% higher than the traditional algorithm.Under the condition of similar fabric width,the detection speed can reach 41 m/min,which can meet the requirements of online real-time detection in industrial environment.

关 键 词：机器学习 实时性 疵点检测 特征提取 分类学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]