基于CUDA计算GLCM特征值和SVM的织布疵点检测  

Fabric Defect Detection Based on Features of GLCM calculated Based on CUDA and SVM

作　　者：万东 孙志刚 肖力 WAN Dong;SUN Zhigang;XIAO Li(School of Automation,Huazhong University of Science and Technology,Wuhan 430074)

机构地区：[1]华中科技大学自动化学院,武汉430074

出　　处：《计算机与数字工程》2018年第4期828-832,共5页Computer & Digital Engineering

摘　　要：针对当前织布疵点检测的准确率和实时性问题,提出一种基于CUDA计算灰度共生矩阵特征值和支持向量机的检测算法;该算法借助GPU基于CUDA架构计算4个方向灰度共生矩阵各自的4种特征值组成16维特征向量输入训练完成的支持向量机模型,实现对各种类型疵点图像和无疵点图像的分类检测。实验结果表明,该算法系统在准确率和实时性上都能满足工业生产的需求。Focusing on the accuracy and real-time online problem of fabric defect detection,a detection algorithm based on features of gray level co-occurrence matrix calculated by cuda and support vector machine is proposed.In the algorithm,the sixteen-dimensional feature vector which contains four kinds of feature of gray level co-occurrence matrix of four directions calculated by cuda with the help of GPU is input to the trained support vector machine to classify the fabric images,some of which contain defects and others of which do not contain them.The experiment results show that the algorithm system can meet the needs of industrial production on the accuracy and real-time performance.

关 键 词：疵点检测 支持向量机 GPU CUDA 灰度共生矩阵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]