带钢表面缺陷实时图像检测与识别关键技术研究

项目名称： 带钢表面缺陷实时图像检测与识别关键技术研究

项目编号： No.60805020

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 杨延西

作者单位： 西安理工大学

项目金额： 22万元

中文摘要： 带钢表面缺陷不仅影响了产品外观，而且降低了产品性能。目前，国内外主要采用机器视觉技术进行带钢表面缺陷检测，多通过CCD摄像机在带钢上扫描成像，经过图像预处理、特征提取等操作，提取出图像几何、灰度、纹理等特征参数，然后进行图像识别判断。该技术不仅能够有效地检测出缺陷并进行分类，而且能够根据缺陷信息判定带钢的质量等级，是目前研究的热点方向。现有方法存在的主要不足有：数据量庞大不利于实时处理；现有图像特征提取方法精度低；基于单一分类器进行缺陷识别，识别率低。本项申请针对现有难题，就其关键技术展开研究：研究高效的检测方法，构建高速实时的海量数字图像信息处理的硬件平台；基于小波多分辨率分析和马尔可夫随机场研究高性能的缺陷特征提取方法；基于支持向量机研究多分类器融合的缺陷自动识别分类技术，实现检测、识别智能化。该项目中关键技术的攻克，有效提高目前检测速度10倍以上，识别精度达到90％以上。

中文关键词： 带钢表面缺陷；表面检测；支持向量机；小波多分辨率分析；马尔可夫随机场

英文摘要： Steel strip surface defects not only affect the product appearance, but also lower the product performance. At present, machine-vision technology is primarily adopted for steel strip surface defect detection at home and abroad, which always gets images with CCD cameras scanning on the strip, and extracts image geometry, gray, texture, and other parameters by image preprocessing, feature extraction, and other operations, then proceed to image recognition judgments. The technology, which is currently a hot research direction, not only can effectively detect and classify defects, but also can determine the quality and grade of steel strip according to defect information. The main drawbacks of existing methods are: the huge volume of data adverse to real-time processing; the low precision of existing image feature extraction method; the low recognition rate of the method based on a single classifier to identify defects. This application directed against the existing problems studies on its key technology for an efficient detection method, constructs high-speed, real-time hardware platform for processing the massive digital image information, adopts high-performance defect-feature extraction methods based on Wavelet Multi-resolution Analysis and Markov Random Field(MRF), and applies Multi-Classifiers Fusion technology based on Support Vector Machines(SVM) to identify and classify defect automatically, and finally achieves a intelligent detection and identification method. The capture of key technologies in the project will improve the current detection rate to more than 10 times and make identification accuracy above 90%.

英文关键词： Steel Strip Surface Defect;Surface Inspection;Support Vector Machines;Wavelet Multi-Resolution Analysis;Markov Random Field