基于HMT模型的带钢表面缺陷检测研究  

RESEARCH ON SURFACE DEFECT DETECTION FOR STRIP STEEL BASED ON HIDDEN MARKOV MODEL

作　　者：徐海燕[1] 郭京[1] 

机构地区：[1]河南职业技术学院,河南郑州450046

出　　处：《计算机应用与软件》2015年第3期325-330,共6页Computer Applications and Software

摘　　要：为了减少带钢表面缺陷检测系统需要处理的数据量,提高系统检测效率,提出基于隐马尔科夫模型的带钢表面检测方法。该方法在检测系统获取图像数据后,先用相对简单的方法分割出图像中的缺陷可疑区域,然后根据带钢表面图像的特点,采用隐马尔可夫树模型(HMT)进行数据分析,并改进HMT模型参数,完成多尺度分割效果融合,获得最终的分割结果。在对带钢缺陷测试样本集的分割中,采用HMT模型为带钢表面图像建立背景和缺陷两个模型,尺度3缺陷检出率达到94.4%,相比高斯混合模型提升了5.5%,误检率达到18.8%,比高斯混合模型降低了2%。To reduce the amount of data required to be processed in surface defect detection system for strip steel and to improve system efficiency,we propose the hidden Markov model-based strip steel surface defect detection method. After the detection system obtaining image data,the method first segments with a relatively simple method the suspicious defects areas in the image,then according to the features of strip steel surface image it uses hidden Markov tree model（ HMT） to analyse the data and improves the parameters of HMT model to complete the fusion of multiscale segmentation effects,and obtains final segmentation result as well. When segmenting strip defect test sample set,the HMT model is adopted to build two models of background and defect for the strip surface image,the detection rate of scale-3 defect reaches up to 94. 4%. Compared with Gaussian mixture model,it improves by 5. 5%,and the false detection rate reaches 18. 8%,2 % lower than that of Gaussian mixture model.

关 键 词：带钢 图像分割 缺陷检测 HMT 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]