基于P-ReliefF特征选择方法的带钢表面缺陷识别

发布时间：2018-04-22 18:19

  本文选题：特征选择 + 带钢表面缺陷　； 参考：《电子测量与仪器学报》2017年07期

【摘要】：带钢表面缺陷纹理的复杂性和多样性、背景纹理中存在的伪缺陷等给现有的带钢表面缺陷特征提取和识别带来了极大的困难。为此,提出了一种新的带钢表面缺陷选择与识别方法。首先,通过各向异性扩散算法对带钢表面的伪缺陷干扰进行抑制;其次,利用提出的P-Relief F方法对表面缺陷特征进行选择,相比传统的Relief F方法,该方法考虑了不同维度特征之间的关联性;最后,利用筛选的特征集和支持向量机(SVM)核分类器对带钢表面缺陷进行分类与识别。实验结果表明,提出的方法能够提取出具有高区分性和鲁棒性的带钢表面缺陷特征,并且对于划痕、褶皱、凸起和污渍等不同类型的带钢表面缺陷,本方法相比传统的方法可以获得更高的识别率。
[Abstract]:The complexity and diversity of strip surface defect texture and the existence of pseudo-defects in background texture make it difficult to extract and identify the existing surface defect features of strip steel. Therefore, a new method for selecting and identifying the surface defects of steel strip is proposed. Firstly, the anisotropic diffusion algorithm is used to suppress the pseudo-defect interference on the strip surface. Secondly, the proposed P-Relief F method is used to select the surface defect characteristics, which is compared with the traditional Relief F method. The correlation between different dimension features is considered in this method. Finally, the surface defects of steel strip are classified and identified using the filtered feature set and support vector machine (SVM) kernel classifier. The experimental results show that the proposed method can extract the characteristics of strip surface defects with high discrimination and robustness, and for different types of strip surface defects, such as scratches, folds, bulges and stains, etc. Compared with the traditional method, this method can obtain higher recognition rate.
【作者单位】： 河北工业大学控制科学与工程学院;河北科技大学电气工程学院;
【基金】：国家自然科学基金(61403119) 河北省自然科学基金(F2014202166) 天津市特派员科技计划(15JCTPJC55500)资助项目
【分类号】：TG335.56;TP391.41
，

本文编号：1788368