R-AdaBoost带钢表面缺陷特征选择算法

发布时间：2018-01-16 20:04

本文关键词：R-AdaBoost带钢表面缺陷特征选择算法　出处：《电子测量与仪器学报》2017年01期 　论文类型：期刊论文

【摘要】：带钢表面缺陷形式的复杂多变给特征的选择带来了困难,为此,提出一种融合特征筛选和样本权值更新的R-Ada Boost特征选择算法。该算法在Ada Boost算法的每个循环中通过Relief算法进行特征的筛选与降维,通过筛选后的特征利用样本的类内类间差去除噪声样本,然后根据Ada Boost的动态权值更新样本库,再利用每个循环优化选择得到的最优特征与弱分类器级联成最终的Ada Boost强分类器,进行带钢表面缺陷的检测与定位。实验结果表明,针对带钢实际生产线上的划痕、褶皱、山脉、污点等多种缺陷,该算法可以有效提取出具有高区分性和独立性的特征,同时提高了缺陷检测算法的准确率。
[Abstract]:It is difficult to select the characteristics because of the complexity and variety of the surface defect form of steel strip. A R-Ada Boost feature selection algorithm based on feature selection and sample weight updating is proposed in Ada. In each cycle of Boost algorithm, Relief algorithm is used to filter features and reduce dimension. The noise samples are removed by the inter-class difference of the samples through the selected features, and the sample library is updated according to the dynamic weights of Ada Boost. Then the optimal feature and weak classifier are cascaded into the final Ada Boost strong classifier to detect and locate the surface defects of steel strip. The experimental results show that. Aiming at the defects such as scratches, folds, mountains, stains and so on in the actual production line of strip steel, the algorithm can effectively extract the features with high distinguishing and independence, and improve the accuracy of the defect detection algorithm at the same time.
【作者单位】：河北工业大学控制科学与工程学院;
【基金】：国家自然科学基金(61403119) 河北省自然科学基金(F2014202166)资助项目
【分类号】：TG142.15;TP391.41
【正文快照】： 1引言由于带钢表面缺陷形成原因的不同,导致缺陷呈现形式存在着多样性、随机性和复杂性等特点[1],现有的缺陷检测方法通过提取缺陷的空域特征和变换域特征[2-3]并将多种类型的特征进行组合以满足多样和复杂缺陷形式的要求,但过高的特征维度难以满足算法实时性的要求。因此,提

【相似文献】