| 刘坤,赵帅帅,屈尔庆,周颖.R-AdaBoost带钢表面缺陷特征选择算法[J].电子测量与仪器学报,2017,31(1):9-14 |
| R-AdaBoost带钢表面缺陷特征选择算法 |
| R AdaBoost strip surface defect feature selection algorithm |
| |
| DOI：10.13382/j.jemi.2017.01.002 |
| 中文关键词: AdaBoost算法 Relief特征筛选 特征选择 缺陷检测 |
| 英文关键词:AdaBoost algorithm relief feature selection feature selection defect detection |
| 基金项目:国家自然科学基金（61403119）、河北省自然科学基金（F2014202166）资助项目 |
|
|
| Author | Institution |
| Liu Kun | School of Control Science and Engineering, Hebei University of Technology, Tianjin 300130, China |
| Zhao Shuaishuai | School of Control Science and Engineering, Hebei University of Technology, Tianjin 300130, China |
| Qu Erqing | School of Control Science and Engineering, Hebei University of Technology, Tianjin 300130, China |
| Zhou Ying | School of Control Science and Engineering, Hebei University of Technology, Tianjin 300130, China |
|
| 摘要点击次数: 11555 |
| 全文下载次数: 47600 |
| 中文摘要: |
| 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 AdaBoost特征选择算法。该算法在AdaBoost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AdaBoost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AdaBoost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。 |
| 英文摘要: |
| The complex and various defects of the steel surface bring great difficulty to the feature extraction and selection. Therefore, this paper proposes a new R AdaBoost future selection method with a fusion of feature selection and sample weights updated. The proposed algorithm selects features and reduces the dimension of features via Relief feature selection according to updated samples in each cyle of AdaBoost algorithm, and uses reduced features to remove noise samples by intra class difference among samples, and then update sample library according to dynamic weight of AdaBoost. The weak classifiers are trained by the resulting optimal features, and combined to generate the final AdaBoost strong classifier, and detect and locate strip surface defects by AdaBoost two classifiers. Aiming at a variety of defects such as scratch, wrinkle, mountain, stain, etc. in the actual strip production line, the experimental results show that the proposed R AdaBoost algorithm can effectively extract features with high distinction and independence and reduce the feature dimension, and simultaneously improve the accuracy of defect detection. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
