邻域粗糙集在轮对踏面缺陷图像特征选择的应用

邻域粗糙集理论能够直接处理数值性数据,无需离散化,已经被应用在属性选择和分类器设计中;文中在分析了轮对踏面缺陷图像的成像特征的基础上,首先从缺陷样本中提取了16个特征值,然后基于邻域粗糙集知识约简的思想,提出了一种轮对踏面缺陷图像特征的选择方法,实现了对原始特征的优化选择,利用径向基函数神经网络和选择的特征值对缺陷进行分类;实验结果表明,该方法可将识别的原始特征下降到原始特征数目的20%左右,擦伤识别率提高了68%。

Application of Neighborhood Rough Sets in Features Selection of Wheelsets Tread Defect Images

The neighborhood rough sets theory can directly deal with numerical attributes,thus the theory has been widely used in domain of attribute reduction and classifier design.In the paper,based on the characteristics of wheelsets tread defects image,sixteen features are extracted from defect samples,on the basis of knowledge reduction of theory in the neighborhood rough sets,a method of selecting wheelsets tread defect image features is proposed,the optimal selection of original features is realized.Defects are classified by radial basis function(RBF) network and features selected.The experimental results show that the method can reduce the 80% redundant features and the recognition rate of tread surface scrapes increases from 30% to 98%.