【摘要】 为了实现用较低的类间特征维数对正常人体外周血白细胞高效分类,本文提出一种基于属性层次关系的彩色白细胞图像类间特异特征选取方法。本文依据形式概念的属性约束关系,定义并权衡属性度数值,留取类间特异性较高的属性,实现层次关系分层优化及可视化,建立基于分层优化层次关系图的知识表示和发现方法。以正常人体外周血白细胞区域特征为形式背景,通过此方法选取类间特异性较高的属性,挖掘白细胞图像六分类类别特异性,将60种类间属性优化为12种,有效降低了特征维数,提高类间特征分类实效性。通过与经典实验结果比对,证明了该方法的可用性和有效性。更多还原

【Abstract】 To increase efficiency of automated leucocyte pattern recognition using lower feature dimensions,a novel inter-class distinctive feature selection method for chromatic leucocyte images was proposed based on attribute hierarchical relationship.According to the attribute constraints in formal concept analysis,we established a knowledge representation and discovery method based on the hierarchical optimal diagram by defining attribute value and visual representation of optimized hierarchical relationship.It was applied to human peripheral blood leucocytes classification and 12 distinctive attributes were simplified from 60inter-class attributes,which contributes significantly to reduced feature dimensions and efficient inter-class feature classification.Compared with the classical experimental data,the inter-class distinctive feature selection method based on hierarchical optimal diagram was proved to be usable and effective for six leucocyte pattern recognition.更多还原