基于l2,p-范数回归的判别子空间聚类算法  

Discriminant subspace clustering algorithm based on l2,p-norm regression

作　　者：支晓斌 毕龙涛 ZHI Xiaobin;BI Longtao(School of Science,Xi'an University of Posts and Telecommunications,Xi'an 710121,China;School of Communication and Information Engineering,Xi'an University of Posts and Telecommunications,Xi'an 710121,China)

机构地区：[1]西安邮电大学理学院,陕西西安710121 [2]西安邮电大学通信与信息工程学院,陕西西安710121

出　　处：《西安邮电大学学报》2020年第3期77-81,共5页Journal of Xi’an University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61671377,61102095,61571361,11401045);陕西省教育厅专项科学研究计划项目(18JK0719);西安邮电大学新星团队项目(xyt2016-01)。

摘　　要：针对基于F范数的判别子空间聚类算法存在对数据适应性差的缺陷,提出一种基于l2,p-范数回归的判别子空间聚类算法。扩展最小二乘线性判别分析的目标函数到l2,p-范数意义下,利用迭代重加权最小二乘法求解目标函数。将基于l2,p-范数的线性判别分析和K-means聚类算法结合到单一的聚类框架中,从而构成广义的判别子空间聚类算法。对比实验结果表明,该算法有效地提高了判别子空间聚类对不同结构数据集的适应性。The present F-norm based discriminant subspace clustering algorithms have poor adaptability.A discriminant subspace clustering(l2,p-DSC)algorithm based on l2,p-norm regression is proposed.In this algorithm,the objective function of Linear Discriminant Analysis based on Least Square(LSLDA)is extended to the sense of l2,p-norm and solved using the iterative re-weighted least squares method.Then,the linear discriminant analysis and K-means clustering algorithm based on the l2,p-norm are combined into a single clustering framework to form a generalized DSC algorithm.Experimental results show that the proposed algorithm can improve the adaptability of discriminant subspace clustering for different structural data sets.

关 键 词：判别子空间聚类 最小二乘判别分析 l2 p-范数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]