摘要
在对样本数据进行降维时,子空间学习模型无法揭示数据结构和处理训练样本外的新样本。提出一种融合表示学习和嵌入子空间学习的降维方法。将低秩表示、加权稀疏表示和低维子空间学习构建到一个统一的框架中,并采用交替优化策略,实现数据表示系数矩阵和数据投影矩阵的同时学习和相互优化,最终达到重建效果最优的降维精度。在3个数据库上的实验结果表明,与PCA、NPE、LRPP等主流方法相比,该方法不仅可以解决无法训练新样本的问题,而且具有较优的分类性能。
When reducing the dimension of sample data,the subspace learning model cannot reveal the data structure or process new samples apart from the training samples.This paper proposes a dimension reduction method that fuses representation learning and embedded subspace learning.The method constructs a unified framework that integrates low-rank representation,weighted sparse representation and low-dimensional subspace learning.Then the alternative optimization strategy is adopted to realize the simultaneous learning and mutual optimization of the coefficient matrix and the data projection matrix,finally achieving the dimension reduction precision that enables optimal reconstruction effect.The experimental results on three databases show that,compared with the mainstream methods such as PCA,NPE and LRPP,this method can solve the problem of new sample training,and has better classification performance.
作者
陶洋
鲍灵浪
胡昊
TAO Yang;BAO Linglang;HU Hao(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
出处
《计算机工程》
CAS
CSCD
北大核心
2021年第6期83-87,97,共6页
Computer Engineering
基金
重庆市自然科学基金(cstc2018jcyjAX0344)。
关键词
低秩表示
稀疏表示
降维
联合学习
图像分类
low rank representation
sparse representation
dimensionality reduction
joint learning
image classification