稀疏约束的L21增量式非负矩阵分解研究  

Research on L21 Incremental Non-Negative Matrix Factorization with Sparsity Constraints

作　　者：杨亮东 赵妍杰 潘正红 YANG Liangdong;ZHAO Yanjie;PAN Zhenghong(Basic Teaching Department,Lanzhou Resources&Environment Voc-Tech University,Lanzhou,Gansu Province,730021 China;School of Public Health,Gansu University of Chinese Medicine,Lanzhou,Gansu Province,730022 China)

机构地区：[1]兰州资源环境职业技术大学基础教学部,甘肃兰州730021 [2]甘肃中医药大学公共卫生学院,甘肃兰州730022

出　　处：《科技资讯》2024年第12期240-244,共5页Science & Technology Information

基　　金：兰州资源环境职业技术大学校级科技项目(项目编号:X2023A-05)。

摘　　要：针对新增数据增大而引起的运算效率增大的现象,提出了一种稀疏约束的增量式非负矩阵分解改进算法。该算法是在加入稀疏条件的情况下对增量数据使用L21范数。首先对初始数据进行经典非负矩阵分解,其次再利用其分解结果参与增量数据的运算,使目标函数在分解计算中具有较好的收敛效果和分解后数据有较好的稀疏度。实验部分主要是将该算法与增量式非负矩阵分解、稀疏约束的增量式非负矩阵分解、经典非负矩阵分解算法进行对比,得出在分解后数据的稀疏度和收敛快慢方面该算法均优于其他3个算法。In order to solve the phenomenon of the increasing opertion efficiency caused by increasing new data,an improved algorithm of incremental non-negative matrix factorization with sparsity constraints is proposed,and the algorithm uses the L21 norm for incremental data with the addition a sparse condition.Firstly,classical non-negative matrix factorization is performed on initial data,and then its factorization results are used to participate in the operation of the incremental data,so that the objective function has a better convergence effect in the calcula-tion of factorization and a better sparsity of the data after factorization.In the experiment,the proposed algorithm is compared with incremental non-negative matrix factorization,incremental non-negative matrix factorization with sparsity constraints and classical non-negative matrix factorization algorithms,and it is concluded that the proposed algorithm is better than the other three algorithms in terms of the sparsity and convergence speed of the data after factorization.

关 键 词：非负矩阵分解 增量式学习 图像识别 稀疏约束 L21范数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]