基于L_(1/2)范数约束增量非负矩阵分解的SAR目标识别  

SAR target recognition via incremental nonnegative matrix factorization with L_(1/2) constraint

作　　者：张慧[1] 党思航 崔宗勇 

机构地区：[1]电子科技大学成都学院电子工程系,成都611731 [2]电子科技大学电子工程学院,成都611731

出　　处：《计算机应用研究》2018年第2期628-631,共4页Application Research of Computers

基　　金：四川省教育厅科研资助项目(16ZB0446)

摘　　要：增量非负矩阵分解(INMF)随目标样本增加逐渐更新分解模型,能够有效解决NMF算法的计算代价随样本增加而成倍增长的问题。然而INMF在使NMF具备增量学习能力的同时,并未考虑NMF分解矩阵的稀疏性对识别性能的提升作用。针对上述问题,提出基于L1/2范数约束的增量非负矩阵分解(L1/2-INMF)算法,并应用于SAR目标识别。L1/2-INMF采用L1/2范数实时约束增量过程中的NMF分解矩阵,能够在不增加计算复杂度的同时,提升识别性能。针对MSTAR数据集的仿真实验结果表明,提出的L1/2-INMF能够解决传统非负矩阵分解方法计算代价随样本增加而增加的问题。As target sample increasing,incremental nonnegative matrix factorization （INMF） gradually updates the decomposition model, which can effectively solve the problem that the NMF algorithm increases the computational cost. However, INMF doesn＇ t consider that decomposition matrices on NMF can improve recognition performance when it changes NMF to have ability of incremental learning. To solve the above problem, this paper proposed incremental nonnegative matrix factorization with L1/2 constraint （L1/2-INMF） ,and applied it in SAR target recognition. L1/2-INMF took the decomposition matrices under L1/2 real time constraint in the incremental process, which could improve recognition performance without increase of computational complexity. According to the simulation results on MSTAR data sets, the proposed L1/2-INMF can solve the problem of traditional NMF method increasing the computational cost as sample increasing, and obtain better recognition rate than INMF.

关 键 词：增量非负矩阵分解 合成孔径雷达 目标识别 L1/2范数约束 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]