基于L_(0)稀疏约束的近似NMF高光谱解混

Approximate NMF hyperspectral unmixing based on Lsparsity constraint

非负矩阵分解(NMF)由于其非负性和分块表征能力,使得该算法大量的应用于机器学习和信号处理等相关领域。经典NMF与线性混合的高光谱模型比较一致,因此在高光谱解混中被广泛应用。因为传统的NMF模型对初值非常敏感,难以保证算法的收敛性。所以,通常对其加入各种稀疏性约束。本文就NMF的L_(0)约束提出了一种联合稀疏特性的近似NMF算法,它分别约束基础矩阵和系数矩阵,并将其与不受约束的NMF技术结合,诸如乘法更新规则或交替的非负最小二乘方案。最后采用真实仿真数据验证了该算法在光谱解混中相对其他算法所具有的优越性和有效性。

Non-negative matrix factorization(NMF)is widely used in hyperspectral unmixing due to its non-negativity and part representation ability, so the algorithm is largely aimed at related fields such as machine learning and signal processing.The classical NMF is more consistent with the linear mixing hyperspectral model.Therefore, it is widely used in hyperspectral unmixing.Since NMF is very sensitive to the initial value, this characteristic cannot be well guaranteed.In order to ensure that the NMF algorithm is more robust, various sparsity constraints are usually added to it.This paper proposes an approximate NMF framework for the L_(0) constraint of NMF,which constrains the fundamental matrix and coefficient matrix respectively, and combines them with unconstrained techniques NMF,such as multiplicative update rules or alternate non-negative least squares schemes.Finally, real simulation data are used to verify the superiority and effectiveness of this algorithm compared with other algorithms in spectral unmixing.