Approximate sparse regularized multilayer NMF for hyperspectral unmixing

The selection of sparse regularization functions directly affects the effect of sparse non-negative matrix factorization of hyperspectral unmixing. At present, the L0 or L1 norms are mainly used as sparse measures. L0 has good sparsity, but it is difficult to solve; L1 is easy to solve, but the sparsity is poor. An approximate sparse model was presented, and was applied to the multi-layer NMF (AL0-MLNMF) in hyperspectral unmixing. The algorithm made the observation matrix multilevel sparse decomposition improve the precision of hyperspectral unmixing, and improve the convergence of the algorithm. The simulation data and real data show that the algorithm can avoid falling into the local extremum and improve the NMF hyperspectral unmixing performance. Algorithm accuracy has greater improvement effect than several other algorithm, RMSE reduces 0.001-1.676 7 and SAD reduces 0.002-0.244 3.