Spectral-spatial stacked autoencoders based on low-rank and sparse matrix decomposition for hyperspectral anomaly detection

Abstract

Nowadays, some algorithms based on deep learning have drawn increasing attention in hyperspectral image (HSI) analysis. In this paper, we propose spectral-spatial stacked autoencoders based on low-rank and sparse matrix decomposition (LRaSMD-SSSAE) for hyperspectral anomaly detection (AD). First, the Go Decomposition (GoDec) algorithm is employed to solve the low-rank background component and the sparse anomaly component. Second, stacked autoencoders (SAE) are employed on the sparse matrix for spectral deep features and on the low-rank matrix for spatial deep features, respectively. Finally, the spectral-spatial feature matrix is established and local Mahalanobis-distance algorithm is employed for the final detection result. Experiments are carried out on real and synthetic HSI, and the results show that the proposed LRaSMD-SSSAE generally outperforms the comparison algorithms.