Multiscale 2-D Singular Spectrum Analysis and Principal Component Analysis for Spatial–Spectral Noise-Robust Feature Extraction and Classification of Hyperspectral Images

In hyperspectral images (HSI), most feature extraction and data classification methods rely on corrected dataset, in which the noisy water absorption bands are removed. This can result not only extra working burden but also information loss from removed bands. To tackle these issues, this article, we propose a novel spatial-spectral framework, multiscale 2-D singular spectrum analysis (2-D-SSA) with principal component (PCA) (2-D-MSSP), for noise-robust of HSI. First, 2-D-SSA is applied to exploit spatial features each spectral band HSI via extracting varying trends within defined windows. Taking extracted trend signals at scale level as input features, PCA employed domain dimensionality reduction extraction. The derived separately classified then fused decision-level efficacy. As our 2-D-MSSP method extract simultaneously remove noise both domains, ensures it be HSI, even uncorrected dataset. Experiments three publicly available datasets have fully validated efficacy robustness proposed approach, when benchmarked 10 state-of-the-art classifiers, including six four deep learning classifiers. addition, quantitative qualitative assessment has approach limited training samples, especially classifying without filtering