An Unsupervised Spectral-Spatial Feature Extraction Method for Hyperspectral Image Classification

The classification of hyperspectral images (HSI) is a wonderful mechanism for analysing mineral and oil exploration, agriculture, military and defence, and diverse land cover in remotely sensed hyperspectral images. Lately, features have been successfully extracted from hyperspectral images (HSI) using singular spectral analysis (SSA), including standard SSA in the spectral domain and 2-D SSA in the area of space. However, there are a number of important limitations, such as the sensitivity to window size, the high processing complication under a large size window, and the inability to take out joint spectral-spatial characteristics. Super pixel wise adaptive singular spectral analysis was suggested to illustrate these problems (Spa SSA). In SpaSSA, each super pixel created by an over segmented HSI receives adaptive application of both conventional and 2-D SSA. According to experimental findings on the various datasets, the suggested In terms of classification sensitivity, specificity, accuracy, and elapsed time, SpaSSA outperforms both SSA and 2D-SSA. By integrating SpaSSA with other methods, land-cover analysis can be made even more precise. Land-cover analysis can be made even more accurate by mixing SSA with the unsupervised principal component analysis (SSA + PCA).