Effect of Spatial Enhancement on Hyperspectral Image Classification

Hyperspectral Images (HSI) are more informative for discriminating class specific objects with the help of spectral signatures. These images are collected over a range of wavelength, which are the key features for spectral classifications. Although a series of literature on exploiting spectral-spatial feature extraction exists, still the effect of enhancement on classification is unexploited so far. In the proposed framework, enhancement with/without filtration is carried out, to find the discriminating ability of samples for applications like classification. Various enhancement techniques like Histogram Equalization (HE), Adaptive Histogram Equalization(AHE) and Singular Value Equalization (SVE) based approaches are exploited and experimentally discussed. Incorporation of a simple Gaussian filter resulted in significantly better classification outcomes. Proposed filter-enhancement approaches can be termed as a spatial enhancement technique, which is more acceptable than the general enhancement techniques. For validation two datasets namely Indian Pines (IPines) and a subset of Salinas (Sal-A) dataset is experimented. Popular Support Vector Machine (SVM) classifier with standard measures like Overall Accuracy (OA), Average Accuracy (AA) and Kappa (K) are evaluated and used for comparison with various enhancement techniques. And the paper suggests for the spatial enhancement prior to classification of HSIs.