Spectral-Spatial Hybrid Mechanism for Feature Detection Using Spectral Correlation

This paper discusses the way in which the contiguous nature of spectral bands was exploited to generate an additional set of features to improve the overall accuracy of feature detection in hyperspectral images. The additional set of features was generated using the correlation between the spectral bands of a neighbourhood of pixels. This incorporates spatial information into the classification algorithm. This method of incorporating spatial information of a Hyperspectral image, and the method of combining it with the spectral information in the classification algorithm could be considered as two novel concepts introduced in this paper. The proposed approach follows an unsupervised procedure and produces results with an increase of more than 5% in the accuracy level, compared to the case where only the spectral data is used.