Integration of High Spectral and High Spatial Resolution Image Data for Accurate Target Detection

Abstract

This study has attempted to generate a fused image that comprises fine spectral characteristics of hyperspectral data and greater spatial details of multispectral data. N-FINDR has been applied to extract pure endmembers from hyperspectral data. The projected methodology in this study has re-sampledspectral bands of endmembers in the green, red and infra-red regions of hyperspectral data to the respective wavelength regions in multispectral data. The abundance of these endmembers in multispectral image hasbeen estimated through Linear Spectral Unmixing of multispectral data taking the re-sampled endmembers as input. New fused pixels are generated after applying Linear Model on hyperspectral data with the estimated abundances. The fused image contains spatial and spectral details of multispectral and hyperspectral data respectively enabling more accurate target identification. Normalized Cross Correlation of fused image pixels has shown more than 80% correlation with hyperspectral spectra.