Adaptive band selection for pan-sharpening of hyperspectral images

摘要

In remote sensing applications, it is important to have a dataset that has both high-spectral and high-spatial resolutions. Hyperspectral (HS) imaging achieves a very high spectral resolution, but its spatial resolution is significantly reduced due to very narrow spectral bands of the sensor. One alternative solution to increase its spatial resolution is pan-sharpening, similar to the pan-sharpening used for multispectral (MS) images. Although many pan-sharpening methods have been widely used to increase the spatial resolution of MS images, adapting such methods for HS pan-sharpening is still very challenging, because almost 90% of the HS bands do not fall within the spectral coverage of the Pan band. It has already been a challenge in MS pan-sharpening to fuse the bands that are beyond the Pan spectral coverage. It is even more challenging for HS pan-sharpening to fuse the bands that are beyond the Pan spectral range. This paper investigates the feasibility of using widely used pan-sharpening methods to fuse the HS bands that are beyond the Pan spectral range. An adaptive band selection method is proposed to help to identify the HS bands across the entire HS spectral range that can be pan-sharpened with high spectral fidelity. This is achieved based on statistical measurements and analyses between Pan and the HS bands. EO-1 satellite images are used for the experiment. Results show that using the adaptive band selection method, appropriate HS bands can be effectively selected across the entire spectral range to produce pan-sharpened HS bands with minimized spectral distortion.