Spatial and Spectral Quality Assessment of Fused Hyperspectral and Multispectral Data

Abstract

Hyperspectral sensors collect images in hundreds of narrow, continuous spectral channels, whereas multispectral sensors collect images in relatively broader wavelength bands. The spatial resolution of a hyperspectral image is, however, lower than that of a multispectral image. This study has integrated the high spectral and high spatial information of hyperspectral (Hyperion) and multispectral (LISS-IV) imagery of the Henry Island of Sundarbans, India. An integrated image has been successfully generated that has 5 m spatial resolution of LISS-IV image and 10 nm spectral resolution of Hyperion image. The prime objective of this study is to obtain an image of enhanced spectral and spatial resolution that would facilitate accurate interpretation and mapping of mangrove types, sea and creek water, pisciculture water, agricultural soil and saline soil of the study area. The methodology adopted considers band remapping of different spectral regions covered by multispectral and hyperspectral images. This study has applied algorithms to restore spatial information of hyperspectral data by integrating hyperspectral bands with those multispectral bands that fall within its range. After comparison of the spectral profile of fused and original hyperspectral image, similarity is found in the spectral curves across the electronic spectrum indicating that the spectral signature of the fused image is maintained. High values of crosscorrelation between the images show that both the spectral and the spatial information are well preserved. The quality evaluation of the fused image is based on quantitative criteria which include spatial and spectral properties that are defined in the image. This study assesses the quality of fused image through utilization of various statistical indicators. Spectral quality of integrated image is assessed using parameters such as spectral discrepancy, correlation coefficient, values of root mean square error (RMSE), Spectral Angle Mapper (SAM), standard deviation and signal-to-noise ratio (SNR) are used for spectral quality assessment of the fused image. Spatial quality assessment has been done using bias and Edge Detection techniques. After analysis of application of these indicators on the fused and the original hyperspectral image, it is observed that the spectral discrepancy shows low discrepancy values in the visible and near-infrared (NIR) part of the electromagnetic spectrum. The correlation coefficient has given good results in the NIR region as compared to visible region. The results obtained from RMSE show that its values in the visible region are lower than that in the NIR region. SAM performs well in the NIR region as compared to the visible region. The standard deviation of fused and hyperspectral images is very similar and their differences are either close to zero or negative. The values of SNR calculated between fused and hyperspectral images vary randomly with the spectral wavelength. The random variability of the signal over the image (i.e. noise) is very less, thus giving a good SNR value. The result obtained from bias shows a good-quality fused and hyperspectral image with the ideal bias values varying with the spectral wavelength of the fused image. The results thus obtained after application of the indices show that the spectral and spatial property of fused image is close to its ideal value. Neural network has been used to assign classes to pixels of the new integrated image.