quotPansharpening with Panchromatic and Multispectral Remote Sensing Dataquot

Abstract

Geospatial technologies have been widely used in various fields since their newlineinception. In particular, the science of remote sensing has helped analysis and newlinesolution of many Earth-related issues. Fusion of satellite data with heterogenous newlineresolutions is a pressing problem in view of the ultrahigh spectral resolution data newlinebeing made available. Development of high resolution spectral data with the help newlineof high spatial resolution panchromatic data is of practical value. Pansharpening is newlinea pixel-level fusion technique resulting in a high resolution multispectral image in newlineterms of both spatial and spectral resolution. The problem lies in maintaining the newlinespectral characteristics of each channel of the XS image when pan image is used to newlineestimate the high spatial resolution XS image. Many techniques have been newlineproposed to address the problem. A popular method involves a sensor-based newlineapproach where correlation among the XS channels and correlation between the newlinepan and spectral channels are incorporated. In this thesis, we take a wholesome newlineapproach based on the pixel values of the reflected data irrespective of the sensor newlinephysics. Three pansharpening methods are proposed based on (1) Spectral newlineConsistency (2) Convex Optimization and (3) Information theoretic approach. The proposed data-centric approach consists of building a linear regression newlinemodel between the pan and multispectral channels. A maximum likelihood solution newlineis implemented to find the regressions coefficients. Using the regression newlinecoefficients, pansharpening with spectral consistency is proposed. Apart from spectral consistency, it is imperative that spectral data distribution is also preserved newlinein pansharpening. With this in mind, combining the objectives of spectral newlineconsistency and variance matching, a convex optimization method is proposed for newlinepansharpening. Finally, the information theoretic approach based on a novel newlineapplication of orthogonal projection of pan data on to the spectral data is proposed newlinewhich is carried over from low to high resolution