Simultaneous retrieval of subpixel proportions and signatures from two-component mixture

Abstract

Subpixel decomposition is of high interest in various applications. The most existing approaches to invert either proportions or signatures usually require another quantity to be known. In this paper, we develop two algorithms to simutaneously retrieve both signatures and proportions without any priori knowledge of both quantities. Singular Value Decomposition (SVD) algorithm is first applied to convert the original data matrix into three parts, then a set of constraints are imposed on them, and two algorithms incorporating constrained nonlinear least squares (NLS) and ordinary least squares (OLS) in SAS package are applied to find realistic rolutions. Simulations designed for evaluations of our algorithms indicate that they are quite effective, and practical solutions still can be achieved even if the signal-to-noise ratio (SNR) decreases up to 1.5:1. Finally a real remotely sensed imagery of forest scane. Advanced Solid-State Array Spectroradiometer (ASAS) data, is implemented in our experiments.