Hyperspectral Data Processing: Algorithm Design and Analysis

30.1 Introduction

One advantage provided by hyperspectral imaging is subpixel detection, which detects targets at subpixel scale. In many applications such as reconnaissance and surveillance, targets of interest may occur with low probabilities or may have relatively small size. The targets of this type are special species in agriculture and ecology, rare minerals in geology, vehicles in a large battlefield, etc. Under these circumstances, spatial-based image processing techniques may not be effective to extract these targets, particularly, when the size of targets is smaller than pixel resolution (i.e., (GSD). These targets are embedded in a single pixel vector and referred to as subpixel targets. In this case, spatial analysis-based techniques are unlikely to find these subpixel targets. We must rely on techniques that make use of their spectral characteristics to detect their presence at subpixel scale. One such technique is constrained energy minimization (CEM) developed in Chapters 2 and 12, and also in Chang (2003a). In this chapter, we investigate an interesting issue associated with subpixel detection. If a subpixel target is detected within a single pixel vector, what is its size? In order to solve this problem, we develop an approach that enables us to reliably estimate the abundance fraction of a subpixel target. Then we can multiply the obtained abundance fraction by GSD to calculate its size. Such an approach is effective only if the true abundance fraction of a subpixel ...