Hyperspectral detection and discrimination using the ACE algorithm

Abstract

One of the fundamental challenges for a hyperspectral imaging system is the detection and discrimination of subpixel objects in background clutter. The background surrounding the object, which acts as interference, provides the major obstacle to successful detection and discrimination. In many applications we look for a single signature and discrimination among different signatures is not required. However, there are important applications where we are interested for multiple signatures. In these cases, the use of spectral discrimination algorithms is both necessary and valuable. In this paper, we develop an approach to spectral discrimination based on the adaptive cosine estimation (ACE) algorithm. The basic idea is to jointly exploit the detection statistics from the various signatures and set a common threshold that ensures larger separation between signatures of interest and background. The operation of the proposed detection-discrimination approach is illustrated using real-world hyperspectral imaging data.