A novel endmember extraction and discrimination algorithm for target detection inhyperspectral imagery

Abstract

Target detection is one of the most important applications of hyperspectral imagery.In this paper, an endmember extraction and discrimination algorithm (EEDA)was presented for hyperspectral target detection. Unlike most of the existingendmember extraction techniques, the EEDA takes advantage of the FastICA-generatedindependent components (ICs) that separate all potential endmember pixels inindividual components and uses the maximum spectral screening (MSS) to selectthe best representative background endmembers and thus does not require priorknowledge of the number of endmembers to be extracted. Besides, it has the ability todiscriminate the endmember of targets of interest from the interfering backgroundendmembers by identifying the IC that contains the most target information. Inorder to demonstrate the utility of the EEDA, the fully constrained least-squares(FCLS) algorithm is implemented to estimate the target abundance fractions ofthe image pixels. Experimental results on the airborne visible/infrared imagingspectrometer (AVIRIS) dataset show that the proposed EEDA in conjunction with theFCLS yields better detection performance compared with two well-known targetdetectors, an adaptive cosine estimator (ACE) and an adaptive matched filter (AMF).