Ground Viewing Perspective Hyperspectral Anomaly Detection; Final rept. for FY 2008

摘要

The U.S. Army Research Laboratory (ARL) has teamed with the Armament Research, Development and Engineering Center (ARDEC) to develop and demonstrate performance of innovative algorithmic approaches for applications requiring autonomous detection and classification of military targets (e.g., ground vehicles, camouflaged personnel) using passive hyperspectral (HS) devices. This report focuses on the first stage of a two-stage algorithm suite features autonomous clutter background characterization (ACBC), adaptive anomaly detection, and constrained subspace target classification, where the first stage highlights anomalous structures in the imagery and the second stage classifies these structures as known materials (targets) or unknown materials (targets or non-targets). The first stage has two main components, ACBC and anomaly detection. The uniqueness of this first stage is that a random sampling model is proposed as a parallel process in order to mitigate the likelihood that samples of targets are erroneously used during imagery testing as clutter- background spectral references. This approach is proposed to handling underlying difficulties (target shape/scale uncertainties) often ignored in the development of autonomous anomaly detection algorithms. Experimental results, using no prior information about the clutter background, are presented for the ACBC/anomaly detection approach testing multiple examples of real HS data cubes.