Characterization of terrain background in LWIR hyperspectral data to aid landmine detection

Spectral target recognition algorithms function by distinguishing specified target spectra from those of the environment in which they are embedded. Typically this discrimination is done by comparing a pixel´s spectral signature to those in a known library or by looking for spectral outliers within a single image. A high probability of detection results when the target and background signature distributions are well known and significantly separated. While much work has gone into defining target signature distributions, significantly less effort has gone into defining background distributions, hi fact, improvements in background definition results in lower false alarm rates and higher probability of detection.