On the Performance of Target Detection Algorithms for Hyperspectral Imagery Analysis

摘要

Target detection is one of the most useful applications of hyperspectral remote sensing. In supervised spectral-analysis based target detection, it is assumed that the spectral signature d of a target to be detected is known a prior. In practice, the signature of a material is varied due to the weather, atmospheric, and background conditions. So it may not exactly match the signature d in a spectral library. In addition, most of pixels in a remote sensing image are mixed pixels. How a target detector handles mixed pixels and detects the target component at the subpixel level is another issue. In this paper, we will investigate the performance of five frequently used target detectors when the prior target spectral information is not precise and targets are embedded at the subpixel level. Detailed computer simulation is performed, based on which preliminary conclusions are drawn. This study is instructive to algorithm selection in practical implementation.