Library Partitioning for Target Detection in Joint Contaminated Surface Detector

Target detection from a set of observations, where the target is from a set of given library spectra is a difficult task. A major reason for false positives or misclassifications in detection is canonical correlations with linear combinations of spectra, a serious problem especially for subspace-based detectors using canonical correlations or least squares. We propose a partitioning scheme for the library spectra by posing the problem as a vertex coloring problem in graph theory to enhance the target detection performance. Simulation results demonstrate its effectiveness for several promising approaches for target detection in Joint Contaminated Surface Detector.