Difference-based target detection using Mahalanobis distance and spectral angle

Two difference-based target detection methods are proposed in this work. In contrast to many target detectors which only calculate the distance between the testing pixel to the target spectrum, the proposed methods calculate the distance of the testing pixel to both of target and of background spectra. In other words, they utilize the difference between target and background computed distances. The first proposed method uses the Mahalanobis distance and benefits the valuable information contained in the statistics of targets and background. The second proposed method uses the kernel-based spectral angle mapper to benefit the advantages of spectral angle and kernel trick to separate targets from background, especially in non-linear cases. The experiments done on three real hyperspectral images indicate the high detection probability of the proposed methods compared to several target detectors.