Hyperspectral Data Processing: Algorithm Design and Analysis

30.4 Concealed Target Detection

As another example for target detection applications, this section investigates a more difficult problem, concealed target detection, which has not received much attention in the past. On many occasions, a concealed target can be camouflaged by man-made objects or shaded by natural background. The main issue in detection of concealed targets is how to remove the shadow covered on or over the target surface to expose the targets for detection. One effective means is band ratio (Robinove, 1982; Crippen, 1988), which can be used for this purpose. However, due to the fact that hyperspectral images are generally collected by hundreds of contiguous spectral bands, it is very challenging to select appropriate bands to be used for band ratio. Exhausting all possible pair combinations does not seem to be practically feasible. Accordingly, one key component in detecting concealed targets is how to select proper pairs of bands to compute their band ratio.

Band selection has been widely used for various purposes for data dimensionality, data compression, feature extraction, etc. (see Chapters 6, 19, 21–23). Many criteria for band selection have been proposed in the past to find bands that retain most of information. For instance, distance measures (Bhattacharyya distance, Jeffreys–Matusita distance), information theoretic approaches (divergence, transformed divergence, and mutual information), and eigen analysis (principal components analysis (PCA)) have been ...