Automated detection of subpixel hyperspectral targets with continuous and discrete wavelet transforms

A major step toward the use of hyperspectral sensors to detect subpixel tar gets is the ability to detect constituent absorption bands within a pixel's hyperspectral curve. This paper introduces the use of multiresolution anal ysis, specifically wavelet transforms, for the automated detection of low a mplitude and overlapping constituent bands in hyperspectral curves. The wav elet approach is evaluated by incorporating it into an automated statistica l classification system, where wavelet coefficients' scalar energies are us ed as features, linear discriminant analysis is used for feature reduction, and maximum likelihood (ML) decisions are used for classification. The sys tem is tested using the leave-one-out procedure on a database of 1000 HYDIC E signals where half contain a subpixel target or additive Gaussian absorpt ion band. Test results show that the continuous and discrete wavelet transf orms are extremely powerful tools in the detection of constituent hands, ev en when the amplitude of the band is only 1% of the amplitude of the backgr ound signal.