Adaptive multichannel discrete wavelet transforms for automated subpixel target detection

This paper investigates the use of adaptive multichannel discrete wavelet transforms (AMDWT) for automated subpixel target detection. The detection system utilizes supervised training in which the system adapts the design of the multichannel wavelet filters (MWFs) for optimum detection of subpixel targets in hyperspectral curves. For this study, the subpixel targets are Gaussian absorption bands, where a specified mean and variance of a band represents a given constituent material. When the system is tested, the optimum MWFs are used to decompose the hyperspectral curves, and wavelet coefficient energy features are extracted. Classification is performed using maximum-likelihood decision boundaries. The experimental results show that the AMDWT is very promising for automated detection of especially low amplitude subpixel targets