Target detection in hyperspectral imagery using forward modeling and in-scene information

This work addresses the problem of detecting and classifying materials and targets in hyperspectral images based on their reflectance spectrum. Accurate target detection in hyperspectral imagery requires a radiative transfer model that maps between the spectral reflectance domain and the measured radiance domain. Such a model can be employed in two ways for detection - using atmospheric compensation, where the measured hyperspectral radiance image is converted to a reflectance image, and using forward modeling, where the target reflectance spectrum is converted to an at-sensor target radiance spectrum. This work presents a forward modeling detection method that utilizes in-scene information to estimate the parameters in the radiative transfer model. Uncertainty in the radiative transfer model and variability of the target spectra are captured using a constrained subspace model for the target. Target detection using library spectra and target rediscovery are evaluated in hyperspectral images of a complex urban scene.