Accurate models for the atmosphere are critical to the performance of signature-based hyperspectral target-detection algorithms. The problem of recovering atmospheric properties from a hyperspectral image is ill-posed in the sense that significantly different atmospheres can generate similar radiance spectra at the sensor for the same ground material. In this project we will develop a software product, ATMOD, that combines MODTRAN forward models, image-based estimates, and weather forecast products for atmospheric modeling. ATMOD will be integrated with state-of-the-art detection algorithms to allow the detection and characterization of low-spectral-contrast targets such as targets concealed by vegetation or sub-pixel gas plumes. ATMOD will consider the visible through the short-wavelength infrared and the long-wavelength infrared spectral regions. The combined modeling and detection algorithms are designed to run in real-time on COTS hardware. We will characterize ATMOD using a large amount of hyperspectral imagery and a large amount of forecast data. We will also consider the transition of ATMOD for a range of applications.
Keywords: Space-Based, Hyperspectral, Remote Sensing, Sub-Pixel Detection, Forecast, Atmospheric Characterization, Clutter Modeling, Adaptive
