High Performance Subpixel Hyperspectral Image Target Identification for Space-Based Sensors

Existing target identification technologies for space-based hyperspectral sensors do not attain the high detection and low false alarm rates required for effective battlefield awareness. The low and sub-pixel nature of targets in these images presents significant challenges to reliably recognize targets and separate them from surrounding clutter. We propose a phenomenologically based approach to hyperspectral image (HSI) target detection that addresses all stages of the processing chain – from optimized background characterization to phenomenologically-based subpixel detection to adaptive threshold estimation – that surpasses existing technologies in separating target from clutter. By incorporating physically meaningful constraints within the design, detection is enhanced while false alarms are suppressed. This approach is specifically designed for low spatial resolution where targets are subpixel in nature. Our approach for automated variable threshold determination is applicable to a wide variety of detection algorithms to provide enhanced performance in operational scenarios. We will demonstrate superior detection performance for our detector relative to current detection technologies. Feasibility of the proposed approach will be demonstrated in Phase I, with Phase II focusing on development and rigorous evaluation of a system prototype.