SBIR-STTR Award

The deployment of hyperspectral and higher-resolution panchromatic imagers on the same platform provides the opportunity to use detailed spectral and spatial models for target detection. In this project, we will address this opportunity by developing a system INV-SE that uses combined spectral/spatial information to advance the operational utility of target detection algorithms. A new spectral detection algorithm that includes scene-based environmental constraints will be combined with an innovative spatial detection algorithm that is invariant to target pose and partial obscuration. We will also develop new methods for acquiring target spectral and spatial models from airborne imagery. A Bayesian technique will be used to rank detection results by integrating spectral and spatial cues. The algorithms will be derived to allow real-time performance on tactically relevant hardware. The new approach will be assessed over a range of data during Phase I. We also provide a detailed strategy for the commercialization of INV-SE for a range of applications.

The deployment of hyperspectral and higher-resolution panchromatic imagers on the same platform provides the opportunity to use detailed spectral and spatial models for target detection. In this project, we address this opportunity by developing a system INV-SE that uses combined spectral/spatial information to advance the operational utility of target detection algorithms. INV-SE consists of a physics-based spectral detection algorithm in combination with an innovative spatial detection algorithm to provide a detection capability that is invariant to the illumination, atmospheric conditions, target pose, and viewing geometry. Target spectral and spatial models can be supplied by the user or recovered from airborne imagery using new algorithms. If specific target spatial models are unavailable, INV-SE will include support for the detection of generic classes of manmade objects. The algorithms have been derived for use with calibrated or uncalibrated spectral sensors. A Bayesian technique is used to rank detection results by integrating the spectral and spatial cues. INV-SE will be assessed using a large set of spectral and panchromatic image data that has been acquired over a wide range of conditions. The Phase II project will result in a real-time INV-SE implementation on tactically relevant hardware. We will also consider the transition of INV-SE for a range of applications.

Keywords:
Hyperspectral, Remote Sensing, Target Detection, Information Fusion, Spatial Processing, Spectral/Spatial, Real-Time, Model Acquisition