NGA University Research Initiative (NURI) - Hyperspectral Algorithm Development

Sponsor(s): National Geospatial-Intelligence Agency (NGA)

Research Team: David Messinger, Natalie Sinisgalli (B.S. student), Harvey Rhody, Bill Hoagland, Rolando Raqueño, Paul Lee

Project Scope: This program, being conducted in conjunction with the Laboratory for Imaging Algorithms and Systems (LIAS) within the Center for Imaging Science, focuses on the development and implementation of physics-based target detection algorithms for use with hyperspectral imagery. The algorithms are implemented into prototype-level software within the IDL/ENVI environment for delivery back to NGA. Three tools were identified for development: subpixel target detection, detection of concealed and contaminated targets, and detection of gaseous effluents.

Project Status: This program will be finalized during 2006, with final software delivery to NGA during the fall / winter. Focus during the past year has been on completing the implementation as well as on testing of the third tool of the set. The algorithm for detection of gaseous effluents in longwave infrared hyperspectral imagery is based on a forward-modeling approach to characterize the spectral signatures of target gas species thought to be in the scene.

Figure 3-1: a) Color Infrared image of the area of interest for testing of the gaseous effluent detection tool. b) Thermal infrared image of same area of facility. c) Full scene detection results for ethan gas. d) Zoom showing a small ethane plume detection. e) Full scene detection results for methane gas. f) Zooms showing two small methane plume detections.

This approach models the signatures in both emission and absorption and over a wide range of concentrations providing a more robust detection scheme. Atmospheric information is included in the forward modeling process removing the need to accurately compensate the imagery for these effects. The gaseous effluent tool was tested against airborne longwave hyperspectral imagery collected by the Advanced Hyperspectral Imager (AHI) built and operated by the University of Hawaii . The scene is of an industrial complex and the experiment was conducted by the Environmental Protection Agency. Figure 3-1 shows a color IR image of part of the facility, a single thermal band image of the same area, and detection results from the physics-based detection scheme. The vertical "stripes" in the detection images are a result of a calibration error in the sensor for those pixels. Detection results are shown for both ethane and methane known to be released in this region of the facility. As can be seen in the figure, three small plumes are detected with very high confidence at various positions in the facility.

Final delivery of the software, including a unique approach to look up table generation and storage, will be accomplished in the winter of 2006.