NASA 1996 SBIR Phase I

PROJECT TITLE : : Anomaly Detection Technique for Beaconless Search and Rescue using Multi-Spectral Remote Sensing

TECHNICAL ABSTRACT (LIMIT 200 WORDS)

The power of hyperspectral imaging is based on the fact that in addition to spatial imagery, it also provides the user spectral signature of each pixel in the image. This helps in detecting subpixel sized targets based on the additive contribution of their spectral signatures to that of the background, when they are too small to be resolved spatially. Conventional "subpixel unmixing" algorithms rely on the use of spectral reference libraries for signature unmixing. However, search and rescue (S&R) operations by their very nature are completely unpredictable in terms of the terrain they must survey, the characteristics of the target to be detected, or the temporal and spatial radiometric variability encountered; robust reference libraries do not exist for every situation.

This program will develop rapid pixel anomaly detection algorithms based on spatio-spectral considerations, and lay the basis for applying multi- and hyperspectral imaging spectroscopy techniques for beaconless S&R operations. The algorithms developed herein abandon the reliance on spectral reference libraries and instead focus on detecting pixel anomalies in the terrain relative to their neighborhood pixels. Hence, the actual terrain characteristics provide the needed reference spectra, making the technique invariant to location, type of target, or spatio-temporal variability. The procedure extracts from an image very few potential pixels that need to be considered for detailed analysis. Robust criteria are established to handle trade-offs related to detection sensitivity and false alarm rate.

As timely global coverage by high resolution multi- and hyper-spectral systems becomes available, this technique will add a powerful new tool in the hands of organizations involved in S&R operations.

The algorithmic paradigm is valuable to other multi- and hyper-spectral applications under current development at OKSI. These include early detection of crop infestation via remote sensing, as well as photodiagnosis of small malignant cells in surroundings of benign cells. NAME AND ADDRESS OF PRINCIPAL Dr. Nahum Gat, Opto-Knowledge Systems, Inc. (OKSI), 1737 Third Street, Manhattan Beach, CA 90266-6308

Opto-Knowledge
Systems, Inc. (OKSI),
1737 Third Street,
Manhattan Beach, CA 90266-6308