Tools for Full Spectrum Analysis of Hyperspectral Data

Hyperspectral sensor systems are currently under consideration for applications related to monitoring the nuclear fuel cycle and other signatures of interest to the nonproliferation community. These applications include the detection of camouflaged and concealed targets, gas plume detection and identification, and terrain classification. Unfortunately, many of the data exploitation tools are not user-friendly and do not fully exploit all the spectral data available. For example, hyperspectral data from different frequency bands are analyzed separately. This project will develop technology to allow the analysis of the full spectrum, thereby increasing the probability of proper identification and reducing false alarms. During Phase I, the requirements were analyzed and several segmentation techniques ¿ Independent Component Analysis (ICA) and Non-Negative Matrix Factorization (NNMF) ¿ were implemented. In addition, the extension of the current Microcorder-rule-based spectral identification algorithm to the long wave infrared (LWIR) was investigated using data collected simultaneously by visible-near-infrared/short-wave-infrared (VNIR/SWIR) and LWIR hyperspectral sensors. In Phase II, software tools for the full spectrum analysis of hyperspectral data will be developed. Besides ICA and NNMF, other full-spectrum, compatible analysis techniques will be implemented. Specifically, the current Microcorder algorithm will be extended to the LWIR, so that rules for spectral matching can be applied over the full spectrum from VNIR to LWIR. These tools will be integrated into a user-friendly analysis package with other existing software, and delivered with the software components level for integration into DOE hyperspectral analysis packages. Commercial Applications and Other Benefits as described by the awardee: The software package should find application in the detection of contaminants, military target detection, and exploration geology. With respect to the detection of contaminants, bringing together the reflection and thermal bands would offer the possibility of discriminating against false alarm sources. For geological remote sensing, the technology could overcome the restrictions of using only the reflection-dominated portions of the spectrum, in which certain key minerals of economic importance are missed.