Chemical detection techniques using mid-wave infrared or long-wave infrared hyper-spectral imagery (HSI), or ultra-spectral point sensors, generate massive data sets that complicate the detection and analysis of chemical plumes. In addition, HSI data sets usually require a long time (days to months) to pull useful information from a scene. In order to benefit non-proliferation activities, new signal processing algorithms are required to detect and classify target chemicals quickly in these massive data sets. This project will develop a two-tier digital signal processing algorithm that can provide substantial improvements in the identification and quantification of chemical compounds from effluent plumes, using hyper-spectral and ultra-spectral data. The first tier uses a principal components analysis (PCA) algorithm to transform and compress the hyper-spectral cube to a set of eigen-images in order to reduce spectral redundancy. The second tier looks down the spectral axis of the uncompressed hyper-cube, and uses a wavelet transform to de-noise and present features to a trained, neural net-based feature extractor, which identifies the chemical constituents of the effluent plume. Phase I will design and develop a trained PCA compression algorithm, a non-trained PCA compression algorithm, and fractal/texture-based image processing algorithms capable of recognizing vapor plumes. A wavelet-based feature extractor and neural net-based classifier for the analysis and classification of target spectra will also be developed. Finally, comprehensive testing of the efficacy/necessity of the two PCA compression algorithms will be conducted.
Commercial Applications and Other Benefits as described by the awardee: The hyper-spectral processing procedure should be applicable to a multitude of fields, such as medicine and biology, where the signal to noise ratio is low.
