Data analysis for classification of UXO filler using pulsed neutron techniques

Irradiating substances with pulsed neutrons results in several types of interactions which cause the emission of gamma rays. The energy of these gamma rays is characteristic of the nuclei with which the reaction occurred, and can therefore be used as an indicator of the presence of an atomic species. The PELAN system uses a pulsed neutron generator, which makes it possible to separate the gamma spectra into inelastic and capture components that are easier to interpret. Historically, the analysis of PELAN data has been based on a least squares method to extract the contribution of different elemental species present in the sample. The approach uses measured response functions for each element of interest, followed by decision rules for the identification of the materials. We have investigated an alternative approach that does not require a model and response functions. Instead, the approach determines features directly from a number of spectra of substances of interest, e.g. explosives and hazardous chemicals. The PCA method has been used to obtain indicators from the spectra. These indicators are then used for detection and identification of substances using the GLRT algorithm. The performance of the data analysis is assessed through ROC curves. A comparison of the two approaches indicates that PCA followed by GLRT technique has better performance and is more robust than the previous approach.

Duke Scholars

Author Leslie M. Collins Electrical and Computer Engineering

Author Stacy L. Tantum Electrical and Computer Engineering