An Intelligent Optimization, Clustering and Classification Framework for Large Scale Photo-Thermal Infrared Imaging Spectroscopy (PT-IRIS) Big Data
Abstract
This project aims to investigate the feasibility of innovative photo-thermal infrared imaging spectroscopy (PT-IRIS) data analysis for remote explosive detection applications. The problem of standoff detection of threat chemicals such as explosives and their precursors has been identified as a DoD-wide challenge in response to the improvised explosive device (IED) problem. Detecting chemical threats at standoff distances by optical approaches offers several advantages. First is the obvious separation of the user from the threat. Second is the potential to interrogate more and larger areas than could possibly be probed using traditional contact-based methods such as swabbing, wanding or sniffing. Third is the potential to interrogate subjects without disrupting the normal flows of people, packages, luggage, mail, vehicles, etc. The Naval Research LaboratoryÕs (NRL) approach to standoff detection is based on eye-safe infrared laser interrogation coupled with infrared sensors or imaging arrays. This approach allows for the exploitation of well-known characteristic infrared absorption features that comprise spectroscopic signatures by which the unknown chemicals can be identified based on similarity to threat library spectra. Advantages of this approach include standoff distance, detection sensitivity and specificity, speed, and eye safety. In order to distinguish illicit analytes, such as explosives, from the substrates on which they rest at standoff distances, our proposed research seeks to develop novel clustering and classification algorithms to mitigate the curse of dimensionality, minimize the overlap across clusters from different classes, and intensify the underrepresented class concepts for improved overall performance and less computational time. A new class of supervised learning method - extended nearest neighbor (ENN) method with various kernel designs on classification will be developed. In addition, inspired by the context-sensitive clustering method where the context class (i.e. backgrounds or confounders) can serve as a boundary within the object class (i.e. targets to be recognized), the PIs will propose a novel semi-supervised based kernel density clustering algorithm to generate non-overlapping sub-clusters. Moreover, the hybrid particle swarm optimization (PSO) and evolutional algorithm (EA) approach investigated by the PI will be used to develop feature selection algorithms for high dimensional feature space in which an intelligent mechanism is required to identify the minimum features that contain the most discriminative information of distinct classes.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 20, 2019
- Source ID
- W911NF1810475
Entities
People
- Nian Zhang
Organizations
- Army Contracting Command
- Office of the Secretary of Defense
- University of the District of Columbia