Explosive Forensic Technology: Hyperspectral Real-Time Threat Anomaly Detection (Hyper Thread)
Abstract
Hyperspectral line scanners provide a wealth of data, from which information can be derived and potential threats can be realized. However, real-time analysis of this data is difficult due to the sheer volume of data that must be processed; therefore, this data has traditionally been post-processed. We used statistical representations of the incoming data by looking at higher-order statistics (skewness and kurtosis) and information theory (entropy) to provide probability distribution function-specific data for each of the incoming spectra, thereby reducing the computational burden. In this work from fiscal year 2020-2021, we show that our statistical representations of the data can be used for anomaly detection. We did this through collection of data, treatment of experimental and simulated spectra, ground-truth development for statistical analysis, and an analysis into the use of pretreatment with our data. Furthermore, we determined that implementation of our algorithm using semi-supervised machine learning results in real-time analysis (100 ms frame rate, 250 spectra per frame) of the hyperspectral data we obtain. This algorithm can be implemented in a scenario when immediate situational awareness is necessary, thereby increasing Warfighter lethality.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2022
- Accession Number
- AD1173642
Entities
People
- Darren K. Emge
- Eric R. Languirand
- Justin M. Curtiss