Excellence in Research: Machine Learning and Deep Learning Algorithms for Standoff Detection of Threat Chemicals by Active Infrared Backscatter Hyperspectral Imaging

The problem of standoff detection of threat chemicals such as explosives has been identified as a national security challenge in response to battlefield challenges posed in detecting improvised explosive devices. Standoff chemical detection refers to the ability to detect and identify hazardous chemicals or substances from a safe distance, without the need for direct contact or physical sampling. The standoff detection approach is based on eye-safe infrared laser interrogation coupled with infrared sensors or imaging arrays. Any standoff detection method should be safe to use around people, sensitive to relevant trace levels of analyte, and able to differentiate between the explosives analytes of interest and benign background chemicals including the substrate involved. This project provides research opportunities for three undergraduate students and one graduate student at the University of the District of Columbia. This project aims to advance the understanding, principles, and applications of a new computational framework that integrates data preprocessing, optimization, and classification techniques to distinguish illicit analytes, such as explosives, from the substrates on which they rest at standoff distances. A new data preprocessing method is designed aimed at addressing the class imbalance and unlabeled data challenges. An endmember bundle extraction approach is developed through a global optimization algorithm that utilizes a multiobjective strategy with particle swarm optimization and evolutionary algorithm. To address the speckle noise challenges, an innovative deep learning-based semantic segmentation algorithm is being developed to mitigate the speckle noises as well as enhance the feature extraction. The project will greatly benefit the artificial intelligence community and defense and security, geospatial, agriculture, and healthcare industry by providing new approaches for hyperspectral image classification and endmember bundle extraction. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.