Joint Beamforming and Automated Target Recognition for 3D SAS
Suren Jayasuriya | Arizona State University
The technical objective for this project is to develop a suite of algorithms for enhanced beamforming and automated target recognition (ATR) for unexploded ordinance in cluttered, occluded, and buried environments underwater. The main technology of interest is three-dimensional synthetic aperture sonar imaging. In particular, physics-based knowledge including scattering models, target modeling, and environmental characterization will be merged with state-of-the-art machine learning-based ATR algorithms, to enhance their performance while providing interpretability to these detectors/classifiers. This project will accomplish new research insights, algorithms, and highlight the feasibility of the technical approach for further research in the future, including testing in simulation, laboratory data, and real sensor data from existing SERDP projects.
The technical approach leverages the paradigm of differentiable programming where machine learning methods can be written alongside physics-based models, and thus complement the relative advantages of each approach. The project team will utilize differential programming to jointly write beamforming and ATR as an end-to-end pipeline to optimize target detection and classification, specifically for buried munitions. This pipeline will be augmented by the physics of underwater acoustics, particularly for objects buried in sediment, to enhance the performance. This objective will be accomplished over a one-year time period to assess the feasibility of this approach, utilizing simulation, lab data, and real sensor data to mitigate risk and assess transition potential including how it can be scaled up and deployed in future efforts for in-the-field operation.
Expected benefits to the Department of Defense and the scientific community include new knowledge of how to merge the physics of underwater acoustics and beamforming with machine learning. New algorithms and methodologies will be developed, as well as proof-of-concept experimental demonstrations. Results will be communicated via regular reporting and scientific/technical papers and presentations.