Booz Allen Hamilton

Synthetic Training Data Generation for Side-Scan Sonar

Side-scan sonar (SSS) imaging is an acoustic imaging system used in underwater mapping and exploration, search-and-recovery, and environmental monitoring. Training sonar image recognition models requires a large volume of labeled SSS images, whose acquisition involves specialized hardware, crews, and lengthy field assignments. To address this, this team presented a synthetic data-generation system for SSS imagery in partnership with Booz-Allen-Hamilton. The solution is a physics-based simulator, built on the Unity game engine and grounded in the preeminent models of acoustic imaging, producing high-quality, automatically labeled SSS imagery that can accelerate sonar image-recognition model development.