Sonar Adapted Low-shot Model for Ordnance Neutralization

During this Phase 1 SBIR project for the U.S. Navy, Vadum developed Sonar Adapted Low-shot Model for Ordnance Neutralization (SALMON) – a sonar automatic target recognition (ATR) platform that uses targeted data generation to improve mine countermeasure (MCM) outcomes while reducing training data needs.

Sonar Adapted Low-shot Model for Ordnance Neutralization

The approach introduces a methodology to identify under-represented visual characteristics in a training dataset to guide a data generation platform. The enhanced training dataset is utilized to feed a sonar adapted ATR model that reduces data requirements by biases the model towards visual characteristics common to sonar imagery. With reduced training data requirements, improved model performance and real-time processing speeds, SALMON will support the warfighter’s capability to rapidly deploy MCM ATR models in dynamic field circumstances.