Deep-Sonar: Spatial-Temporal Deep Learning for Sonar Systems

Air-deployed wide-area search sonar systems are widely adopted for military, scientific research, and civilian services. In the MAC sonar system, while MIMO increases data rate relative to SIMO, the clutter effect also increases and transfers from one mission environment to another. To address the critical need, Intelligent Automation Inc. (IAI) proposes the design and implementation of the Deep-Sonar system, which provides an unsupervised/semi-supervised spatial-temporal deep learning solution for the MAC sonar system. The Deep-Sonar system will provide a high-accuracy solution on analyzing MIMO sonar data in a flexible setting to identify targets in different (shallow or deep water) environments. We will prototype the proposed algorithmic workflow of Deep-Sonar and test it with representative scenarios in IAIs Ontology-based, Distributed, Composable Simulator that has been developed for Navy. We will demonstrate the feasibility of Deep-Sonar by evaluating its performance under realistic underwater scenarios with respect to different Tx/Rx configurations and clutter levels. Deep-Sonar will reduce the clutter from each ping by at least 25% by application of semi-supervised and unsupervised spatial-temporal deep learning to MIMO sonar data that will reduce the number of clutter events for 4 transmitters to 1 transmitter level.