Deep Learning for Clutter Reduction in Multi-static Coherent Active Sonar Systems

Pattern Recognition on Bayesian Estimates (PROBE) is a set of software enhancements for the Multistatic Active Coherent System (MAC) which improve clutter screening and reduce operator workload by using a convolutional deep neural network that interprets the outputs of a Likelihood Ratio Tracker (LRT). PROBE will improve MAC operations by providing earlier contact follower creation, improving track classification speed, adding additional contextual information to operator displays, calculating additional snippet scores and enhancing automatic ping control algorithms. The key innovation is the application of the neural network to a small section of the LRT state which returns multiple class probabilities for the surveyed region of the field. In this way, additional information for field-level processing and the operator review will be made available and be leveraged for performance improvements.In Phase II, SSC will begin incorporating at-sea data sets into the neural network training process; support a step 1 Advanced Processor Build evaluation of the PROBE improvements; develop real-time prototype PROBE software and display; and be available to support an at-sea test.