default DW-CNNs: Sonar-Based Deep Learning for Mine Countermeasures Operations (Software Release and User Guide - Version 1.0)

By

DW-CNNs: Sonar-Based Deep Learning for Mine Countermeasures Operations (Software Release and User Guide - Version 1.0). Williams, David P.  CMRE-SP-2019-002. October 2019

This document acts as a stand-alone reference for using convolutional neural networks (CNNs) to perform automatic target recognition (ATR) in mine countermeasures (MCM) applications. To familiarize the nonexpert, an accessible, highlevel overview of CNNs is first provided. Then the design and training of a set of three CNNs, called DW-CNNs, is explained. These DW-CNNs were trained using synthetic aperture sonar (SAS) data collected by CMRE's MUSCLE system over a period of years in multiple geographical locations and environments. How these DWCNNs can now be exploited with sonar data collected by other similar side-looking sonar systems is explained. Accompanying this user guide is MATLAB-based DWCNN software. Those interested only in learning how to use the software can proceed directly to Sec. 4.