Automatic Object Classification for Low-Frequency Active Sonar using Convolutional Neural Networks

Abstract

Neural Networks are proposed to classify underwater objects from active sonar system data collected for underwater surveillance. The raw signal is processed, transformed in the time-frequency domain and classified (object of interest/clutter). The values of the neural network parameters (weights and biases) are learned using data collected during two sea trials with an Echo-Repeater as an object of interest. The classifier is then validated using data from a third sea trial in different geographical locations and environmental conditions. In our validation dataset, the CNN classifier significantly reduces the number of false alarms and outperform traditional feature-based classifier that we previously developed.