Autonomous underwater glider based embedded real-time marine mammal detection and classification.

Abstract

Autonomous marine vehicles offer the potential to provide low-cost data suitable for passive acoustic monitoring applications of marine mammals. Due to their extremely low-power consumption and long range, gliders are an attractive option for long-term deployments. Challenges related to power availability, payload size, and weight have previously restricted the viability of marine mammal monitoring. As an example, the wide bandwidth of odontocete echolocation clicks requires a high sampling rate and poses challenges with respect to limitations in power, size, and weight of the deployed system. Recent developments in commercial off-the-shelf hardware driven by the mobile phone industry’s need for multimedia-rich smart phones have resulted in low-power architectures capable of performing computationally demanding signal processing and stochastic recognition tasks in real time. We describe our work on a small form-factor, light-weight package used to perform real-time passive acoustic detection and classification. The system detects echolocation clicks using Teager energy. Echolocation clicks are then classified using cepstral features processed by a Gaussian mixture model. [Work sponsored by ONR.]