A wave glider-based, towed hydrophone array system for autonomous, real-time, passive acoustic marine mammal monitoring.

An autonomous surface vehicle known as a wave glider, instrumented with a low-power towed hydrophone array and embedded digital signal processor, is demonstrated as a viable low-noise system for the passive acoustic monitoring of marine mammals. Other key design elements include high spatial resolution beamforming on a 32-channel towed hydrophone array, deep array deployment depth, vertical motion isolation, and bandwidth-efficient real-time acoustic data transmission. Using at-sea data collected during a simultaneous deployment of three wave glider-based acoustic detection systems near Stellwagen Bank National Marine Sanctuary in September 2019, the capability of a low-frequency towed hydrophone array to spatially reject noise and to resolve baleen whale vocalizations from anthropogenic acoustic clutter is demonstrated. In particular, mean measured array gain of 15.3 dB at the aperture design frequency results in a post-beamformer signal-to-noise ratio that significantly exceeds that of a single hydrophone. Further, it is shown that with overlapping detections on multiple collaborating systems, precise localization of vocalizing individuals is achievable at long ranges. Last, model predictions showing a 4× detection range, or 16× area coverage, advantage of a 32-channel towed array over a single hydrophone against the North Atlantic right whale upcall are presented for the continental shelf environment south of Martha's Vineyard.

Machine learning-based classification of recreational fishing vessel kinematics from broadband striation patterns.

Machine learning is applied to the classification of underwater noise for rapid identification of surface vessel opening and closing behavior. The classification feature employed is the broadband striation pattern observed in a vessel's acoustic spectrogram measured at a nearby hydrophone. Convolutional neural networks are particularly well-suited to the recognition of textures such as interference patterns in broadband noise radiated from moving vessels. Such patterns are known to encode information related to the motion of its source. Rapid understanding of target kinematics through machine learning can provide powerful and informative cues as to the identity and behavior of a detected surface vessel.

Receiver operating characteristic for a spectrogram correlator-based humpback whale detector-classifier.

This paper presents recent experimental results and a discussion of system enhancements made to the real-time autonomous humpback whale detector-classifier algorithm first presented by Abbot et al. [J. Acoust. Soc. Am. 127, 2894-2903 (2010)]. In February 2010, a second-generation system was deployed in an experiment conducted off of leeward Kauai during which 26 h of humpback vocalizations were recorded via sonobuoy and processed in real time. These data have been analyzed along with 40 h of humpbacks-absent data collected from the same location during July-August 2009. The extensive whales-absent data set in particular has enabled the quantification of system false alarm rates and the measurement of receiver operating characteristic curves. The performance impact of three enhancements incorporated into the second-generation system are discussed, including (1) a method to eliminate redundancy in the kernel library, (2) increased use of contextual analysis, and (3) the augmentation of the training data with more recent humpback vocalizations. It will be shown that the performance of the real-time system was improved to yield a probability of correct classification of 0.93 and a probability of false alarm of 0.004 over the 66 h of independent test data.

On whether azimuthal isotropy and alongshelf translational invariance are present in low-frequency acoustic propagation along the New Jersey shelfbreak.

To understand the issues associated with the presence (or lack) of azimuthal isotropy and horizontal (along isobath) invariance of low-frequency (center frequencies of 600 Hz and 900 Hz) acoustic propagation in a shelfbreak environment, a series of experiments were conducted under the Autonomous Wide-Aperture Cluster for Surveillance component of the Shallow Water 2006 experiment. Transmission loss data reported here were from two mobile acoustic sources executing (nearly) circular tracks transmitting to sonobuoy receivers in the circle centers, and from one 12.5 km alongshelf acoustic track. The circle radii were 7.5 km. Data are from September 8, 2006. Details of the acoustic and environmental measurements are presented. Simple analytic and computer models are used to assess the variability expected due to the ocean and seabed conditions encountered. A comparison of model results and data is made, which shows preliminary consistency between the data and the models, but also points towards further work that should be undertaken specifically in enlarging the range and frequency parameter space, and in looking at integrated transmission loss.

A real-time method for autonomous passive acoustic detection-classification of humpback whales.

This paper describes a method for real-time, autonomous, joint detection-classification of humpback whale vocalizations. The approach adapts the spectrogram correlation method used by Mellinger and Clark [J. Acoust. Soc. Am. 107, 3518-3529 (2000)] for bowhead whale endnote detection to the humpback whale problem. The objective is the implementation of a system to determine the presence or absence of humpback whales with passive acoustic methods and to perform this classification with low false alarm rate in real time. Multiple correlation kernels are used due to the diversity of humpback song. The approach also takes advantage of the fact that humpbacks tend to vocalize repeatedly for extended periods of time, and identification is declared only when multiple song units are detected within a fixed time interval. Humpback whale vocalizations from Alaska, Hawaii, and Stellwagen Bank were used to train the algorithm. It was then tested on independent data obtained off Kaena Point, Hawaii in February and March of 2009. Results show that the algorithm successfully classified humpback whales autonomously in real time, with a measured probability of correct classification in excess of 74% and a measured probability of false alarm below 1%.