Seabed type and source parameters predictions using ship spectrograms in convolutional neural networks.

Broadband spectrograms from surface ships are employed in convolutional neural networks (CNNs) to predict the seabed type, ship speed, and closest point of approach (CPA) range. Three CNN architectures of differing size and depth are trained on different representations of the spectrograms. Multitask learning is employed; the seabed type prediction comes from classification, and the ship speed and CPA range are estimated via regression. Due to the lack of labeled field data, the CNNs are trained on synthetic data generated using measured sound speed profiles, four seabed types, and a random distribution of source parameters. Additional synthetic datasets are used to evaluate the ability of the trained CNNs to interpolate and extrapolate source parameters. The trained models are then applied to a measured data sample from the 2017 Seabed Characterization Experiment (SBCEX 2017). While the largest network provides slightly more accurate predictions on tests with synthetic data, the smallest network generalized better to the measured data sample. With regard to the input data type, complex pressure spectral values gave the most accurate and consistent results for the ship speed and CPA predictions with the smallest network, whereas using absolute values of the pressure provided more accurate results compared to the expected seabed types.

Seabed classification from merchant ship-radiated noise using a physics-based ensemble of deep learning algorithms.

Merchant ship-radiated noise, recorded on a single receiver in the 360-1100 Hz frequency band over 20 min, is employed for seabed classification using an ensemble of deep learning (DL) algorithms. Five different convolutional neural network architectures and one residual neural network are trained on synthetic data generated using 34 seabed types, which span from soft-muddy to hard-sandy environments. The accuracy of all of the networks using fivefold cross-validation was above 97%. Furthermore, the impact of the sound speed and water depth mismatch on the predictions is evaluated using five simulated test cases, where the deeper and more complex architectures proved to be more robust against this variability. In addition, to assess the generalizability performance of the ensemble DL, the networks were tested on data measured on three vertical line arrays in the Seabed Characterization Experiment in 2017, where 94% of the predictions indicated that mud over sand environments inferred in previous geoacoustic inversions for the same area were the most likely sediments. This work presents evidence that the ensemble of DL algorithms has learned how the signature of the sediments is encoded in the ship-radiated noise, providing a unified classification result when tested on data collected at-sea.

Seabed and range estimation of impulsive time series using a convolutional neural network.

In ocean acoustics, many types of optimizations have been employed to locate acoustic sources and estimate the properties of the seabed. How these tasks can take advantage of recent advances in deep learning remains as open questions, especially due to the lack of labeled field data. In this work, a Convolutional Neural Network (CNN) is used to find seabed type and source range simultaneously from 1 s pressure time series from impulsive sounds. Simulated data are used to train the CNN before application to signals from a single hydrophone signal during the 2017 Seabed Characterization Experiment. The training data includes four seabeds representing deep mud, mud over sand, sandy silt, and sand, and a wide range of source parameters. When applied to measured data, the trained CNN predicts expected seabed types and obtains ranges within 0.5 km when the source-receiver range is greater than 5 km, showing the potential for such algorithms to address these problems.

The Airy phase of explosive sounds in shallow water.

The Airy phase is identified in the received signals from explosive charges deployed in a shallow water acoustic experiment conducted in the New England Mudpatch region during the spring of 2017. Measured and modeled time-frequency dispersion curves are compared and a geoacoustic sensitivity study utilizing marginal probability distributions for the sound speed in five sediment layers is performed. The analysis suggests that inclusion of the Airy phase frequency and arrival time in a geoacoustic-inversion method could lower the uncertainty of sound speed parameter estimation in a multi-layer sediment as compared to methods that do not include the Airy phase structure.

Geoacoustic inversion using low frequency broadband acoustic measurements from L-shaped arrays in the Shallow Water 2006 Experiment.

When using geoacoustic inversion methods, one objective function may not result in a unique solution of the inversion problem because of the ambiguity among the unknown parameters. This paper utilizes acoustic normal mode dispersion curves, mode shapes, and modal-based longitudinal horizontal coherence to define a three-objective optimization problem for geoacoustic parameter estimation. This inversion scheme is applied to long-range combustive sound source data obtained from L-shaped arrays deployed on the New Jersey continental shelf in the summer of 2006. Based on the sub-bottom layering structure from the Compressed High-Intensity Radiated Pulse reflection survey at the experimental site, a two-layer (sand ridge overlaying a half-space basement) range-independent sediment model is utilized. The ambiguities of the sound speed, density, and depth of the sand ridge layer are partially removed by minimizing these objective functions. The inverted seabed sound speed over a frequency range of 15-170 Hz is comparable to the ones from direct measurements and other inversion methods in the same general area. The inverted seabed attenuation shows a nonlinear frequency dependence expressed as αb=0.26f1.55(dB/m) from 50 to 500 Hz or αb=0.32f1.65(dB/m) from 50 to 250 Hz, where f is in kHz.

Statistical inference of seabed sound-speed structure in the Gulf of Oman Basin.

Addressed is the statistical inference of the sound-speed depth profile of a thick soft seabed from broadband sound propagation data recorded in the Gulf of Oman Basin in 1977. The acoustic data are in the form of time series signals recorded on a sparse vertical line array and generated by explosive sources deployed along a 280 km track. The acoustic data offer a unique opportunity to study a deep-water bottom-limited thickly sedimented environment because of the large number of time series measurements, very low seabed attenuation, and auxiliary measurements. A maximum entropy method is employed to obtain a conditional posterior probability distribution (PPD) for the sound-speed ratio and the near-surface sound-speed gradient. The multiple data samples allow for a determination of the average error constraint value required to uniquely specify the PPD for each data sample. Two complicating features of the statistical inference study are addressed: (1) the need to develop an error function that can both utilize the measured multipath arrival structure and mitigate the effects of data errors and (2) the effect of small bathymetric slopes on the structure of the bottom interacting arrivals.

Evidence of three-dimensional acoustic propagation in the Catoche Tongue.

This paper presents observations of two classes of acoustic arrivals recorded on a sparsely populated vertical line array (VLA) moored in the center of the Catoche Tongue, a major reentrant in the Campeche Bank in the southeastern Gulf of Mexico. The acoustic signals were generated by signals underwater sound (SUS) located 50-80 km from the VLA. The first class of arrivals was identified as resulting from a direct (non-horizontally refracted) path. Then following a quiescent period, a second, more diffuse class of arrivals is observed and is believed to be the result of horizontal refraction from the margin of the Tongue. A spectral analysis of the measured data revealed that both classes of arrivals were characterized by the source spectrum associated with SUS. Additionally, the difference in time between the onset of the first and second class of arrivals observed as a function of range from the VLA is consistent with the relative difference in the length of the direct and refracted paths. The observations are further supported by a three-dimensional (3D) acoustic propagation computation that reproduces many of the features of the measured data and provides additional insight into the details of the 3D propagation.

An impulsive source with variable output and stable bandwidth for underwater acoustic experiments.

The Combustive Sound Source (CSS) is being developed as an environmentally friendly source to be used in ocean acoustics research and surveys. It has the ability to maintain the same wide bandwidth signal over a 20 dB drop in source level. The CSS consists of a submersible combustion chamber filled with a fuel/oxidizer mixture. The mixture is ignited and the ensuing combustion and bubble activity radiates an impulsive, thus broadband, acoustic pulse. The ability to control pulse amplitude while maintaining bandwidth is demonstrated.

Acoustic ducting by shelf water streamers at the New England shelfbreak.

Greater sound speed variability has been observed at the New England shelfbreak due to a greater influence from the Gulf Stream with increased meander amplitudes and frequency of Warm Core Ring (WCR) generation. Consequently, underwater sound propagation in the area also becomes more variable. This paper presents field observations of an acoustic near-surface ducting condition induced by shelf water streamers that are related to WCRs. The field observations also reveal the subsequent disappearance of the streamer duct due to the passage of a WCR filament. These two water column conditions are investigated with sound propagation measurements and numerical simulations.

Scattering in a Pekeris waveguide from a rough bottom using a two-way coupled mode approach.

Scattering from a rough ocean bottom is described numerically with a two-way coupled-mode formalism that contains scattering effects to all orders and provides an exact solution to the wave equation. Both scattered field and direct blast components are computed within the formalism framework. A comparison of the scattered component solution from the coupled mode with the Born approximation (BA) solution for scattering from a rough bottom Pekeris waveguide shows that the BA predicts correctly the scattered field levels but not detailed structure. The transition from direct blast to scattered field dominance is identified in the total field time series.

Validating deep learning seabed classification via acoustic similarity.

While seabed characterization methods have often focused on estimating individual sediment parameters, deep learning suggests a class-based approach focusing on the overall acoustic effect. A deep learning classifier-trained on 1D synthetic waveforms from underwater explosive sources-can distinguish 13 seabed classes. These classes are distinct according to a proposed metric of acoustic similarity. When tested on seabeds not used in training, the classifier obtains 96% accuracy for matching such a seabed to one of the top-3 most acoustically similar classes from the 13 training seabeds. This approach quantifies the performance of a seabed classifier in the face of real seabed variability.

Multichannel array diagnosis using noise cross-correlation.

A practical application of noise cross-correlation for the diagnosis of a multichannel ocean hydrophone array is derived. Acoustic data were recorded on a horizontal line array on the New Jersey Shelf while Tropical Storm Ernesto passed through. Results obtained from active source measurements reveal that signals from several hydrophones, which were recorded on certain channels before the storm, are recorded on different channels after the storm. Noise cross-correlation of data recorded during the storm show when, and in what manner, these changes took place.

Geoacoustic inversion with ships as sources.

Estimation of geoacoustic parameters using acoustic data from a surface ship was performed for a shallow water region in the Gulf of Mexico. The data were recorded from hydrophones in a bottom mounted, horizontal line array (HLA). The techniques developed to produce the geoacoustic inversion are described, and an efficient method for geoacoustic inversion with broadband beam cross-spectral data is demonstrated. The performance of cost functions that involve coherent or incoherent sums over frequency and one or multiple time segments is discussed. Successful inversions for the first sediment layer sound speed and thickness and some of the parameters for the deeper layers were obtained with the surface ship at nominal ranges of 20, 30, or 50 water depths. The data for these inversions were beam cross-spectra from four subapertures of the HLA spanning a little more than two water depths. The subaperture beams included ten frequencies equally spaced in the 120-200 Hz band. The values of the geoacoustic parameters from the inversions are validated by comparisons with geophysical observations and with the parameter values from previous inversions by other invesigators, and by comparing transmission loss (TL) measured in the experiment with modeled TL based on the inverted geoacoustic parameters.