The remote sensing-based inversion of sound speed profile (SSP) enables the acquisition of high-spatial-resolution SSP without in situ measurements. The spatial division of the inversion grid is crucial for the accuracy of results, determining both the number of samples and the consistency of inversion relationships. The result of our research is the introduction of a physics-inspired self-organizing map (PISOM) that facilitates SSP inversion by clustering samples according to the physical perturbation law. The linear physical relationship between sea surface parameters and the SSP drives dimensionality reduction for the SOM, resulting in the clustering of samples exhibiting similar disturbance laws. Subsequently, samples within each cluster are generalized to construct the topology of the solution space for SSP reconstruction. The PISOM method significantly improves accuracy compared with the SOM method without clustering. The PISOM has an SSP reconstruction error of less than 2 m/s in 25% of cases, while the SOM method has none. The transmission loss calculation also shows promising results, with an error of only 0.5 dB at 30 km, 5.5 dB smaller than that of the SOM method. A physical interpretation of the neural network processing confirms that physics-inspired clustering can bring better precision gains than the previous spatial grid.

Coastal acoustic tomography (CAT) is a remote sensing technique that utilizes acoustic methodologies to measure the dynamic characteristics of the ocean in expansive marine domains. This approach leverages the speed of sound propagation to derive vital ocean parameters such as temperature and salinity by inversely estimating the acoustic ray speed during its traversal through the aquatic medium. Concurrently, analyzing the speed of different acoustic waves in their round-trip propagation enables the inverse estimation of dynamic hydrographic features, including flow velocity and directional attributes. An accurate forecasting of inversion answers in CAT rapidly contributes to a comprehensive analysis of the evolving ocean environment and its inherent characteristics. Graph neural network (GNN) is a new network architecture with strong spatial modeling and extraordinary generalization. We proposed a novel method: employing GraphSAGE to predict inversion answers in OAT, using experimental datasets collected at the Huangcai Reservoir for prediction. The results show an average error 0.01% for sound speed prediction and 0.29% for temperature predictions along each station pairwise. This adequately fulfills the real-time and exigent requirements for practical deployment.

Helmholtz resonator(HR) has already been demonstrated both theoretically and experimentally to be a metamaterial with negative mass density and negative bulk modulus simultaneously. The HR can resonate at a frequency corresponding to a wavelength much longer than its geometrical parameters. At this time, the incident acoustic energy can be located. Therefore, the HR structures are considered to be good choices for controlling low-frequency sound waves. Furthermore, existing results indicate that the wide forbidden band could be formed by a one-dimensional structure shunted with detuned HRs. Based on these aforementioned theories, a man-made acoustical structure with broadband low-frequency sound insulation effect is designed by circularly inbuilt HRs. Beyond this structure's surface, a two-dimensional quiet zone can be created. With the same simulated model, an experimental structure is fabricated based on PVC plastic material. The structure consists of five layerd circular plates. In the top four plates, two kinds of holes are drilled. The smaller holes in the top plate act as shot necks of the HR, while the bigger holes in the middle three plates serve as the cavities of the HR. They can construct 60 resonators with different resonant frequencies. Experiments are carried out to study its sound insulation properties. In the experiments, three kinds of HRs with resonant frequencies 785, 840 and 890 Hz from inner loop to outer loop, respectively, are formed. The experimental results are very coincident with the simulation results from the software of COMSOL Multiphysics based on finite element method, which shows that this structure has an excellent sound insulation effect in a frequency band of 680-1050 Hz, and the maximum insulation sound pressure level can reach 41 dB. Meanwhile, the distribution of the two-dimensional sound field is measured. The results point out that the range of the insulation area can be changed with the incident frequency. In addition, the sound insulation effect is sensitive to the resonant state of the HRs. When all of the resonators at the same loop resonate simultaneously, the insulation sound pressure level will be higher. On the contrary, the insulation sound pressure level will be lower because of the energy leaking through the positions where the HRs do not resonate with the others. This work will be of help for designing new sound protection devices for low-frequency sound waves.

The sonar image has limitations on the physical spatial resolution due to system configuration and underwater environment, which often leads to challenges for underwater targets detection. Here, the deep learning method is applied to enhance the physical spatial resolution of underwater sonar images. Specifically, the U-shaped end-to-end neural network which contains down-sampling and up-sampling parts is proposed to improve the physical spatial resolution limited by the array aperture. The single target and multiple cases are considered separately. In both cases, the normalized loss on the testing sets declines rapidly, and the predicted high-resolution images own great agreement with the ground truth eventually. Further improvements in resolution are focused on, that is, compressing the predicted high-resolution image to its physical spatial resolution limitation. The results show that the trained end-to-end neural network could map high resolution targets to the impulse responses at the same location and amplitude with an uncertain target number. The proposed convolutional neural network approach could give a practical alternative to improve the physical spatial resolution of underwater sonar images.

Acoustic waves are essential tools for guiding underwater activities. For many years, numerical modeling of ocean acoustic propagation has been a major research focus in underwater acoustics. Normal mode theory, one of the earliest and most extensively studied methods in this field, is renowned for its well-established theoretical framework. The core of normal mode theory involves the numerical solution of modal equations. In classical normal mode models, these equations are typically discretized using low-order finite difference methods, which, while broadly applicable, suffer from a limited convergence rate. The spectral element method, widely used in the seismic field, is recognized for its spectral precision and flexibility. In this article, we propose a normal mode model discretized using the spectral element method. The weak form of the modal equation directly satisfies boundary and interface conditions without requiring additional operations. The entire computational domain can be divided into segments of varying number and length, configured according to environmental conditions. The perfectly matched layer technique is employed to simulate acoustic half-space boundary conditions, effectively addressing the high computational costs and numerical instability associated with traditional artificial absorbing layers. Based on these algorithms, we have developed a numerical program (SEM). This research verifies the accuracy of the spectral element model through three different types of numerical experiments.

Reliable ship-radiated-noise recognition depends on both model design and evaluation protocol. On DeepShip, the common recording-level split still places training-seen ship identities in the test set; we define this overlap as a contamination rate of $68.1\%$. A ship-level protocol reduces contamination to 0 under our metadata-based identity grouping, and a controlled comparison with the same architecture and hyperparameters shows a $7.11$-percentage-point (pp) accuracy difference. We also refine the constant-Q harmonic coefficient (CQHC) by formulating envelope–harmonic separation as non-negative constrained least squares and replacing the original one-step approximation with non-negative harmonic projection and regularized closed-form back-substitution. The three inputs, whose vertical axes represent linear frequency, Mel bands, and harmonic order, are encoded separately and fused at an intermediate level by soft mixture-of-experts gating. Bayesian hybrid attention provides posterior-based predictive uncertainty. Under the DeepShip recording-level protocol, accuracy is $88.64\%$, exceeding input stacking and decision averaging by $3.83$ and $2.50$ pp. The expected calibration error decreases from $0.0724$ to $0.0413$, and rejection at $90\%$ coverage reduces the error by $27.0\%$. At the recording level, confidence-weighted voting reaches $94.17\%$ over 120 test recordings. Without dataset-specific tuning, ShipsEar accuracy is $98.63\%$ with a segment-random split and $95.05\%$ with recording-disjoint evaluation.