Underwater acoustic target recognition is a hot research area in acoustic signal processing. With the development of deep learning, feature extraction and neural network computation have become two major steps of recognition. Due to the complexity of the marine environment, traditional feature extraction cannot express the characteristics of the targets well. In this paper, we propose an underwater acoustic target recognition approach named VFR. VFR adopts a novel feature extraction method by fusing three-dimensional FBank features, and inputs the extracted features into a residual network, instead of the classical CNN network, plus cross-domain pre-training to perform target recognition. The experimental results show that VFR achieves 98.5% recognition accuracy on the randomly divided ShipsEar dataset and 93.8% on the time-divided dataset, respectively, which are better than state-of-the-art results.

The accuracy and efficiency of sound field calculations highly concern issues of hydroacoustics. Recently, one-dimensional spectral methods have shown high-precision characteristics when solving the sound field but can solve only simplified models of underwater acoustic propagation, thus their application range is small. Therefore, it is necessary to directly calculate the two-dimensional Helmholtz equation of ocean acoustic propagation. Here, we use the Chebyshev–Galerkin and Chebyshev collocation methods to solve the two-dimensional Helmholtz model equation. Then, the Chebyshev collocation method is used to model ocean acoustic propagation because, unlike the Galerkin method, the collocation method does not need stringent boundary conditions. Compared with the mature Kraken program, the Chebyshev collocation method exhibits a higher numerical accuracy. However, the shortcoming of the collocation method is that the computational efficiency cannot satisfy the requirements of real-time applications due to the large number of calculations. Then, we implemented the parallel code of the collocation method, which could effectively improve calculation effectiveness.

Inverting seawater sound speed profiles (SSPs) using Bayesian methods enables optimal parameter estimation and provides a quantitative assessment of uncertainty by analyzing the posterior distribution of target parameters. However, in nonlinear geophysical inversion problems like acoustic tomography, calculating the posterior distribution remains challenging. In this study, a Bayesian framework is used to construct the posterior distribution of target parameters based on acoustic travel-time data and prior information. A Hamiltonian Monte Carlo (HMC) approach is developed for SSP inversion, offering an effective solution to the computational issues associated with complex posterior distributions. The HMC algorithm has a strong physical basis in exploring distributions, allowing for accurate characterization of physical correlations among target parameters. It also achieves sufficient sampling of heavy-tailed probabilities, enabling a thorough analysis of the target distribution characteristics and overcoming the low efficiency often seen in traditional methods. The SSP dataset was created using temperature–salinity profile data from the Hybrid Coordinate Ocean Model (HYCOM) and empirical formulas for SSP. Experiments with acoustic propagation time data from the Kuroshio Extension System Study (KESS) confirmed the feasibility of the HMC method in SSP inversion.

Magnetic anomaly detection (MAD) is a core technology for the detection of ferromagnetic targets, yet traditional methods such as the Orthogonal basis function (OBF) detector suffer from severe performance degradation in low signal-to-noise ratio (SNR) and Gaussian colored noise environments. To address this issue, this paper proposes a novel neural network architecture based on manual feature extraction, namely the Partitioned Classification Network with 42 features (PCN-42). First, a total of 42 magnetic anomaly features belonging to three categories, i.e., time–frequency features, statistical features, and magnetic moment features, are extracted from magnetic measurement data. Subsequently, each category of features is processed independently by parallel sub-networks followed by feature fusion. Finally, the performance of the proposed method for magnetic anomaly signal detection is verified under colored noise simulation conditions combined with measured geomagnetic noise. In the experiments, the target is modeled as a magnetic dipole with moment 200 $\mathrm{A}\cdot {\mathrm{m}}^2$, closest point of approach (CPA) ranges from 300 to 800 m, and noise conditions span Gaussian colored noise ($\alpha$ = 0.5, 0.8, 1.0) plus measured geomagnetic background. Simulation results demonstrate that compared with the OBF detector, the detection probability of the PCN-42 detector is improved by 45–75 percentage points, reaching over 80% in different magnetic moment directions and approximately 90% at CPA = 450 m. Experiments with measured noise further validate the effectiveness of the proposed method. This method significantly improves the detection probability of magnetic anomaly signals in colored noise environments. MAD serves a broad range of civilian purposes, including mineral exploration, buried pipeline monitoring, archeological prospection, and humanitarian unexploded ordnance (UXO) clearance.

This article presents a semantic tracker which simultaneously tracks a single target and recognises its category. In general, it is hard to design a tracking model suitable for all object categories, e.g., a rigid tracker for a car is not suitable for a deformable gymnast. Category-based trackers usually achieve superior tracking performance for the objects of that specific category, but have difficulties being generalised. Therefore, we propose a novel unified robust tracking framework which explicitly encodes both generic features and category-based features. The tracker consists of a shared convolutional network (NetS), which feeds into two parallel networks, NetC for classification and NetT for tracking. NetS is pre-trained on ImageNet to serve as a generic feature extractor across the different object categories for NetC and NetT. NetC utilises those features within fully connected layers to classify the object category. NetT has multiple branches, corresponding to multiple categories, to distinguish the tracked object from the background. Since each branch in NetT is trained by the videos of a specific category or groups of similar categories, NetT encodes category-based features for tracking. During online tracking, NetC and NetT jointly determine the target regions with the right category and foreground labels for target estimation. To improve the robustness and precision, NetC and NetT inter-supervise each other and trigger network adaptation when their outputs are ambiguous for the same image regions (i.e., when the category label contradicts the foreground/background classification). We have compared the performance of our tracker to other state-of-the-art trackers on a large-scale tracking benchmark (100 sequences)---the obtained results demonstrate the effectiveness of our proposed tracker as it outperformed other 38 state-of-the-art tracking algorithms.

The ocean sound speed profile (SSP) governs underwater acoustic propagation but remains difficult to observe continuously along a moving survey track using direct profiling instruments. This paper investigates differential acoustic propagation delays as remote observations that integrate propagation information along the acoustic paths for SSP retrieval along a survey track. A single vessel transmits coded signals, and a towed array records direct arrivals and arrivals reflected from the seabed. Their differential delays, together with water depth, are inverted using a physics-guided delay-denoising artificial neural network (PDANN), in which the SSP is represented by empirical orthogonal functions and constrained through a differentiable acoustic propagation model. Under 10 ms Gaussian delay perturbations, PDANN achieved an SSP RMSE of 0.611 m/s compared with 1.210 m/s for AETNN and 2.367 m/s for an ANN trained on clean data. Adding the Physical Module reduced the SSP RMSE from 0.759 to 0.611 m/s and the forward prediction RMSE for acoustic delays from 2.020 to 0.998 ms. Independent field evaluation using available CTD and XBT references yielded SSP errors of approximately 0.65 to 1.4 m/s. Sequential inversions further produced a spatially continuous representation of sound speed along the track. These results support active acoustic remote sensing as a practical approach for rapid SSP retrieval along a survey track under the tested configuration using a single vessel.