Deep learning (DL) approaches in underwater acoustic localization (UAL) have gained a great deal of popularity. While numerous works are devoted to improving the localization precision, they neglect another critical challenge inherent in the DL-based UAL problem, i.e., the model's practicality. Advanced DL models generally exhibit extremely high complexity, requiring a large amount of computational resources and resulting in slow inference time. Unfortunately, the limited processing power and real-time demands in oceanic applications make the deployment of complex DL models exceedingly challenging. To address this challenge, this article proposes a lightweight UAL framework based on knowledge distillation (KD) techniques, which effectively reduces the size of a deep UAL model while maintaining competitive performance. Specifically, a dedicated teacher network is designed using attention mechanisms and convolutional neural networks (CNNs). Then, the KD is performed to distill the knowledge from the teacher network into a lightweight student model, such as a three-layer CNN. In practical deployment, only the lightweight student model will be utilized. With the proposed lightweight framework, the student model has 98.68% fewer model parameters and is 87.4% faster in inference time compared to the teacher network, while the prediction accuracy drops to only 1.07% (97.55% $\rightarrow$ 96.48%). In addition, the generalization ability of the student model is examined through transfer learning, where the model is transferred between two different ocean environments. The student model demonstrates a stronger generalization ability compared to the model without the KD process, as it can quickly adapt itself to a new application environment using just 10% of the data.

Deep-learning source localization methods trained on synthetic replicas calculated by acoustic propagation models (referred to as model-based source localization methods via deep learning) suffer from environmental mismatches between replicas and measurements, which degrade the localization performance. This article introduces deep transfer learning (DTL), specifically the unsupervised domain adaptation (UDA) and fine-tuning methods for source localization using ships of opportunity (SOO) to alleviate the influence of mismatched replicas. First, the range estimation network is trained using simulated replicas and unlabeled SOO data. Subsequently, the parameters of the ranging network are transferred to a depth estimation network and fine-tuned accordingly. Simulation results demonstrate that the proposed method (UDA-DTL) exhibits greater robustness to array tilt and sound speed profile mismatches in the direct arrival zone compared to the Bartlett processor and the ResNet estimator. Experimental data from the South China Sea also verified the validity of the proposed UDA-DTL estimator. The results demonstrate that supplementing model-based supervised learning with unlabeled SOO, the impact of unknown uncertainty on the localization performance is alleviated. This study highlights the potential to utilize unlabeled Big Data in applications of underwater source localization.

Estimating the line spectra of ship-radiated noise is a crucial remote sensing technique for detecting and recognizing underwater acoustic targets. Improving the signal-to-noise ratio (SNR) makes the low-frequency components of the target signal more prominent. This enhancement aids in the detection of underwater acoustic signals using sonar. Based on the characteristics of low-frequency narrow-band line spectra signals in underwater target radiated noise, we propose a dual-stream deep learning network with frequency characteristics transformation (DS_FCTNet) for line spectra estimation. The dual streams predict amplitude and phase masks separately and use an information exchange module to swap learn features between the amplitude and phase spectra, aiding in better phase information reconstruction and signal denoising. Additionally, a frequency characteristics transformation module is employed to extract convolutional features between channels, obtaining global correlations of the amplitude spectrum and enhancing the ability to learn target signal features. Through experimental analysis on ShipsEar, a dataset of underwater acoustic signals by hydrophones deployed in shallow water, the effectiveness and rationality of different modules within DS_FCTNet are verified.Under low SNR conditions and with unknown ship types, the proposed DS_FCTNet model exhibits the best line spectrum enhancement compared to methods such as SEGAN and DPT_FSNet. Specifically, SDR and SSNR are improved by 14.77 dB and 13.58 dB, respectively, enabling the detection of weaker target signals and laying the foundation for target localization and recognition applications.

Direction of arrival (DOA) estimation methods for wideband signals based on sparse Bayesian learning (SBL) have shown impressive performance. It generally assumes that wideband signals share the same frequency band, whereas deteriorating the performance of DOA estimation when target signals locate at different bands. To deal with this problem, we propose an off-grid DOA estimation framework, namely FBC-SBL, based on SBL incorporating frequency band clustering (FBC). FBC can exploit the cluster structure of signal distribution in frequency domain by applying a specific clustering algorithm. Compared to other existing algorithms, FBC-SBL can infer the frequency band occupation thereby estimating DOAs more accurately. Specifically, the whole frequency band is firstly divided into finer subbands and the variance vector of each subband is estimated. Then, FBC gathers the subbands with similar signal distribution. Finally, the estimated DOAs can be obtained by analyzing each cluster through SBL framework. Numerical simulations and real data experiments show that our method has superior performance compared to other algorithms, and achieves optimal average absolute error of 1.6394 degree in real data experiments.

A complex-valued neural process method, combined with modal depth functions (MDFs) of the ocean waveguide, is proposed to reconstruct the acoustic field. Neural networks are used to describe complex Gaussian processes, modeling the distribution of the acoustic field at different depths. The network parameters are optimized through a meta-learning strategy, preventing overfitting under small sample conditions (sample size equals the number of array elements) and mitigating the slow reconstruction speed of Gaussian processes (GPs), while denoising and interpolating sparsely distributed acoustic field data, generating dense field data for virtual receiver arrays. The predicted field is then integrated with the matched field processing (MFP) method for passive source localization. Validation on the SWellEx-96 waveguide shows significant improvements in localization performance and reduces sidelobes of ambiguity surface compared to traditional MFP and GP-based MFP. Moreover, the proposed kernel based on MDFs outperforms the Gaussian kernel in describing ocean waveguide characteristics. Because of the feature representation of multi-modal mapping, this kernel enhances acoustic field prediction performance and improves the accuracy and robustness of MFP. Simulated and real data are used to verify the validity.