Ocean sound speed is important for underwater acoustic applications, such as communications, navigation and localization, where the assumption of uniformly distributed sound speed profiles (SSPs) is generally used and greatly degrades the performance of underwater acoustic systems. The acquisition of SSPs is necessary for the corrections of the sound ray propagation paths. However, the inversion of SSPs is challenging due to the intricate relations of interrelated physical ocean elements and suffers from the high costs of calculations and hardware deployments. This paper proposes a novel sound speed inversion method based on multi-source ocean remote sensing observations and machine learning, which adapts to large-scale sea regions. Firstly, the datasets of SSPs are generated utilizing the Argo thermohaline profiles and the empirical formulas of the sound speed. Then, the SSPs are analyzed utilizing the empirical orthogonal functions (EOFs) to reduce the dimensions of the feature space as well as the computational load. Considering the nonlinear regression relations of SSPs and the observed datasets, a general framework for sound speed inversion is formulated, which combines the designed machine learning models with the reduced-dimensional feature representations, multi-source ocean remote sensing observations and water temperature data. After being well trained, the proposed machine learning models realize the accurate inversion of the targeted ocean region by inputting the real-time ocean environmental data. The experiments verify the advantages of the proposed method in terms of the accuracy and effectiveness compared with conventional methods.

Introduction: Various types of ships sail at sea, and identifying maritime ship types through shipradiated noise is one of the tasks of ocean observation. The ocean environment is complex and changeable, such rapid environmental changes underline the difficulties of obtaining a huge amount of samples. Meanwhile, the length of each sample has a decisive influence on the classification results, but there is no universal sampling length selection standard.Methods: This study proposes an effective framework for ship-radiated noise classification. The framework includes: i) A comprehensive judgment method based on multiple features for sample length selecting. ii) One-dimensional deep convolution generative adversarial network (1-DDCGAN) model to augment the training datasets for small sample problem. iii) One-dimensional convolution neural network (CNN) trained by generated data and real data for ship-radiated noise classification. On this basis, a onedimensional residual network (ResNet) is designed to improve classification accuracy.Results: Experiments are performed to verify the proposed framework using public datasets. After data augmentation, statistical parameters are used to measure the similarity between the original samples and the generated samples. Then, the generated samples are integrated into the training set. The convergence speed of the network is clearly accelerated, and the classification accuracy is significantly improved in the one-dimensional CNN and ResNet.Discussion: In this study, we propose an effective framework for the lack of scientific sample length selection and lack of sample number in the classification of ship-radiated noise, but there aret still some problems: high complexity, structural redundancy, poor adaptability, and so on. They are also long-standing problems in this field that needs to be solved urgently.

The performance of the underwater acoustic (UWA) orthogonal frequency division multiplexing (OFDM) system is often restrained by time-varying channels with large delays. The existing frequency domain equalizers do not work well because of the high complexity and difficulty of finding the real-time signal-to-noise ratio. To solve these problems, we propose a low-complexity neural network (NN)-based scheme for joint equalization and detection. A simple NN structure is built to yield the detected symbols with the joint input of the segmented channel response and received symbol. The coherence bandwidth is investigated to find the optimal hyperparameters. By being completely trained offline with real channels, the proposed detector is applied independently in both simulations and sea trials. The results show that the proposed detector outperforms the ZF and MMSE equalizers and extreme learning machine (ELM)-based detectors in both the strongly reflected channels of the pool and time-variant channels of the shallow sea. The complexity of the proposed network is lower than the MMSE and ELM-based receiver.

Ultrasonic Non-Destructive Testing (UT) technology serves as a fundamental quality control tool in critical industrial sectors including aerospace and petrochemical industries. However, defect data acquired during practical inspection processes suffer from limited sample sizes, class imbalance, and high annotation costs, which severely hinder the development and performance optimization of deep learning-based intelligent detection algorithms. To address this challenge, this paper proposes the Physics-Constrained Denoising Diffusion Probabilistic Model (PC-DDPM) for high-fidelity ultrasonic defect signal synthesis. The proposed method achieves significant improvements in generation quality through three key aspects. A physics-parameter conditional injection module based on a cross-attention mechanism is designed to incorporate physical parameters such as amplitude, frequency, and attenuation coefficient into the diffusion process. In addition, a multi-dimensional physics-constrained loss function, including energy conservation, frequency rationality, and time–frequency consistency, is constructed to ensure that generated signals conform to fundamental physical laws of ultrasonic wave propagation. Furthermore, an adaptive denoising scheduling strategy based on signal complexity is introduced to dynamically adjust diffusion step sizes, balancing generation quality and computational efficiency. Experimental results demonstrate that PC-DDPM achieves a 63.7% reduction in time-domain MSE, a spectral similarity of 0.957, and a time-frequency correlation of 0.943 compared to existing baselines. Augmenting training data with generated signals raises four-class defect classification accuracy to 95.0%, validating practical effectiveness in real inspection scenarios.