Interference-structure-based passive source localization provides computational efficiency and physical interpret-ability but suffers performance degradation for weak acoustic targets,where low signal-to-noise ratios(SNRs)obscure the observable interference patterns necessary for reliable depth estimation.Narrowband time-window limitations and environ-mental mismatches further introduce blind zones and nonlinear distortions between the source depth and the measured interference structure.To mitigate these effects,this study proposes a hybrid depth estimation method that integrates arrival-angle interference features with a deep learning(DL)framework based on a residual network.Sound intensity in the beam domain is employed to exploit multidimensional and nonlinear relationships in the acoustic field,thereby enhancing robustness under weak target and low SNR conditions.Simulation and sea experiment results demonstrate that the proposed method achieves improved depth estimation accuracy compared with conventional interference-based techniques.The findings indicate that the hybrid interference-DL approach effectively extends the applicability of interference-structure-based localization to weak acoustic targets.

This paper proposed DLR (Dual-Label-Reversed) Ensemble Learning Strategy, a universal underwater acoustic target detection strategy with a transfer learning architecture, which can ensemble two transferred target detection models to enhance the detection accuracy. An acoustic data feature extraction strategy is employed to extract comprehensive features ranging from time/frequency domain to dedicated auditory parameters. A target detection model transfer strategy is proposed to get original transferred model and DLR transferred model from source domain to target domain. Then, the final detection can be made by the DLR ensemble learning strategy, which ensemble the output of two transferred model. We evaluate the proposed strategies using real underwater acoustic signal data. Experimental results show that the proposed algorithm can achieve a detection accuracy comparable with that trained with 2000 samples using only 200 labeled samples.

A vertical line array (VLA) deployed at the seabed bottom captures the arrival angle interference structure in the frequency-beam domain resulting from the Direct (D) and Surface-Reflected (SR) arrivals of a broadband source. This interference structure, sensitive to the source's depth, serves as a basis for depth estimation. In order to address limitations related to bandwidth and nonlinear errors stemming from environmental differences, and to enhance the applicability of interference structures, a hybrid source localization method based on deep learning is proposed. This method employs an optimized residual network (ORN) to effectively extract and evaluate features from the frequency-beam domain sound intensity matrix. Simulated and experimental datasets are used to test the performance of the proposed method, and results suggest that the performance of the ORN model is much better than those of multi-Fourier transform approach (MSTDE), matching field processing (MFP) and traditional Convolutional Neural Network (CNN) models.

When employing the conventional beamforming (CBF) for the estimation of the direction of arrival of the Direct rays, one can observe a corresponding relationship between the arrival angle and the source distance, which can be used for range estimation. In the actual deep ocean environment, the arrival angle matched location method performs effectively in solving range estimation problems, although its performance is susceptible to the signal-to-noise ratio (SNR). To enhance the environmental adaptability and expand the application range of the source ranging method using the arrival structures in the beam do-main received by a vertical line array (VLA), we introduce a deep transfer learning (DTL) based source ranging method. Initially, a pre-trained model is established using simulation data generated under various SNRs through an ocean ambient noise model. Then high SNR experimental data is employed for DTL of the pre-trained model to fine tune the parameters. Finally, the experimental datasets are used to test the performance of the proposed method, and results suggest that the performance of the deep transferred model is much better than those of the traditional arrival angle matched location method and the model trained on noise-free data.

The acquiring geoacoustic model and associated geoacoustic parameters is vital for sound propagation modelling in a range-dependent environment. Conventional sequential methods, e.g., particle filtering and Kalman filtering have been widely used for geoacoustic inversion in range-dependent environment, which encounter challenges when confronted with unknown geoacoustic models that intrinsically varying with range. As an attempt to estimate geoacoustic model and associated geoacoustic parameters as well, a trans-dimensional particle filtering method is presented here. This method integrates birth-death rules into the filtering process, enabling automatic selection of appropriate geoacoustic models and estimating geoacoustic parameters in the same time. Numerical simulations were conducted based on an environment model of the South China Sea, where a sea trial was conducted in 2022. Numerical results demonstrate the efficiency in accurately estimating the geoacoustic model variations and associated parameters. Preliminary results of sea trial data processing are also presented here.

The you-only-look-once (YOLO) model identifies objects in complex images by framing detection as a regression problem with spatially separated boundaries and class probabilities. Object detection from complex images is somewhat similar to underwater source detection from acoustic data, e.g., time-frequency distributions. Herein, YOLO is modified for joint source detection and azimuth estimation in a multi-interfering underwater acoustic environment. The modified you-only-look-once (M-YOLO) input is a frequency-beam domain (FBD) sample containing the target and multi-interfering spectra at different azimuths, generated from the received data of a towed horizontal line array. M-YOLO processes the whole FBD sample using a single-regression neural network and directly outputs the target-existence probability and spectrum azimuth. Model performance is assessed on both simulated and at-sea data. Simulation results reveal the strong robustness of M-YOLO toward different signal-to-noise ratios and mismatched ocean environments. As tested on the data collected in an actual multi-interfering environment, M-YOLO achieved near-100% target detection and a root mean square error of 0.54° in azimuth estimation.

Underwater acoustic technology is essential for ocean observation, exploration and exploitation, and its development is based on an accurate predication of underwater acoustic wave propagation. In shallow sea environments, the geoacoustic parameters, such as the seabed structure, the sound speeds, the densities, and the sound speed attenuations in seabed layers, would significantly affect the acoustic wave propagation characteristics. To obtain more accurate inversion results for these parameters, this study presents an inversion method using the waveguide characteristic impedance based on the Bayesian approach. In the inversion, the vertical waveguide characteristic impedance, which is the ratio of the pressure over the vertical particle velocity, is set as the matching object. The nonlinear Bayesian theory is used to invert the above geoacoustic parameters and analysis the uncertainty of the inversion results. The numerical studies and the sea experiment processing haven shown the validity of this inversion method. The numerical studies also proved that the vertical waveguide characteristic impedance is more sensitive to the geoacoustic parameters than that of single acoustic pressure or single vertical particle velocity, and the error of simulation inversion is within 3%. The sea experiment processing showed that the seabed layered structure and geoacoustic parameters can be accurately determined by this method. The root mean square between the vertical waveguide characteristic impedance and the measured impedance is 0.38dB, and the inversion results accurately represent the seabed characteristics in the experimental sea area.

复杂海洋环境中信道的传输特性、时空变化、频散效应等一定程度上制约了主动声呐目标方位估计的性能.该文引入卷积神经网络,提出了适用于主动声呐中目标方位的高精度估计方法.仿真声场环境为浅海负梯度,主动发射信号为具有多普勒不变性质的双曲调频信号,水平线列阵作为接收装置,目标按仿真路线运动.该文利用Kraken进行声场数据仿真,并对接收的信号在频域做均匀加权常规波束形成,进而进行卷积神经网络的模型训练和测试.数值仿真研究表明,该文所用方法可以有效估计目标波达方向,对信噪比具有一定的鲁棒性.

为解决海底沉积层分层结构未知时的反演问题,提出了一种变维粒子滤波方法,利用声场的互谱密度,估计沉积层分层结构以及地声参数.仿真结果表明:变维粒子滤波在沉积层分层结构未知时,能有效反演沉积层层数以及地声参数,粒子的并行计算能使其相较于可逆跳蒙特卡洛马尔可夫链(rjMCMC)更加高效.利用变维粒子滤波,对南海垂直线阵列接收到的线性调频信号进行处理,反演结果与rjMCMC反演得到的沉积层层数和地声参数结果相近,说明了此方法能有效估计沉积层层数的同时反演浅海地声参数,得到可靠的参数后验概率密度.

In underwater acoustics, the performance of sonar echo detection is limited when the echo-to-noise ratio or echo-to-reverberation ratio is low. As an attempt to improve the echo detection rate and maintain a low false alarm rate, a method based on two-dimensional matched filtering (2-D-MF) and convolutional neural network (CNN) is proposed. The 2-D-MF divides the replica signal into multiple sub-replicas, each with different frequency components, and utilizes the sub-replicas to perform the matched filtering individually, obtaining 2-D-MF features that better represent the amplitude-frequency characteristics of the echo signal. The CNN is utilized as an echo detector to extract echo information from the 2-D-MF features and determine the presence of an echo. The proposed method is tested using data collected in the South China Sea, 2021. During the experiment, a transducer transmitted linear frequency modulation (LFM) signals, and a transponder, acting as an analog target, forwarded the LFM signals as echoes. The detection results demonstrate that this method can improve the echo detection rate by approximately 7% while maintaining a constant false alarm rate of 1‰.

The relationship between modal elevation angle and the relative arrival time between modes, derived from exploiting modal dispersion, provides source information that is less susceptible to environmental influences. However, the standard method based on modal dispersion has limitations for application. To overcome this, we propose a hybrid method for passive source ranging of low-frequency underwater acoustic-pulse signals in a range-independent shallow-water waveguide. Our method leverages deep learning, utilizing the intermediate results from the standard method as inputs, and short-time conventional beamforming to transform signals received by a vertical line array into a beam-time-domain sound-intensity map. The source range is estimated using an attention-based regression model with a ResNet backbone that has been trained on the beam-time-domain sound-intensity map. Our experimental results demonstrate the superiority of the proposed method, with a mean relative-error reduction of 71%, mean root-squared error reduction of 2.25 km, and an accuracy of 85%, compared to matched-field processing.

The Sound Speed Profiles (SSPs) in sea water have obvious time evolution characteristics, and their prediction can be regarded as a nonlinear time series prediction. Recurrent Neural Networks (RNN), a type of deep neural network designed for sequence modeling, can capture nonlinear relationships flexibly. Attention Mechanism (AM), which selects the most critical information for the current task, can describe the nonlinear relationships in space and temporal dimensions. In this paper, RNN and AM are used to construct a multivariate time series prediction model to learn the historical SSPs and predict the time-varying full-sea SSPs in shallow sea environment. Experiments on real sound speed data show that the proposed method is effective and outperforms other methods, and provides a new idea for the combination of physical model and machine learning in underwater acoustics.

A direction of arrival (DOA) estimation method based on a convolutional neural network (CNN) using an acoustic vector sensor is proposed to distinguish multiple surface ships in a selected frequency band. The cross-spectrum of the pressure and particle velocity are provided as inputs to the CNN, which is trained using data obtained by employing an acoustic propagation model under different environmental and source parameters. By learning the characteristics of acoustic propagation, the multisource distinguishing performance of the CNN is improved. The proposed method is experimentally validated using real data.

Direction-of-arrival (DOA) estimation for underwater acoustic sources is usually affected by multisource interference and ambient noise. In this study, DOA estimation is achieved by using a conventional beamformer modified by attention mechanism (A-CBF) which explores the spatial spectrum for DOA estimation that can focuses more on the peak of the desired signal while suppressing other peaks caused by interference and noise. The coefficients in A-CBF are learned by a neural network trained by array-received signals. On the basis of the above concept, the neural network determines the presence of the target in the received signals. From data obtained during a 2020 sea trial, the A-CBF model was trained by using a small amount of experiment data. The processing results demonstrate its performance of DOA estimation and target detection through suppressing multisource interference and focusing on the beams of the target ship in the spatial spectrum.

海底地声参数作为海洋声信道的重要组成部分,很大程度上决定了海洋声传播特性.地声参数可以通过反演算法获得,与距离相关的地声反演问题近年来是研究的热点.粒子滤波是一种高效的序贯寻优算法,可以在海底声学特性随距离缓慢变化情形下,有效解决地声参数的估计问题.但当海底声学特性随距离变化剧烈,如沉积层分层情况发生改变时,传统的粒子滤波则可能失效.为解决此问题,有文献使用了带桥接重采样的粒子滤波反演地声参数.文章在此基础上进一步改进了粒子的采样方式,并将此方法应用到基于海底反射系数的反演中.通过仿真数据处理结果证明,文中提出的改进粒子滤波在地声参数随距离变化剧烈的情况下仍具有较好性能,可以准确地估计海底特性随距离的变化规律和海底沉积层的声学参数.

Underwater acoustic target recognition based on ship-radiated noise is difficult owing to the complex marine environment and the interference by multiple targets. As an important technology for target recognition, deep-learning has high accuracy but poor interpretability. In this study, an attention-based neural network (ABNN) is proposed for target recognition in the pressure spectrogram with multi-source interference using an attention module to inspect the inner workings of the neural network. From data obtained during a September 2020 sea trial, the ABNN exhibited a gradual focus on the frequency-domain feature of the target ship and suppressed environmental noises and marine vessel interference, which led to high accuracy in the target detection and recognition.

Supervised classification algorithms are often used for marine noise classification. However, limited by insufficient labeled samples, the performance of the supervised classification method is typically influenced. To alleviate the limitations of insufficient labeled samples, in this paper, a semi-supervised noise classification method based on an auto-encoder (AE) has been proposed using radiated noise of four kinds of ships. This method takes a two-step training process, including unsupervised pre-training and supervised fine-tuning, making full use of unlabeled data and limited labeled data, respectively, which reduces reliance on label information for noise classification. The performance of this method is compared with traditional backpropagation neural networks (BPNN) and support vector machines (SVM). Experimental data analysis demonstrates that the semi-supervised noise classification method has improved the accuracy with different amounts of labeled samples, especially when labeled samples are relatively rare.

Non-uniform distribution of sound-speed profile (SSP) significantly impacts deep-sea sound propagation. A simple and efficient method is proposed for estimating spatial non-uniform SSP using empirical orthogonal function and towed temperature-depth sensors (TDs). Processing results of a deep-sea experiment show that this method requires only 2 towed TDs, and the root-mean-square error (RMSE) of reconstructed SSP is 1.0377 m/s. Results of the sound-field simulation using different SSPs show the necessity of considering non-uniform distribution of SSP, and the RMSE of transmission loss calculated using the reconstructed SSP is less than 2 dB when source depth is 50–350 m.

Source ranging based on ship-radiated noise is a crucial task in many practical applications. Deep neural networks (DNNs) have shown outstanding performance but poor interpretability on source ranging, leading to the heavily hidden risks of blind trust in the AI black box. In this study, an attention-based convolutional neural network (ABCNN) is proposed for the ship ranging in an attempt to visualize the features of concern in neural networks. Acoustic data of four ships were collected during a sea trial conducted in January 2021 to validate the ship ranging performance of ABCNN. Results showed high accuracy in ship ranging using synthetic data and part of the experimental data as a training set for the proposed method. The attention mechanism visualized a concentration on the inherent features of ships and the waveguide effect of underwater acoustic channels.

A modified convolutional neural network (CNN) is proposed to enhance the reliability of source ranging based on acoustic field data received by a vertical array. Compared to the traditional method, the output layer is modified by outputting Gauss regression sequences, expressed using a Gaussian probability distribution form centered on the actual distance. The processed results of deep-sea experimental data confirmed that the ranging performance of the CNN with a Gauss regression output was better than that using single regression and classification outputs. The mean relative error between the predicted distance and the actual value was ~2.77%, and the positioning accuracy with 10% and 5% error was 99.56% and 90.14%, respectively.