As seabed exploration activities increase, side-scan sonar (SSS) is being used more widely. However, distortion and noise during the acoustic pulse’s travel through water can blur target details and cause feature loss in images, making target recognition more challenging. In this paper, we improve the YOLO model in two aspects: lightweight design and accuracy enhancement. The lightweight design is essential for reducing computational complexity and resource consumption, allowing the model to be more efficient on edge devices with limited processing power and storage. Thus, meeting our need to deploy SSS target detection algorithms on unmanned surface vessel (USV) for real-time target detection. Firstly, we replace the original complex convolutional method in the C2f module with a combination of partial convolution (PConv) and pointwise convolution (PWConv), reducing redundant computations and memory access while maintaining high accuracy. In addition, we add an adaptive scale spatial fusion (ASSF) module using 3D convolution to combine feature maps of different sizes, maximizing the extraction of invariant features across various scales. Finally, we use an improved multi-head self-attention (MHSA) mechanism in the detection head, replacing the original complex convolution structure, to enhance the model’s ability to focus on important features with low computational load. To validate the detection performance of the model, we conducted experiments on the combined side-scan sonar dataset (SSSD). The results show that our proposed SS-YOLO model achieves average accuracies of 92.4% (mAP 0.5) and 64.7% (mAP 0.5:0.95), outperforming the original YOLOv8 model by 4.4% and 3%, respectively. In terms of model complexity, the improved SS-YOLO model has 2.55 M of parameters and 6.4 G of FLOPs, significantly lower than those of the original YOLOv8 model and similar detection models.

The Inertial Navigation System (INS) and Doppler Velocity Log (DVL) combination navigation system has been widely used in Autonomous Underwater Vehicles (AUVs) due to its independence, stealth, and high accuracy. Compared to the standalone INS or DVL, the integrated system provides continuous and accurate navigation information. However, the underwater environment is complex, and system noise and observation noise may not satisfy the condition of mutual independence. In addition, the DVL may produce abnormal measurement values during operation. In this study, an Event-Triggered Correlation Noise Filter (ETCNF) method was designed for fusing INS and DVL data. An auxiliary matrix was introduced to decouple the correlated noise, allowing novel state estimation. Moreover, the event-triggered mechanism detected and eliminated abnormal values in DVL measurements, which improved the positioning accuracy and robustness of the INS/DVL integrated system. Finally, simulation experiments were conducted to verify the effectiveness and superiority of the proposed algorithm.

Geoacoustic inversion typically involves the collection of far-field underwater acoustic data to obtain seabed geoacoustic parameters using empirical formulas and matched field inversion (MFI) techniques. However, acoustic data propagated over long distances can introduce inevitable errors into inversion results, and traditional MFI techniques suffer from low computational efficiency. Although deep learning technologies have been applied to geoacoustic inversion, conventional deep neural network (DNN) models struggle to capture the long and short-term dependencies in bottom reflection data, leading to suboptimal inversion accuracy. These issues present challenges in rapidly and accurately acquiring geoacoustic parameters over large areas. To address this, we propose a near-field bottom reflection signal collection method (NBRC), collecting bottom reflection signals over a wide range of grazing angles by drifting. Utilizing the characteristics of near-field sound propagation, we constructed the Bottom Reflection Coefficient Sequence (BRCS) dataset using the wavenumber integration method. We then introduce a novel deep learning model, Self-Attention Geoacoustics (SAG), based on multi-head self-attention (MHSA) mechanisms, which improves inversion accuracy. Additionally, we propose an adaptive-weight multi-task learning (AW-MTL) training strategy, significantly enhancing the prediction accuracy of sound attenuation. Experimental results demonstrate that our method outperforms conventional geoacoustic inversion methods based on MFI and DNNs in terms of efficiency and accuracy, proving the superiority of our approach.

Implicit Full Waveform Inversion (IFWI) has emerged as a promising framework for reconstructing complex subsurface structures. However, conventional methods suffer from a fundamental bias conflict, in which the inherent spectral bias of neural networks, such as the low-frequency bias of MLPs and the high-frequency bias of SIRENs, conflicts with the physics-based gradient’s requirement for progressive frequency reconstruction. To resolve this, we propose Encoding-IFWI, a Decoupled Spectral Pre-conditioning framework. Unlike existing approaches that couple spectral bias with network architecture, our method utilizes Fourier encoding to engineer a neutral, broadband kernel basis. Theoretical analysis based on the Neural Tangent Kernel (NTK) reveals that this design empowers the physics-based gradient to select requisite frequency components, allowing the wave equation residuals to autonomously guide the spectral evolution of the model. Comprehensive validation on Marmousi, Overthrust, and SEG/EAGE Salt models confirms that Encoding-IFWI significantly improves convergence speed and imaging resolution, offering a robust solution that effectively balances large-scale structural continuity with fine-scale geological details.