水下无人系统学报

AUV traversal planning method based on AMDQN

PANG Zhouqi, LIN Xiaobo, GENG Shicheng, HAO Chengpeng

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0028

[Abstract](81) [FullText HTML] (61) [PDF 1982KB](15)

Abstract:
To improve the traversal path planning capability of autonomous underwater vehicle(AUV), this paper proposes an AUV traversal path planning algorithm based on advanced M-DQN (AMDQN). First, this paper establishes a traversal planning environment suitable for practical scenarios. On the one hand, the ray coverage method is used instead of the traditional rectangular grid modeling method to improve modeling accuracy; on the other hand, the optional action set of the AUV is constrained to limit its maneuverability. Second, each component of the algorithm is carefully designed under the above environment. Specifically, in the state space design, this paper fuses vague global environmental information, accurate local environmental information centered on the AUV, and the AUV’s own position information to systematically represent the environment. In the reward function design, “rule penalty” and “edge guidance” are introduced, enabling the AUV to stably improve coverage along environmental edges. For the parameter update method, an adaptive temperature parameter update strategy is designed based on M-DQN, while the multi-step reward and a “dueling” network architecture are introduced to alleviate training variance. During the training process, “soft reset” of fully connected layer parameters is adopted to mitigate the local optimum problem. Finally, simulation results show that the proposed algorithm has stronger environmental adaptability and higher traversal coverage compared with traditional methods and unimproved reinforcement learning algorithms.

Multi-Agent Collaborative Search Path Planning Based on Reinforcement Learning

HE Congwei, XIE Yong, CHEN Yutao

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0163

[Abstract](118) [FullText HTML] (56) [PDF 1773KB](19)

Abstract:
To address the problem of optimizing multi-agent collaborative search path planning with maximizing cumulative detection probability, a multi-agent reinforcement learning model is developed. A multi-agent collaborative search algorithm based on Deep Q-Network, Joint-DQN, is proposed. It enhances the collaboration efficiency and stability among multiple platforms by designing an experience knowledge sharing mechanism; introduces a conflict detection mechanism and imposes penalties, effectively addressing the frequent path conflicts in multi-agent collaboration research; and designs a composite reward function to improve search coverage and reduce the rate of duplicate searches. Simulation experimental results demonstrate that this algorithm can effectively guide search platforms to avoid obstacles while efficiently moving in the direction where the target is most likely to be found, both in static and dynamic target scenarios. It rapidly enhances the cumulative detection probability while conducting efficient search, providing theoretical support and valuable guidance for multi-agent collaborative search path planning.

Adaptive Modal Decomposition Method for Underwater Explosion-Induced Shock Vibrations

LONG Yiyang, ZHU Wei, WEN Jun, JIA Xiyu, MA Feng

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0004

[Abstract](91) [FullText HTML] (56)

Abstract:
Structural shock–vibration signals induced by underwater explosions exhibit strong non-stationarity and broadband superposition. Conventional modal decomposition methods are prone to mode mixing and energy leakage, making it difficult to achieve stable frequency-band separation. To address these issues, an adaptive modal decomposition method for underwater-explosion-induced shock vibration, termed adaptive modal decomposition (AMD), is proposed. The method is built on frequency-domain parametric modeling, where the response spectrum is represented by basis functions with local support and adjustable scale, and the dominant frequency components are adaptively extracted and reconstructed via the joint optimization of center frequency, bandwidth, and amplitude parameters together with pruning constraints. A numerical simulation of an underwater-explosion stiffened-plate structure is used as an example, and AMD is systematically compared with empirical mode decomposition(EMD), empirical mode decomposition(VMD) and empirical mode decomposition(EWT). The results indicate that AMD achieves near-lossless reconstruction, with a maximum inter-modal cross-correlation coefficient of approximately 17.7% and an average spectral overlap ratio of 2.7%, both significantly lower than those of the compared methods, demonstrating its effectiveness for shock-vibration signal analysis under underwater explosion.

A Delay-Robust Multi-Agent Reinforcement Learning Approach for Cooperative Target Encirclement

FU Songchen, BAI Letian, ZHAO Shaojing, MENG Aomeng, LIANG Hong, LI Ta

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0173

[Abstract](109) [FullText HTML] (59) [PDF 1552KB](24)

Abstract:
To address delayed observations caused by acoustic communication in cooperative operations of multiple unmanned underwater vehicles (UUVs), a delay-robust multi-agent reinforcement learning based cooperative encirclement method is proposed. First, the mechanism of delayed observations is analyzed under common communication scenarios. Second, a turbulent flow field is modeled using the two-dimensional Navier–Stokes equations to construct a simulation environment that reflects realistic task settings. Then, a delay-robust multi-agent reinforcement learning method is proposed, and its constituent modules are described in detail. On this basis, the reward functions for the encirclement task, the network architectures, and the training procedures are designed, and ablation studies are conducted under different tasks and delay conditions. Experimental results demonstrate that the proposed method effectively handles delayed observations and maintains strong performance under varying delay levels, approaching the theoretical upper bound of the no-delay case in some tasks. Furthermore, the ablation results verify the effectiveness of each module in mitigating delayed observations, providing new theoretical support and a practical methodology for multi-UUV cooperative strategies under delayed observations.

AUV Terrain-Following Method Based on B-Spline Planning and Meta-Learning LTV-MPC

ZHU Yingjiao, YAN Tianhong, LIU Yingying, LIU Yili

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0011

[Abstract](86) [FullText HTML] (55) [PDF 1733KB](22)

Abstract:
To address the difficulty of simultaneously satisfying tracking accuracy, bottom-clearance safety and actuator constraints for autonomous underwater vehicle(AUV) navigating close to complex seabed terrain, this paper proposes a terrain-following method based on sliding-window B-spline planning and meta-learning-based adaptive linear time-varying model predictive control(LTV-MPC). The planning layer employs cubic B-spline under slope constraints to generate a reference path that maintains a prescribed altitude margin, combined with a sliding-window mechanism to achieve a balance between local optimization and global continuity. The control layer implements LTV-MPC with a meta-learning network that adaptively adjusts cost weights based on terrain curvature, depth and pitch errors to adapt to varying tracking difficulty under different terrain conditions. The meta-network parameters are trained offline via Bayesian optimization on closed-loop simulation data from multiple synthetic terrains, learning the mapping relationship between terrain features and optimal weights, requiring only forward inference online with high computational efficiency. In the specified simulation scenarios, validations on synthetic and real seabed terrain data demonstrate that, compared with LOS-PID, the proposed method reduces root mean square(RMS) tracking error by 42.5% and 31.3% under the two terrain scenarios respectively, compared with standard LTV-MPC, RMS tracking error is reduced by 4.5% and 6.1% respectively, while satisfying pitch angle and rudder angle constraints, effectively solving the balance problem between tracking accuracy and safety under complex terrain conditions.

Design and Application of Vehicle Mounted Industrial CT Nondestructive Testing System and Underwater Unmanned Equipment Defect Recognition Software

MA Junli, WANG Shuai, LIU Wenbin, GAO Haoyang, LI Chenhui, ZHANG Xitong, SU Chenghai

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0171

[Abstract](103) [FullText HTML] (69) [PDF 4796KB](23)

Abstract:
To satisfy the inspection demands of unmanned undersea equipment that are stored in remote areas, unsuitable for frequent transportation and difficult to conduct centralized testing, this paper builds a vehicle-mounted industrial computed tomography(CT) nondestructive defect detection and identification system with hardware subsystems including power supply, X-ray source, detection, acquisition and transmission. Combined with deep neural networks, an image recognition software for defects in industrial CT inspection products is developed to process inspection images and realize the identification and marking of internal product defects. Practical test results demonstrate that the proposed system can effectively detect and identify pores, cracks, debonding and other internal defects in high-energy fillers of unmanned undersea equipment. The imaging accuracy exceeds 98%, and both the missed detection rate and false alarm rate of defect identification are no more than 5%. This system can adapt to different materials and shapes of the products, and can operate in various environments. It has a promising application prospect.

Dual Propulsion Motor Control of Unmanned Lifeboat Based on Sliding Mode Allocation Controller

WANG Zhen, TANG Yu

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0170

[Abstract](102) [FullText HTML] (55) [PDF 1226KB](25)

Abstract:
The unmanned lifeboat is prone to yawing due to external water flow interference because of its light weight. In order to address the issue, this paper designs a cross-coupled speed cooperative control method for dual propulsion motors based on a sliding mode speed distributor. Firstly, it establishes a motor control model for the turning radius and rotational speed difference of the unmanned lifeboat. Secondly, it designs a dual-propulsion motor speed cooperative control system based on the cross-coupling control. Finally, it designs a speed distributor based on the sliding mode control, which can adjust the given speeds of the two propulsion motors to resist the external water flow interference so as to achieve the smooth navigation. The simulation results prove that the proposed control method can allocate the different set speeds for the dual-propulsion motor based on the actual situation. And this control method can control the course of the unmanned lifeboat more accurately.

Monitoring Paradigm for Deepwater Subsea Pipeline Laying and Key Underwater Wireless Optical Communication Technologies

LI Ziqing, HE Ning, CHEN Jianyi, WEI Jiaguang, FENG Xiaowei, HE Wenxuan, LIAO Peixuan, XIE Chenhua

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0159

[Abstract](76) [FullText HTML] (44) [PDF 13349KB](25)

Abstract:
To address the high multi-vessel coordination cost, limited timeliness, and stringent constraints of tethered operations in near-bottom monitoring of the touchdown point(TDP) during deepwater pipeline laying, an integrated monitoring architecture comprising an unmanned surface vehicle(USV), a towed/telemetry module(TMS), and an autonomous/remote underwater vehicle(ARV) is developed, together with a vertical underwater wireless optical communication(UWOC) scheme. A vertical UWOC approach based on quasi-omnidirectional LED array emission and adaptive receive-threshold control is proposed and evaluated. First, depth-dependent absorption and scattering attenuation coefficients are established from the chlorophyll concentration profile; under Lambertian, quasi-omnidirectional LED-array boundary conditions, Monte Carlo photon tracing with the Henyey–Greenstein(HG) phase function is introduced to overcome the bias of the constant-parameter Beer–Lambert approximation in spatially varying media, yielding the spatial distribution of received power and the 90% confidence coverage radius. Second, a hardware–software integrated implementation is completed: the transmitter employs blue/green LED arrays with secondary optics, the receiver adopts a large-aperture photomultiplier tube(PMT) with a narrowband filter, and OOK/IM-DD is used as the signaling scheme. At the software layer, a sliding-window adaptive threshold and gain control method(CFAR+AGC) jointly adjusts transmit power and receive gain, reducing reliance on high-precision pointing, acquisition, and tracking (PAT). Water-tank experiments verify link stability, pointing tolerance, and multi-rate performance(6~20 Mbps); open-sea trials in the Wenchang 16-2 field achieve a stable communication range of approximately 17 m under a water attenuation coefficient of 0.58 m⁻¹, with error-free transmission at 6.25 Mbps and robustness to relative motion and ambient-light fluctuations. The results demonstrate that the proposed closed-loop “layered channel–quasi-omnidirectional emission–adaptive reception” approach has strong transferability and engineering effectiveness, enabling continuous deepwater TDP monitoring without additional MSV deployment, thereby reducing cost and improving operational safety and timeliness.

Online optimization and path-following control of AUVs for piecewise linear paths

CHEN Shuwen, YAN Tianhong, ZHANG Yuxuan, WANG Rixian

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0166

[Abstract](112) [FullText HTML] (66) [PDF 1795KB](25)

Abstract:
Aiming at the problem of excessive tracking error caused by tangential jumps at switching points during the piecewise linear path tracking of autonomous underwater vehicle(AUV), this paper proposes an online path optimization method based on overlapping sliding windows. This method overlaps the windows to enable the new window to inherit the latter part of the optimization results from the previous window, thereby mitigating the end effects and decision myopia. The optimization of the path in the window gives priority to the parameterized cubic Bézier curve(PCBC) algorithm. When this algorithm is not applicable, an improved particle swarm optimization(PSO) is used to optimize the control points of the cubic Bézier curve, so as to obtain optimized path segments that meet the performance requirements. To achieve high-precision tracking of the optimized path segments by the AUV, this paper designs an improved line-of-sight(LOS) guidance method based on adaptive look-ahead distance, which dynamically adjusts the look-ahead distance by fusing tracking error, path curvature and navigation speed to improve guidance accuracy. Simulation analysis and sea trials verify the superiority of the improved LOS and the engineering practicability of its integration with the path optimization method.

Oblique Ice-breaking Load and Motion Characteristics Analysis of Water-exiting Vehicles

YE Yonghao, HE Baiyan, PEI Jinliang, ZHANG Yigan, QU Zehui, LIU Huaping, ZHANG Junhui, QI Runchao

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0002

[Abstract](120) [FullText HTML] (60) [PDF 3892KB](28)

Abstract:
Underwater vehicles possessing water-exit ice-breaking capabilities hold significant application value for polar scientific research and resource exploration. However, existing research primarily focuses on vertical ice-breaking, with a notable lack of investigation into the impact of the oblique angle on ice-breaking performance. Therefore, a numerical model for the oblique water-exit and ice-breaking process of a vehicle is established based on the Arbitrary Lagrangian-Eulerian (ALE) fluid-structure interaction algorithm. The effects of the oblique angle, initial velocity, and ice thickness on the load and motion characteristics of the vehicle are systematically analyzed. The results indicate that during the initial stage of ice-breaking, the impact of the vehicle's conical head induces intense local stress concentration in the ice sheet. This leads to the early initiation of radial cracks at the top surface, followed by failure originating from the center. The center of the resultant force on the vehicle deviates from its axis, causing an exacerbated deflection along the initial oblique direction, and a trend that becomes more pronounced as the initial velocity and ice thickness increases. For the θ=10° case, the vehicle’s attitude is governed by the ice-breaking kinetic energy: under conditions of low velocity and thick ice, the attitude exhibits a “deflection-recovery” pattern; conversely, under high velocity and thin ice conditions, it transitions to a “deflection-steady flight” pattern. The findings of this research provide a valuable theoretical reference for the design and development of polar cross-media vehicles in the future.

High-Fidelity Seafloor 3D Reconstruction Based on Cross-dimensional Gaussian Normal Transition Field

ZHAO Ximan, CHU Xuanhe, CHEN Han, LIU Siyuan

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0167

[Abstract](108) [FullText HTML] (71) [PDF 10699KB](29)

Abstract:
The demand for high-fidelity scene reconstruction of the seafloor is growing in fields such as marine scientific surveying and underwater environmental exploration. As an advanced explicit scene representation method, 3D Gaussian Splatting shows significant application potential in scene reconstruction and novel view synthesis. However, influenced by factors such as blurring effects in underwater medium, the results often exhibit defects like medium artifacts and structural distortion, severely limiting their applicability in complex underwater environments. To address these challenges, we propose a high-fidelity seafloor scene reconstruction based on cross-dimensional Gaussian normal transition field. First, we establish a cross-dimensional and normal transition system for Gaussian primitives, enhancing its capability for detailed geometric modeling of complex structures. Second, we introduce a Gaussian opacity-weighted filtering model to suppress reconstruction artifacts caused by medium blurring effects. Finally, experimental results across multiple underwater scenes demonstrate our method's capability for efficient scene reconstruction and novel view synthesis in underwater environments.

A Physics-Based Method for Drifting Buoy Trajectory Backtracking with Uncertainty Quantification

LI Hui, WANG Shui, XIONG Xinquan, YUE Peng, MO Lihong, WAN Wanghua

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0014

[Abstract](110) [FullText HTML] (55) [PDF 1779KB](35)

Abstract:
High-precision trajectory backtracking technology for drifting buoys is urgently needed for maritime search and rescue (SAR) and pollution source tracing, yet traditional Lagrangian models exhibit significant errors in complex marine environments. This study proposes a physics-driven trajectory backtracking model for marine drifting buoys, which innovatively introduces a dynamic diffusion coefficient based on autocorrelation analysis of wind and current field time series to optimize the subgrid-scale velocity compensation mechanism in random walk models. The model integrates a drift dynamics framework incorporating wind forcing, ocean currents, and Coriolis force, and combines Monte Carlo simulation with kernel density estimation to quantify the spatiotemporal uncertainties in trajectory backtracking. Validated using North Atlantic Ocean Internet of Things buoy data across four typical marine environments—tropical open ocean, current convergence zones, temperate westerlies, and nearshore complex terrain—the proposed model achieves 72-hour trajectory backtracking errors of 3.9-5.8 km. Compared with traditional Lagrangian models, accuracy improvements of 74%, 55%, 59%, and 22% are achieved in the four sea areas respectively, effectively addressing the trajectory backtracking accuracy problem in complex marine environments and providing reliable technical support for maritime emergency rescue and pollution source localization.

A TCN-Attention-Based Pseudo-Velocity Measurement Generation Method for Loosely Coupled SINS/DVL Integrated Navigation

WANG Guoxiang, HAN Xingcheng, GAO Shengwen, WANG Ling

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0047

[Abstract](106) [FullText HTML] (68) [PDF 3674KB](37)

Abstract:
DVL unavailability degrades the accuracy of loosely coupled SINS/DVL integrated navigation for autonomous underwater vehicles. To address this problem, a pseudo-DVL velocity measurement generation method based on a temporal convolutional network with an attention mechanism (TCN-Attention) is proposed. The method uses the angular velocity and specific force measured by the inertial measurement unit (IMU), together with the attitude, position, and velocity obtained from inertial navigation computation, as sequential inputs. During the DVL-available stage, supervised samples are constructed using DVL velocity for offline network training. During the DVL-unavailable stage, the trained model outputs pseudo-velocity measurements, which are incorporated into the extended Kalman filter (EKF) update to suppress inertial error accumulation. Causal dilated convolutions are adopted to extract temporal features, and an attention mechanism is introduced to enhance the representation of key dynamic segments such as turning, acceleration, and deceleration. Simulation results based on 16 trajectory datasets show that, compared with the temporal convolutional network (TCN) and the gated recurrent unit with attention model (GRU-Attention), the proposed method achieves better performance in east- and north-velocity errors as well as absolute trajectory error, and reconstructs trajectories closer to the ground truth, demonstrating its effectiveness and robustness under continuous DVL-outage conditions.

A Review of Energy and Propulsion Technology Development for Autonomous Undersea Vehicles

FENG Shuai, LIU Weijie, YANG Jian, WENG Weiguo

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0169

[Abstract](375) [FullText HTML] (128) [PDF 14330KB](134)

Abstract:
Autonomous undersea vehicles(AUVs) play a pivotal role in ocean engineering, marine scientific exploration, and military operations. Among their core subsystems, the energy and power system is particularly critical, as its performance directly determines the vehicle’s endurance, operational range, and overall efficiency. This study classifies AUVs from multiple perspectives and examines the principal characteristics and applications of their energy and power systems. Particular emphasis is placed on key enabling technologies, including high-energy-density battery systems, underwater charging methods, high-density hydrogen and oxygen storage, and advanced battery management. Finally, the paper outlines prospective directions for energy and power technologies in AUVs, with the aim of providing valuable insights for the future development of their energy systems.

Probability assessment and prediction method for escape area of underwater lost targets

SUN Jihong, ZHENG Yi, YANG Xiangfeng

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0062

[Abstract](123) [FullText HTML] (69) [PDF 1430KB](37)

Abstract:
In the complex environment where underwater high-speed maneuvering vehicles face both real and decoy targets, incorrect target selection may lead to the escape of actual high-value targets. Traditional non-targeted re-search strategies are inefficient, necessitating an in-depth analysis of high-value target countermeasure strategies, comprehensive consideration of their maneuverability and tactical choices, and the formulation of targeted counter-countermeasure and re-search strategies. Based on the analysis of high-value target countermeasure strategies, this paper constructs a target escape area model and proposes a probability assessment method for lost target escape areas, aiming to predict the escape probability of targets within the countermeasure area. This method helps improve the success rate of target re-search, reduce the underwater high-speed maneuvering vehicle's search time, and enhance overall search efficiency.

Modeling and Analysis of Underwater Vehicles Wake-Induced Electromagnetic Fields and Internal Wave Characteristics in a Stratified Ocean

WANG Xiangjin, ZHANG Jiansheng, WANG Xintong, YAN Linbo, LAN Qing

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0162

[Abstract](83) [FullText HTML] (74) [PDF 1390KB](46)

Abstract:
To counter the threat of underwater vehicle stealth and meet the demand for non-acoustic detection, this study investigates the influence mechanism of underwater vehicle wakes in density-stratified ocean environments. Most existing studies are based on the uniform fluid assumption, neglecting the effects of internal waves induced by stratification. This paper establishes a novel mathematical model for the velocity field of an underwater vehicle wake in a stratified fluid, decomposing the wake into a linear superposition of surface wave and internal wave components. Based on electromagnetic induction theory, the expression for the induced electromagnetic field is derived. Through numerical simulations, the spatial distribution, attenuation patterns, and component contributions of the induced magnetic field are analyzed for underwater vehicles at depths ranging from 10 m to 50 m. The results indicate that in a stratified environment, the surface wave-induced magnetic field has a high peak value (0.15 nT in the near-field) but decays rapidly with distance. In contrast, the internal wave-induced magnetic field has a lower peak value (0.006 nT in the near-field) but is more stable and decays slowly, becoming dominant in the far-field. Furthermore, as the submergence depth increases, the contribution of the internal wave component grows significantly (reaching 84.9% in the near-field at a depth of 50 m). This study reveals, from both theoretical and simulation perspectives, that internal waves are the key physical mechanism for far-field detection, providing a new theoretical basis for developing non-acoustic detection technologies for underwater vehicles.

Research on Factors Affecting Leakage Current of Lithium Reserve Battery Packs

GAO Xinlong, JIA Bin, LIN Pei, CHENG Haichao, LI Xuehai

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0083

[Abstract](133) [FullText HTML] (55) [PDF 995KB](58)

Abstract:
The distribution of leakage current of lithium reserve battery packs where the battery cells are in a common electrolyte state was studied, using the equivalent circuit simulation calculation method. The effects of the series cells quantity, electrolyte conductivity, and the structural characteristics of the injection tube on the leakage current were analyzed. A leakage current experimental device for battery modules was established. By comparing the measured results with the simulation results, the validity of the simulation calculation method was confirmed.

A Fast Algorithm for High-Frequency Approximate Scattering Acoustic Field of Complex Multi-Targets Based on the Planar Elements Method

MENG Lining, ZHAO Jingshu, LI Zhenhui, LIU Rui

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0090

[Abstract](167) [FullText HTML] (90) [PDF 5190KB](70)

Abstract:
With the development of unmanned undersea vehicle(UUV) cluster operations, the detection and recognition of complex multi-targets underwater have garnered significant attention. To address this, we established a fast calculation model based on the Planar Elements method to improve both computational accuracy and efficiency. Initially, the method was employed to calculate the target characteristics of a dual-target model. The accuracy of this method was validated by comparing the calculation results with physical field simulations and experimental measurements from an anechoic water tank.To enhance computational efficiency, the OpenMP parallel algorithm was introduced. By optimizing the loop iteration scheduling mechanism according to the varying computational difficulties of complex multi-target scattering acoustic field characteristics under different incident angles and frequencies, high thread load balance was achieved, yielding a 5.3x speedup. This fast algorithm was then applied to investigate more complex multi-target models. By analyzing the angle-frequency maps of target characteristics, the regular variations in the high-frequency scattering acoustic field characteristics of multi-targets with increasing frequency were obtained, revealing that target strength exhibits extrema at certain angles. Meanwhile, high-frequency interference fringes were observed. The correlation between scattering characteristics and geometric positions was analyzed. The research results provide a theoretical reference for underwater target acoustic detection and characteristic studies.

Cetacean call recognition and classification model based on multimodal MAE data augmentation network

LIU Yueyue, NIU Qiuna, SUN Yue, WANG Jingjing, SHI Wei

, Available online , doi: 10.11993/j.issn.2096-3920.2026-0052

[Abstract](192) [FullText HTML] (105) [PDF 5139KB](83)

Abstract:
Passive acoustic monitoring-based call recognition and classification are essential means for marine animal conservation and population surveys. To address the issues of data scarcity and inter-class imbalance in call recognition and classification, data augmentation methods hold significant practical value and research importance. However, marine animal calls contain rich acoustic information, and relying solely on frequency-domain feature extraction lacks the capability to model audio structure and semantics, making it difficult to effectively capture the deep features of calls. To this end, this paper proposes a data augmentation network based on a multimodal masked autoencoder (MAE-MF), which breaks through the limitations of single-modal information. The network employs Mel-spectrograms as the primary modality, integrates temporal features and frame-level statistical metrics to form multimodal inputs, and incorporates semantic labels as conditional guidance for reconstruction. To scientifically validate the effectiveness and practical value of the proposed data augmentation network, a cetacean call recognition and classification model is further constructed based on the MAE-MF network. Experimental results on the Watkins dataset demonstrate superior performance of the proposed method, with improved spectrogram reconstruction quality compared to mainstream algorithms. The proposed method achieves an average recognition accuracy of 97.6% across six cetacean species, representing an improvement of 6.72 percentage points over the baseline MAE method. This scheme effectively alleviates the inter-class imbalance issue and provides reliable technical support for cetacean conservation research.

Anti-Disturbance Control for Underwater Propulsion Motor at Low Speed Based on Hybrid Resolver and High-Frequency Injection Observation

WANG Yu, DUAN Luobao, CUI Jialun, WANG Yuankui

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0144

[Abstract](241) [FullText HTML] (125) [PDF 2247KB](139)

Abstract:
The low-speed control performance constitutes a fundamental prerequisite for unmanned underwater vehicle propulsion systems to execute critical missions such as deep-sea exploration and military reconnaissance effectively. To address the need for enhanced control capabilities during low-speed operations, this paper systematically examines limitations in permanent magnet synchronous motor drive systems employing both position-sensor-based schemes and sensorless schemes. Resolvers introduce position detection errors under harsh environmental conditions, while among dominant sensorless solutions, back-electromotive-force observers contain inherent observation dead zones near zero speed. Although high-frequency signal injection methods improve low-speed observation performance, their estimation accuracy remains susceptible to motor parameter variations. Crucially, the accuracy of all sensorless schemes exhibits critical dependence on current sampling precision, making such approaches vulnerable to severe engineering challenges in complex interference-intensive operating conditions. To resolve these issues, this paper proposes a hybrid observation-based low-speed anti-disturbance control strategy that integrates resolver technology with high-frequency square-wave injection. By applying hardware redundancy and information fusion techniques, the methodology achieves comprehensive integration between the absolute position reference provided by resolvers and dynamic observations generated through high-frequency square-wave injection. This synthesis establishes an advantage-complementary observation architecture that significantly enhances system robustness in difficult scenarios: low-speed operations, variable loading conditions, and signal interference contexts. Simulation results verify the capability of the method to suppress detection error interference arising from position sensors and current sensors concurrently, enabling stable and precise rotor position estimation. The framework therefore delivers a high-reliability control solution for underwater equipment propulsion systems.

Design and simulation of mechanical biomimetic fish tail driven by EAP material

WANG Sijiao, ZHANG Haoyi, CHENG Yanlin, CAO Kaiming

, Available online , doi: 10.11993/j.issn.2096-3920.2025-0164

[Abstract](294) [FullText HTML] (187) [PDF 1812KB](161)

Abstract:
Against the backdrop of advancing marine conservation and exploration, traditional underwater propulsion systems are often hampered by inherent drawbacks such as structural complexity and low motion efficiency. In contrast, flexible materials have emerged as a research focus in underwater actuation due to their superior adaptability, high safety, and remarkable flexibility. Leveraging the favorable core properties of Electroactive Polymer (EAP), namely its high energy density and efficient electromechanical coupling, this study introduces a novel biomimetic caudal fin actuator. This design incorporates a spring element to harness flexural deformation and elastic recovery, effectively simulating the cyclic contraction and relaxation dynamics characteristic of the Body and/or Caudal Fin (BCF) propulsion mode in fish, thereby achieving continuous, compliant changes akin to tail musculature. Based on hydrodynamic theory, the coupled interaction mechanism between fin kinematics and thrust generation is systematically analyzed. An instantaneous mechanical model for fin-ray oscillation is developed and solved by incorporating experimental data. A three-dimensional numerical simulation model is established using Fluent software. The validity of the proposed mechanical model is confirmed through comparative analysis between the computational results from dynamic meshing and the model's predictions. This work provides reliable theoretical support and experimental evidence for the design and development of new biomimetic robotic fish driven by this innovative actuator.

Latest Articles

Journal News> >

Download> >