Assessing the propagation of oil spills within port areas is essential for ensuring adequate environmental protection and management. The behavior of an oil spill depends on multiple interacting factors, including oil type, volume, release location, prevailing hydrodynamic and meteorological conditions, the trajectory and fate of the oil, and the characteristics of the affected coastal zone. Therefore, the development of systems capable of rapidly monitoring and predicting oil spill movement is fundamental to enabling timely and effective emergency responses. Both stochastic and numerical modeling approaches can be employed for this purpose. In particular, integrating predictive technologies that can identify which port infrastructures are likely to be impacted during an oil spill event can significantly enhance preparedness and response strategies. The primary objective of this study is to implement a two-stage extreme gradient boosting (XGBoost) framework to enhance oil spill prediction and management capabilities. The methodology was applied to the Port of Augusta (Sicily, Italy), one of the most strategic industrial and energy hubs in the Mediterranean. The proposed framework consists of two sequential stages: a binary classification model to predict whether oil will reach a given segment, followed by a regression model to estimate the arrival time of the impact. Several model configurations were tested, including segment-specific models, a global model trained on all segments, and a global model incorporating segment identifiers as categorical features to account for spatial heterogeneity. Results show that while segment-specific models achieved high classification accuracy, they exhibited limited generalization in predicting arrival times. The global modeling strategy significantly improved regression performance, and the inclusion of spatial identifiers further enhanced both classification and timing accuracy. The final configuration achieved binary classification accuracy close to 99% and reduced mean absolute error in arrival time predictions to approximately 0.6 hours. These findings confirm the effectiveness of gradient-boosting methods in capturing complex nonlinear relationships in coastal oil spill processes, while also demonstrating that computational efficiency can be improved without compromising predictive performance, supporting the integration of machine learning into operational oil spill risk assessment systems.
Underwater acoustic propagation is strongly influenced by variations in physical properties across different media in the sea, with the seafloor media exerting a particularly critical role. The geoacoustic parameters of seafloor sediments are therefore of central importance to underwater acoustics and ocean engineering. However, the high cost and complexity of in situ measurements make accurate acquisition of such parameters difficult, particularly for large-scale seafloor mapping. To address these challenges, this study proposes a method for accurately inverting the sediment geoacoustic parameters from acoustic backscattering. First, a forward model of acoustic backscattering on the sediment interface was developed based on the approximate fluid density model, establishing the physical relationship between sediment parameters and acoustic backscattering strength. The model laid the foundation for subsequent sediment parameters inversion. To solve the multidimensional inversion problem, a multilayer perceptron network was employed to develop an inversion model from backscattering strength. The performance of the inversion method is evaluated using the mean squared error. The maximum inversion error is within 1.184 dB, which confirms the effectiveness of the proposed approach in investigating sediment parameters. Furthermore, a set of sonar data, which acquired from Qiandao Lake (Chun’an County, Zhejiang Province, China), was used to calculate the lake floor backscattering strength and conduct geoacoustic parameter inversion tests. Comparison with in situ measurements confirms that, with the backscattering strength obtained from the lake data, the method successfully inverted the sediment parameters for large-scale seafloor, it is cost-effective, reliable, and accuracy sufficient for practical engineering applications.
Underwater acoustic target recognition, which is a key area in ocean sensing technology, is challenged by the complexities of acoustic channels, environmental noise, and target diversity. Owing to inadequate shallow feature characterization and rigid model architectures, it is difficult for traditional methods to achieve robust and accurate recognition in dynamic and variable hydroacoustic environments. To address this limitation, we propose an attention bifocal network deep recognition framework. The core of this framework consists of three innovative and synergistic modules. The subband concatenation constant Q-transform (SC-CQT) was designed to inject domain-specific prior knowledge into the time–frequency transformation process explicitly, thereby enabling the active reinforcement of key discriminative features. The dual-axis attention refiner is introduced to enrich feature maps with dynamic and contextually relevant information by employing orthogonal decoupling of the frequency and time dimensions, in addition to multiscale semantic perception. Finally, these enhanced features are then provided as inputs to the core decision layer of the attention-guided hierarchical bifocal network (ABN). Drawing inspiration from the cognitive mechanisms of biological vision, the proposed network utilizes a layered architecture with dual pathways for overview and focus, which facilitates the efficient allocation of computational resources and refinement of the decision-making process. Experimental results reveal that the proposed framework achieved recognition accuracies of 85.25% and 96.42% on two open-source data sets.
To address detection failures caused by image distortion and insufficient 3-D information acquisition when using monocular cameras in underwater environments, we introduce MaRIne, an underwater monocular real-time object detection and geometric estimation framework based on motion inversion. We propose an innovative object detection compensation mechanism that dynamically estimates the position of the target bounding box through the motion inversion technique. By employing a Kalman filter framework, we achieve efficient fusion of multimodal data by integrating object detection results with high-frequency motion inversion outputs. In addition, we establish a geometric model constrained by six degrees-of-freedom motion, which enables robust estimation of the target’s geometric parameters via cross-time-step matching of bounding boxes. Experimental results show that our method improves the accuracy of underwater object detection by 40% compared to YOLOv8n, while being 14 times faster. Furthermore, it enables real-time and reliable estimation of targets’ geometric parameters, achieving a mean absolute percentage error of less than 12.5% for target’s depth estimation.
In this study, inertial sensors were used to develop a real-time measurement system to estimate the attitude and tension of mooring chains on floating offshore wind power platforms. This system integrates a mooring chain attitude sensing device, an attitude estimation algorithm, and a tension estimation algorithm to measure the attitude and tension of mooring chains. The mooring chain attitude sensing device integrates inertial sensors, signal transmission modules, and microcontrollers, which are uniformly installed along a mooring chain. These sensors transmit dynamic signals to a human–machine interface to evaluate mooring chain curvature and tension variation. The attitude estimation algorithm uses an extended Kalman filter to merge acceleration, angular velocity, and magnetic field data collected from the sensors, thereby enabling the calculation of mooring chain movement to determine the mooring chain attitude. The tension estimation algorithm analyzes acceleration signals to assess the stress condition of the mooring chain and to determine whether it is exposed to excessive tension or is at risk of rupture. In this study, numerical simulation software was used to validate the reliability of tension estimation. Experimental analysis revealed that, under scaled testing conditions simulating the extreme environment of a 50-year typhoon return period in Taiwan, the estimated tension values deviated by 13.5% from the simulated values. Under a 1-year typhoon return period condition, this deviation increased to 27.9%. In summary, the proposed system enables the real-time monitoring of mooring chain curvature and mechanical stress. It transforms conventional numerical-parameter-based underwater cable monitoring into visualized imaging, thereby enabling engineers to effectively interpret the dynamic behavior of mooring chains exposed to wave-induced forces.
A combined maneuvering–seakeeping framework is developed for the efficient prediction of ship maneuvering performance in regular waves. The seakeeping module employs a frequency-domain 3-D Rankine panel method with double-body linearization and direct pressure integration to precompute second-order wave drift forces, organized in a multidimensional database parameterized by forward speed, lateral velocity, wave heading, and wavelength. A generalized upwind difference scheme, derived through a Taylor series expansion on nonuniform meshes, supports free-surface discretization with near-hull refinement and far-field coarsening within a single computation. The maneuvering module solves a 4-degree-of-freedom (DOF) maneuvering modeling group (MMG) model (surge, sway, yaw, and roll) in the time domain, retrieving the precomputed drift forces via database interpolation. Once the database is constructed offline, each maneuvering simulation requires only 4–6 s, compared with approximately 30 min per maneuver for online coupling approaches. The framework is validated on the S175 container ship at three levels: first-order roll RAOs and second-order drift forces against published seakeeping experiments; calm-water turning and zig-zag indices against free-running model tests, with all predictions satisfying the IMO MSC.137(76) criteria; and wave-induced trajectory drift in regular waves against basin measurements. The method is then applied to a new-build luxury cruise ship to illustrate its use at the design stage, where beam seas emerge as the most adverse condition and produce the largest departures from the calm-water baseline. The wave-induced trends remain qualitatively consistent across the two hulls, supporting the framework’s utility for early stage parametric assessment pending experimental confirmation on the new hull.
This article proposes a high-resolution wideband direction-of-arrival (DOA) estimation method based on an improved real-valued multiple signal classification (IRV-MUSIC) processing framework. While real-valued (RV) matrices inherently preserve both the signal and its complex conjugate components—which may introduce phase ambiguity—we strategically leverage this property to constrain the spectral search range. Furthermore, complex-valued operations are converted into RV operations, thereby achieving significant computational savings. Building upon traditional RV MUSIC DOA estimation methods, we introduce an improvement strategy involving the reorganization of the received signal’s RV covariance matrix into a centrosymmetric structure. This restructuring effectively enhances the angular resolution and estimation accuracy of the algorithm. Furthermore, the proposed method integrates wideband signal processing techniques to enable rapid and efficient discrimination of wideband target signal DOAs. Simulation results demonstrate that the algorithm substantially improves DOA estimation performance while simultaneously reducing computational complexity.
Localization of underwater cooperative targets plays a critical role in various fields of oceanic engineering and has received tremendous attention. This study aims to address the challenging problem of direction-of-arrival (DOA) estimation for a single cooperative target in shallow-water multipath channels. Although traditional underwater acoustic (UWA) DOA estimation strategies provide effective solutions, performance is greatly affected by random multipath propagation uncertainties and noise in UWA channels, especially in shallow-water acoustic channels. In this article, a novel method is proposed to achieve multipath decoupling shallow-water acoustic DOA estimation via virtual sources reconstruction (MD-VSR). First, multipath is modeled as the superposition of direct paths from multiple virtual sources, thereby decoupling the channel parameters from the DOAs in the received signal formulation. Second, with one array element as reference, Doppler shift and amplitude of each direct path are estimated through a spectral peak search, while time delay is obtained in a sequential manner to avoid phase ambiguity. Subsequently, the direct signals are reconstructed and used to derive elementwise equivalent array responses, which are combined to form an equivalent array response matrix. The DOAs of the virtual sources are then calculated using a minimum-variance optimal weighted fusion scheme, from which the final target DOA is selected based on minimal time delay and amplitude attenuation. Numerical simulations and at-sea experimental results demonstrate that the proposed MD-VSR method achieves a lower root-mean-square error compared to conventional methods in multipath channels.
This article explores the issue of event-triggered control with prescribed performance for autonomous underwater vehicles subject to input saturation and external disturbances. To avoid the problem of violating the behavioral boundaries, a self-adjustable prescribed performance function (SAPPF) is designed on the basis of a finite-time prescribed performance function (FTPPF), which appropriately expands the behavioral boundaries through the use of auxiliary systems and performance indicator functions, and the behavioral boundaries revert back to the original form of FTPPF when the system tends to stabilize. The SAPPF mentioned in this article can relax the initial conditions while guaranteeing the amount of overshoot and convergence time of the system. To save communication resources, a relative threshold event-triggered mechanism is incorporated between the controller and the actuator. In conclusion, the simulation results validate the proposed control strategy’s effectiveness in this research.
This study investigates a novel drag-reduction strategy for submersibles during ascent, achieved by altering the wake structure with surface-mounted flexible attachments in varying numbers. Previous studies have mainly focused on the influence of flexible attachment length on wake modulation, while the effect of attachment number on wake evolution and quantitative drag reduction remains insufficiently understood. To address this knowledge gap, the flow structures around submersibles equipped with different numbers of flexible attachments were analyzed and compared at a Reynolds number of 109 561, with a particular focus on wake characteristics such as Reynolds stress and turbulent kinetic energy (TKE). The wake structure was then analyzed using discrete wavelet transform (DWT) and continuous wavelet transform (CWT) methods. The results demonstrate that flexible attachments stabilize the wake, reduce turbulence intensity, and improve the spatial uniformity of TKE. A maximum drag coefficient reduction of 5.3% is obtained with 20 flexible attachments. Wavelet analyses, which include both DWT and CWT, reveal a dual-scale modulation mechanism: the flexible attachments preferentially break down large-scale vortices while promoting energy cascade toward smaller scales, thereby enhancing turbulent energy dissipation. This multiscale flow modulation ultimately leads to a more stable wake structure, as evidenced by spatial correlation analyses showing significant suppression of velocity fluctuations. However, further increasing the number of flexible attachments beyond this optimum induces wake broadening and adverse secondary flows, resulting in a rebound of drag.
With the development of integrated space–air–land–underwater networks, cross-medium communication has become increasingly important for ocean exploration and monitoring. As a promising solution, dual-hop radio frequency (RF)-underwater wireless optical communication (UWOC) transmission can support long-distance communication with high bandwidth and low latency. This article investigates the secrecy performance of a dual-hop RF-UWOC downlink system, where the first hop is an RF link from a satellite to a multiantenna relay in the presence of eavesdroppers, and the second hop is a UWOC link with underwater turbulence modeled by a lognormal distribution. Specifically, the random distribution characteristics of each node’s signal-to-noise ratio are first studied. Then, approximate and closed-form expressions are derived for the first hop’s nonzero secrecy capacity and the second hop’s outage probability, respectively. The validity of the proposed model is validated, and the influence of assorted parameters on the system’s secrecy performance is examined through Monte Carlo simulation methods.
Underwater acoustic communication often suffers from severe delay and Doppler spreads, compounded by limited bandwidth. This article investigates sparse channel estimation in an orthogonal signal division multiplexing (OSDM) system. While sparse signal processing is typically framed as an $\ell _{1}$-norm optimization problem, the $\ell_{0}$-norm formulation is more directly related to sparsity but is generally NP-hard. We introduce a global optimal solution for $\ell _{0}$-norm optimization in OSDM, develop an associated channel estimation method, and compare its communication performance against $\ell _{1}$-norm optimization. Performance is evaluated through numerical simulations in a static setting and sea trials in a dynamic environment. The simulation results indicate that $\ell _{0}$-norm-based channel estimation outperforms $\ell _{1}$-norm optimization under ideal threshold tuning. In contrast, experimental results show that the two methods yield comparable performance with practical Stein’s unbiased risk estimate-based threshold tuning. This suggests that the practical advantage of $\ell _{0}$-norm optimization is limited by the discrepancy between the threshold that minimizes channel estimation error and the threshold that maximizes demodulation performance.
Underwater acoustic communication is a well-established technology supporting data transmission over several hundreds of meters, up to kilometers. The link reliability is affected by severe channel impairments, including frequency selectivity, multipath and time-variability, making signal detection very challenging. Hence, effective channel equalization is mandatory to compensate the signal distortion at the receiver. In this contribution, a cooperative least mean square adaptive rule is described with application to fractionally spaced channel equalization schemes based on lattice filtering structures. Classical lattice-based adaptation relies on backward prediction errors only, while the novel cooperative adaptive rule presented here, which stems from a certain property of the infinite impulse response minimum mean square error equalizer, uses both the backward as well as the forward prediction errors, thus fully exploiting the statistical content of the received signal. Numerical results obtained from very challenging time-varying underwater scenarios not only confirm the merits of fractionally spaced, lattice-based, channel equalization scheme but also the performance improvement obtainable using the novel cooperative adaptive scheme.
Receiver hyperparameters are parameters set prior to the receiver’s operation and critically influence the performance of underwater acoustic communication systems (UAC). They include detection thresholds, filter lengths and cutoff frequencies, equalization architecture, learning rates, phase-locked loop (PLL) gains, etc. Selecting these parameters is challenging because their optimal values depend strongly on the channel state, which fluctuates rapidly in underwater environments. Traditional static hyperparameter selection methods become suboptimal under dynamic conditions. This article introduces an online Bayesian optimization (BO) framework for real-time adaptive tuning of receiver hyperparameters in UAC systems. The proposed approach leverages a time-varying Gaussian process (TV-GP) model to dynamically track optimal hyperparameters, and adapt to channel variations at a packet-level resolution. In addition, the proposed method leverages two coordinated BO agents: 1) an exploratory “Scout;” and 2) a greedy “Wise.” The resulting Wise-and-Scout (WandS) algorithm balances exploration to follow fast channel changes with exploitation to select high-confidence, high-performance hyperparameter values. As a proof of concept, we apply the method to a decision-feedback equalizer (DFE) with a PLL. We introduce a selection policy to avoid unstable hyperparameter regions, called “hallucinations.” These regions can make the DFE diverge, even when the output signal appears to have a high signal-to-noise ratio. The proposed method is validated using replay of channels measured at sea, as well as simulations based on Bellhop ray-tracing and maximum entropy channel models. Results demonstrate that the adaptive tuning approach achieves lower packet error rates compared to channel-estimation-based heuristics and fixed hyperparameter configurations across diverse channel conditions.
Underwater gliders are vital for sustained ocean monitoring, yet limited satellite communication and reliance on skilled pilots constrain fleet scalability. Automated anomaly detection can reduce operator burden and improve mission reliability. To address this challenge, this study proposes a spatio-temporal graph neural network (STGNN) adapted for underwater glider telemetry. The model treats each sensor as a graph node and couples a fixed intersensor structure with temporal dynamics to detect anomalous windows within the sensor channels. Its performance is then contextualized through a comparative evaluation against deep and classical baselines under a pooled cross-mission protocol with configurable manual mission splits, together with principled threshold calibration through quantile, rolling-statistics, and extreme value strategies. It also has explicit latency and false-alarm reporting. Experiments conducted on Slocum G2 deployments with synthetic fault injection show that the proposed STGNN is the strongest overall model in the evaluation suite with an F1 score of 0.986. The codebase and evaluation protocol are provided for general use.
An autonomous underwater vehicle (AUV) is an unsupervised, self-powered underwater robotic platform programmed to operate without human intervention. Autonomous underwater gliders (AUGs) are a special class of AUV, engineered for extended, energy-efficient ocean observation. These gliders use variable buoyancy for propulsion. The buoyancy change initiates a heave motion, and the attached wings create a pitching moment due to lift and drag forces acting on them. This interaction converts the vertical heave motion into forward horizontal motion, enabling three degrees of freedom: surge, heave, and pitch. Various factors, such as the design and efficiency of the buoyancy engine, power management, and control systems, influence the AUGs' performance. Optimizing these factors is crucial to enhancing glider performance in complex underwater environments and ensuring long-duration, reliable missions. This article presents an optimization strategy for buoyancy control to enhance the gliding efficiency, thereby improving the underwater glider's overall performance, energy efficiency, and endurance. The model dynamics of the RoBuoy glider platform, developed in the Robotics Laboratory at the Indian Institute of Technology Madras, were used for this study. A comparative analysis is conducted for the symmetrical and asymmetrical actuation of the linear actuators used for buoyancy control, focusing on glider parameters, such as glide velocity, pitch angle, surge velocity, and angle of attack, to optimize buoyancy for mission-specific requirements. An energy model for calculating the energy consumed by the buoyancy engine is formulated. In addition, different optimization scenarios are explored to assess energy consumption and enhance the overall energy efficiency of the underwater glider. The results from the analysis show that the glider's energy consumption can be reduced up to 40% depending on the mission.
Inspection, Maintenance, and Repair tasks on underwater structures are risky, costly, and complex, typically requiring professional divers or remotely operated vehicles. This work presents a framework for autonomous object grasping without relying on visual markers such as ArUco tags. The proposed solution uses an Intervention Autonomous Underwater Vehicle equipped with a multicamera system and a deep learning vision module. The neural network model was trained with images of target objects captured in simulation and dry environments rather than underwater conditions. Using these pretrained networks, the robot can accurately detect and grasp objects underwater despite the differences between training and operational environments. The framework also integrates a natural language processing module powered by a Large Language Model within the Robot Operating System (ROS) through ChatGPT’s API. This enables intuitive human–robot interaction, allowing operators to issue manipulation commands in natural language specifying target objects. The ROS-based intelligent agent interprets these commands and launches the autonomous behaviors required for precise grasping tasks. Experiments conducted in simulation and controlled water tank environments demonstrate the effectiveness and adaptability of the proposed system for underwater manipulation.
Inertial navigation systems are specialized navigation apparatuses that equip almost all autonomous underwater and surface vehicles. They require precise initial alignment, i.e., determination of their initial attitude, which is typically achieved: (a) in quasi-static conditions (whenever possible); and (b) in two stages: Coarse Alignment (CA), using methods like TRI-axis Attitude Determination (TRIAD), and Fine Alignment (FA), via Zero Velocity Update (ZVU)-based Extended Kalman Filtering (EKF). However, conventional methods suffer from slow convergence and limited bias estimability. In response, this paper introduces: (a) an optimized version of a recently proposed CA method, namely, TRIAD with Coarse Bias Estimation (TRIAD-CBE); and (b) a novel FA EKF observation model that incorporates Non-Orthonormality (NON) error constraints derived from TRIAD, directly linking these errors to the inertial sensor biases. As validated through extensive Monte Carlo (MC) simulations, as well as real-world experiments using two Inertial Measurement Units (IMUs) of different grades, our approaches substantially accelerate the convergence of misalignment and bias estimates (from minutes to seconds), while maintaining accuracy/precision comparable to traditional techniques.
