In the navigation of unmanned surface vehicles (USVs), various types of obstacles may be encountered, which can be categorized into real-time collision avoidance among multiple USVs and obstacle avoidance between USVs and other obstacles. Most existing autonomous obstacle avoidance algorithms do not account for the nonlinear motion characteristics of USVs, often resulting in non-compliance with the International Regulations for Preventing Collisions at Sea (COLREGs) and a tendency to fall into local optima. To address these issues, this paper proposes a path planning algorithm that integrates the dynamic window approach (DWA) considering nonlinear characteristics with COLREGs, making the USV's motion trajectory more applicable to practical engineering scenarios. A kinematic mathematical model is established based on the motion characteristics of USVs, and an evaluation function for the optimal path is constructed using DWA. The fully informed search algorithm (FISA) is employed to select the optimal set of velocities and steering angles from the velocity sampling set, based on different cost calculation methods. The USVs use a laser radar for local obstacle detection, enabling real-time dynamic obstacle avoidance. To address the real-time collision avoidance problem among multiple USVs in open waters, the algorithm filters out COLREGs-compliant avoidance maneuvers during path planning. The correctness and feasibility of the fusion algorithm were verified through comparative simulations. In the simulated environment model, the influence of ocean currents on the USV was introduced, and multiple sets of experiments under different conditions were conducted to compare the motion trajectories, average travel distances, and average travel times of the USV. The simulation results indicate that the USV can perform accurate obstacle avoidance when encountering various types of obstacles. Compared to the traditional DWA algorithm, the proposed approach demonstrates advantages in terms of travel distance and travel time, while still achieving effective obstacle avoidance.

This paper presents a novel trajectory tracking controller for an underactuated unmanned surface vehicle (USV). The controller incorporates an event-triggered extended state observer (ETESO), a minimum learning parameter neural network, an integral non-singular terminal sliding mode (INTSM) control strategy, and a dynamic event-triggered mechanism (DETM). Firstly, an ETESO is developed to estimate unmeasurable velocities and lumped disturbances, differentiating it from most existing extended state observers without the necessity for real-time output measurements. To further alleviate the communication burden and minimize actuator wear, a DETM with an adjustable threshold is introduced. In contrast to traditional event-triggered methods, which employ fixed threshold parameters, this mechanism allows for online adaptive updates of the triggering thresholds, thereby enhancing resource efficiency. Additionally, an INTSM is designed to ensure rapid convergence of the position and velocity errors of the USV. To effectively counteract external disturbances and internal modeling uncertainties, a minimum learning parameter (MLP) neural network algorithm is implemented to approximate and compensate for these uncertainties. Finally, using Lyapunov's theory, it is demonstrated that all signals within the closed-loop tracking control system remain bounded. Simulation results are given to illustrate the effectiveness of theoretical results.

Aiming at the problems of the poor real-time detection ability of smart devices for underwater targets and the difficulty of deploying complex model algorithms on embedded devices with limited resources, this paper proposed a lightweight feature extraction module GGS, which combines traditional algorithm downsampling with deep separable convolution downsampling and introduces a parameter-free attention mechanism to extract features from input data at multiple scales and focus on target information. The paper constructs the GGS-PF-YOLOv5 network by replacing the backbone extraction network in YOLOv5 with the GGS module and the GGS-PANET by combining the GGS module with depth separable convolution for feature fusion, the PfAAMLayer non-parameter attention mechanism is introduced to enhance the feature extraction capabilities of the model by focusing on the feature information of the two dimensions of channel and space, improving the identification accuracy of sea cucumbers while keeping the number of network parameters low. Experimental results show that the proposed network, with remarkably few network parameters, outperforms the original YOLOv5 model regarding detection speed while achieving comparable accuracy. The GGS-PF-YOLOv5 model reduces the parameter volume by 95% compared to YOLOv5s source code while doubling detection speed, with a weight file size of only 1.08M, 92% smaller than the source code. The (Map50) only decreases by 1.5% compared to YOLOv5s, which indicates that this paper proposed model can achieve real-time target detection on low-power embedded devices while maintaining high levels of accuracy.

Addressing the challenges of strong coupling, nonlinearity, and time-varying characteristics in underwater manipulator systems, along with the inevitable environmental disturbances such as gravity, buoyancy, and hydrodynamic forces that reduce system state convergence speed and affect control performance, an underwater manipulator dynamic model integrating the Lagrange equation, Morison equation, and nonlinear friction forces is built. A linear active disturbance rejection control (LADRC) strategy incorporating a hierarchical cascade linear extended state observer (HCLESO) is proposed. According to LADRC theory, the underwater manipulator system is approximated as an independent subsystem, which reduces the coupling effect between each joint of the manipulator arm. To address the incomplete estimation issue of the Linear Extended State Observer (LESO), a HCLESO is designed to estimate the total system disturbance, suppress high-frequency measurement noise, and reduce the disturbance observation residual, thereby improving the control precision and robustness of the underwater manipulator. The simulation results indicate that the proposed control strategy significantly improves system performance in terms of smaller control inputs, reduced overshoot, and lower steady-state errors. Compared to conventional PD control, the proposed strategy exhibits superior control accuracy and faster response. Compared to the standard LADRC, the introduction of HCLESO provides notable enhancements. With HCLESO ( p = 2), the IAE for joints q ₁ and q ₂ is reduced by 50.1% and 64.0%, respectively, while the ITAE decreases by 36.2% and 50.0%. Further improvements are observed with HCLESO ( p = 3), where the IAE for joints q ₁ and q ₂ is reduced by 50.1% and 65.1%, respectively, and the ITAE by 46.2% and 57.8%.

For marine ranching, efficiently and accurately detecting, tracking, and enumeration of benthic organisms can help farmers understand the growth and population changes of marine products, avoid high-risk tasks, and analyze changes in the marine ecological environment. To address the problems of target occlusion, low detection accuracy, and numerous small targets in existing marine organism detection models in complex seabed environments, an improved YOLOv5+DeepSORT algorithm for detecting and tracking benthic organisms is proposed. This algorithm integrates the Global Context Block attention mechanism with the BottleneckCSP module to form a new BottleneckCSPGC module, enhancing feature extraction capabilities. Replace the original loss function with the Normalized Wasserstein Distance (NWD) loss function to improve the detection accuracy of small targets. Finally, experimental results show that the accuracy on the underwater dataset reached 87.1% mAP@0.5 and 53.3% mAP@0.5:0.95, which are 1.8% and 4.0% higher than YOLOv5, respectively. The use of DeepSORT for tracking and counting provides technical support for marine ranching supervision.