A Smart Navigation Control Method and System Based on an Unmanned Fishery Research Vessel

Abstract

This application relates to the field of intelligent navigation control technology, and more particularly to an intelligent navigation control method and system for an unmanned research vessel used in fisheries. The method acquires environmental parameters and hull operating status parameters of the operating water area, and utilizes a multi-sensor fusion algorithm to achieve real-time perception of the environment and hull status, generating a navigation status information set. A surface obstacle detection model is established based on a deep semantic segmentation algorithm to achieve obstacle recognition. Combining the obstacle detection results and navigation status information, an improved depth-deterministic strategy gradient algorithm is used for path planning to generate the optimal navigation path. Finally, a model predictive control algorithm is used to decouple speed and heading control, and precise navigation control is achieved through dynamic adjustment of propeller speed and rudder angle. This invention realizes intelligent autonomous navigation and high-precision path tracking for the unmanned research vessel, improves its obstacle avoidance capability and energy efficiency in complex waters, and enhances operational safety and mission execution efficiency.

Description

Technical Field

[0001] This application relates to the field of intelligent navigation control technology, and in particular to an intelligent navigation control method and system based on an unmanned research vessel for fisheries. Background Technology

[0002] In related technologies, with the development of marine scientific research and intelligent fisheries, unmanned research vessels (UAVs) have become an important platform for conducting fisheries resource surveys, environmental monitoring, and water quality analysis. Compared with traditional manned research vessels, UAVs have advantages such as small size, low cost, maneuverability, high speed, high energy efficiency, long endurance, and adaptability to harsh environments. They can perform multi-source data acquisition tasks in various complex environments, including nearshore, offshore, and inland waterways. By carrying a multi-sensor system, they can achieve real-time monitoring of multi-dimensional parameters such as meteorology, wind speed and direction, ultraviolet radiation intensity, temperature and humidity, gas concentration, and underwater acoustic signals, providing data support for fisheries resource assessment and environmental protection.

[0003] However, existing unmanned research vessels for fisheries still face multiple challenges in autonomous navigation and mission execution in complex aquatic environments. First, these vessels often operate in waters with densely distributed obstacles and strong environmental disturbances, such as randomly distributed islands, aquaculture areas, and floating debris. Traditional sonar obstacle avoidance methods have limited accuracy in obstacle identification, easily leading to misjudgments or path planning failures, making it difficult to guarantee navigation safety and real-time performance. Second, the operating range of a single research vessel is limited, preventing efficient coverage of wide-area data. Furthermore, they often rely on single sensors or simple weighted fusion for environmental perception, resulting in low research efficiency and long mission execution cycles. Third, existing path planning methods mostly employ fixed heuristic algorithms or rule-based planning strategies, lacking adaptability to environmental changes and vessel status, making it difficult to achieve comprehensive optimal path planning that balances navigation safety, energy consumption, and time optimization.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0005] The main objective of this application is to propose an intelligent navigation control method and system based on an unmanned research vessel for fisheries, which can realize intelligent autonomous navigation and high-precision path tracking functions of the unmanned research vessel, improve the obstacle avoidance ability of the unmanned research vessel in complex waters and reduce energy consumption.

[0006] To achieve the above objectives, one aspect of this application proposes an intelligent navigation control method based on an unmanned research vessel for fisheries, the method comprising the following steps:

[0007] Based on a multi-sensor fusion algorithm, the environmental parameters and hull operation status parameters of the fishery unmanned scientific research vessel in the target operating waters are perceived and features are extracted in real time to generate a navigation status information set.

[0008] The navigation status information set is semantically segmented, and obstacle detection results are output.

[0009] Based on the obstacle detection results, the navigation path of the fishery unmanned research vessel is optimized using an improved depth deterministic strategy gradient algorithm, and the optimal navigation path is output.

[0010] Based on the optimal navigation path, a model predictive control algorithm is used to decouple the speed and heading control, and output speed control commands and heading control commands.

[0011] The real-time navigation status of the unmanned fishery research vessel is controlled according to the speed control command and the heading control command.

[0012] In some embodiments, the navigation status information set includes wind speed, wind direction, wave height, surface current parameters, ultraviolet radiation intensity, temperature and humidity, speed information, heading information, and hull attitude angle information.

[0013] In some embodiments, the semantic segmentation of the navigation state information set and the output of obstacle detection results include:

[0014] The navigation status information set is preprocessed to obtain an image sequence containing water surface areas and obstacle areas;

[0015] Construct a deep semantic segmentation network model that includes encoding and decoding layers;

[0016] The image sequence is semantically segmented using a deep semantic segmentation network model to generate a semantic segmentation map.

[0017] Based on the semantic segmentation map, the spatial location and distribution density of obstacles in the hull coordinate system are calculated using a spatial clustering algorithm to form obstacle detection results; the obstacle detection results include the type, location and distribution information of obstacles on the water surface.

[0018] In some embodiments, the encoding layer uses MobileNetV2 as the backbone network and extracts multi-scale contextual features by combining convolution and dilated convolution; the decoding layer performs boundary information recovery and pixel classification reconstruction based on the deconvolution structure to generate a semantic segmentation map.

[0019] In some embodiments, the step of optimizing the navigation path of the fishery unmanned research vessel based on the obstacle detection results and using an improved depth deterministic policy gradient algorithm to output the optimal navigation path includes:

[0020] Based on the obstacle detection results and the navigation status information set, a state-action mapping set is generated; the state-action mapping set is used to describe the state space and action space that correspond to the hull operation state and control actions.

[0021] Based on the state-action mapping set, a path planning model based on the improved deep deterministic policy gradient algorithm is constructed.

[0022] During the training process of the path planning model, a priority experience replay mechanism based on path importance scoring is adopted to update the experience samples with weights, thereby obtaining a priority experience sample set.

[0023] An attention mechanism is introduced based on the priority experience sample set to weight and fuse state features and action features to generate a fused feature tensor.

[0024] Based on the fused feature tensor, a reward function is constructed, and the path planning model is iteratively optimized using a deterministic policy gradient update rule to obtain the optimal policy parameters. The reward function includes a range cost term and an energy consumption cost term, which are used to comprehensively evaluate the travel distance and propulsion energy consumption.

[0025] The optimal navigation path is generated based on the path planning model corresponding to the optimal strategy parameters.

[0026] In some embodiments, the path planning model includes a policy network and a value network; the policy network is used to output continuous control actions to generate candidate paths, and the value network is used to receive state-action pair inputs, calculate the corresponding action value function, and output the path value evaluation result.

[0027] In some embodiments, the step of decoupling speed and heading control using a model predictive control algorithm based on the optimal navigation path and outputting speed control commands and heading control commands includes:

[0028] Based on the optimal navigation path, a hull optimization model based on the model predictive control algorithm is constructed;

[0029] Based on the aforementioned hull optimization model, a speed prediction sub-model and a heading prediction sub-model are constructed.

[0030] In the joint control phase of the speed prediction sub-model and the speed prediction sub-model, a speed-heading decoupled control strategy is adopted to coordinate and optimize longitudinal propulsion and lateral maneuvering, and generate a control increment sequence.

[0031] Based on the control increment sequence, the propeller speed and rudder angle input are rolled and corrected, and the speed control command and heading control command are output.

[0032] In some embodiments, the hull optimization model utilizes the ship dynamics equations and performs rolling calculations on the six degrees of freedom of the ship using the fourth-order Runge-Kutta method, forming a set of predicted hull motion states based on the rolling calculation results; the six degrees of freedom states include hull position, velocity, acceleration, bow angle, angular velocity, and angular acceleration.

[0033] In some embodiments, the speed prediction sub-model is used to predict the thrust-speed response relationship in the longitudinal direction, and the speed control sequence is obtained by rolling solution by combining the thrust power constraint and the propeller speed change constraint.

[0034] The heading prediction sub-model is used to predict the heading angle-rudder angle response relationship in the lateral direction. Combining rudder angle constraints and lateral deviation constraints, the heading correction sequence is obtained by rolling solution.

[0035] To achieve the above objectives, another aspect of this application proposes an intelligent navigation control system based on an unmanned research vessel for fisheries, the system comprising:

[0036] The data acquisition module is used to obtain environmental parameters and vessel operating status parameters of the target operating area;

[0037] The data fusion processing module is used to perform real-time perception and feature extraction of environmental parameters and hull operating status based on multi-sensor fusion algorithms, and generate a navigation status information set.

[0038] The surface obstacle detection module is used to establish a surface obstacle detection model using a deep semantic segmentation algorithm, perform semantic segmentation on the navigation state information set, and output obstacle detection results.

[0039] The path planning module is used to establish a path planning model based on the obstacle detection results and the navigation state information set, optimize the navigation path of the hull, and output the optimal navigation path.

[0040] The navigation control module is used to decouple the speed and heading based on the optimal navigation path using a model predictive control algorithm, and output speed control commands and heading control commands.

[0041] The execution control module is used to output the real-time navigation control status of the fishery unmanned research vessel according to the speed control command and the heading control command.

[0042] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0043] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0044] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0045] The embodiments of this application include at least the following beneficial effects: This application provides an intelligent navigation control method and system based on an unmanned research vessel for fisheries. By constructing an intelligent navigation control system based on multi-sensor fusion, this application realizes the autonomous perception and stable control of the unmanned research vessel in complex water environments. By fusing environmental parameters and hull operating state parameters, and performing semantic segmentation on the navigation state information set to obtain obstacle detection results, the obstacle avoidance capability of the unmanned research vessel during autonomous navigation can be improved. Then, a path planning model based on the improved Deep Deterministic Policy Gradient (DDPG) algorithm is adopted to optimize the navigation path of the unmanned research vessel by combining the obstacle detection results. This enables the global policy optimization function of the hull navigation path, improves path tracking capability and tracking accuracy, and reduces navigation energy consumption. At the same time, the introduction of the Model Predictive Control (MPC) algorithm to decouple speed and heading control effectively suppresses track deviation and attitude oscillation caused by coupling interference, thereby improving navigation stability and control accuracy. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating an intelligent navigation control method based on an unmanned research vessel for fisheries, provided in an embodiment of this application.

[0047] Figure 2 This is a schematic diagram of the structure of a fishery unmanned research vessel that integrates atmospheric resource detection and intelligent cruise.

[0048] Figure 3 This is a schematic diagram of the sensor system distribution of a fishery unmanned research vessel that integrates atmospheric resource detection and intelligent cruise.

[0049] Figure 4 yes Figure 1 A flowchart illustrating step S3 in the process;

[0050] Figure 5 yes Figure 1 A flowchart illustrating step S4 in the process;

[0051] Figure 6 This is a schematic diagram of the structure of an intelligent navigation control system based on an unmanned research vessel for fisheries, provided in an embodiment of this application;

[0052] Attached figures: 101. Propeller; 102. Counterweight compartment; 103. Main hatch; 104. Main compartment; 105. Appendages; 106. Wind speed sensor; 107. GPS; 108. Wind direction sensor; 109. Ultraviolet and air quality sensors, etc.; 110. Bottom-mounted sonar. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0054] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0055] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0057] This application provides an intelligent navigation control method and system for an unmanned research vessel used in fisheries. This solution constructs an intelligent navigation control system based on multi-sensor fusion, enabling the unmanned research vessel to achieve autonomous perception and stable control in complex aquatic environments. By fusing environmental parameters and vessel operating status information, it can comprehensively acquire multi-source data such as wind speed, wave height, water flow speed, and vessel attitude, significantly improving environmental perception accuracy and dynamic response capabilities. A surface obstacle detection model established using a deep semantic segmentation algorithm can accurately identify and segment various types of obstacles, improving the real-time performance and reliability of obstacle avoidance and avoiding recognition errors caused by visual noise or changes in lighting. A path planning model based on the improved Deep Deterministic Policy Gradient (DDPG) algorithm, combined with navigation status information and obstacle detection results, achieves global policy optimization of the vessel's navigation path, resulting in an optimal balance between safety, energy consumption, and travel time. Meanwhile, the introduction of model predictive control (MPC) algorithm to decouple speed and heading control, through propeller speed adjustment and dynamic rudder angle correction, effectively suppresses track deviation and attitude oscillation caused by coupling interference, and significantly improves navigation stability and control accuracy.

[0058] This application provides an intelligent navigation control method and system based on an unmanned research vessel for fisheries, relating to the field of intelligent navigation control technology. The intelligent navigation control method and system based on an unmanned research vessel for fisheries provided in this application can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster composed of multiple physical servers, or a distributed system. It can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing an intelligent navigation control method and system based on an unmanned research vessel for fisheries, but is not limited to the above forms.

[0059] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0060] Please see Figure 1-6 As shown, this invention relates to an intelligent navigation control method and system based on an unmanned research vessel for fisheries.

[0061] Figure 1 This is an optional flowchart of an intelligent navigation control method based on an unmanned research vessel for fisheries, provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S1 to S6:

[0062] S1: Based on the multi-sensor fusion algorithm, the environmental parameters and hull operation status parameters of the fishery unmanned scientific research vessel in the target operating water area are perceived and features are extracted in real time to generate a navigation status information set; among which, the navigation status information set includes wind speed, wind direction, wave height, water surface current parameters, ultraviolet intensity, temperature and humidity, speed information, heading information and hull attitude angle information.

[0063] In this embodiment, this application provides a Figure 2 and Figure 3 The image shows an unmanned research vessel for fisheries that integrates atmospheric resource monitoring and intelligent navigation.

[0064] like Figure 2As shown, a fishery unmanned research vessel integrating atmospheric resource detection and intelligent navigation mainly includes a propeller 101, a ballast tank 102, a main hatch 103, a main compartment 104 (which mainly includes energy units and control equipment), and appendages 105. The hull adopts a catamaran structure design, consisting of two hulls, fixed together by a connecting frame. The overall length is approximately 4 meters, with a length-to-width ratio of 9 and a spacing ratio of 0.55. The catamaran structure provides excellent stability and wave resistance, effectively reducing lateral rolling and improving attitude maintenance under wind and wave conditions. Each hull is equipped with an OT80 series 8kg thruster at the stern, with an input power of 360W, saving approximately 25% energy compared to traditional thrusters, providing a sufficient energy foundation for subsequent multi-sensor collaborative sensing and data processing.

[0065] The fore and aft hulls each have ballast tanks 102, used to adjust the overall center of gravity, ensuring lateral and longitudinal balance under varying loads and sea conditions. The main compartment 104, located below the main hatch 103, is the core control and power supply area, housing an energy supply unit, battery management unit, and data processing unit. The energy supply unit provides a stable power supply to the host computer, slave computer, and sensor systems. The host computer uses a Jetson Orin Nano development board, and the slave computer uses an STM32F407VET6 chip, with high-speed data exchange via CAN bus and UART.

[0066] like Figure 2 As shown, appendages 105 are also installed on both sides of the hull. These appendages are detachable floating structures with a bottom made of lightweight waterproof material and an internal cavity structure. Appendages 105 are fixed to the main hull via connecting frames to enhance the hull's lateral stability and reduce the impact of waves on the main hull's pitch and roll. In practical applications, the appendages can also serve as sensor mounting platforms for deploying lateral current sensors or auxiliary camera modules, giving the unmanned surface vessel a wider field of view and higher measurement accuracy during missions. The presence of the appendages allows the entire vessel to maintain its attitude balance when operating in high-wave, high-altitude areas, ensuring the upper sensor array remains horizontal, thereby improving the reliability of environmental parameter measurements.

[0067] Figure 3 This is a schematic diagram showing the distribution of the sensor system on the hull. From... Figure 3It is known that the wind speed sensor 106 and the wind direction sensor 108 are respectively installed on the central gimbal of the upper part of the hull, and are used to measure the wind speed and wind direction of the target operating water area in real time; the GPS 107 is fixed at the center of the top of the hull, providing high-precision navigation position and time synchronization information; the ultraviolet and air quality sensor 109 is deployed at the front of the top of the hull, and is used to detect parameters such as ultraviolet intensity, gas concentration in the air and ambient temperature and humidity; the sonar probe 110 is installed at the bottom of the hull, and is used to detect underwater obstacles, trawl nets, buoy supports and other structures to assist in underwater obstacle avoidance; in addition, the hull attitude measurement unit (IMU) is integrated in the main cabin, and is used to acquire speed, heading and attitude angle information, including pitch angle and roll angle.

[0068] All sensor signals were acquired by an STM32 F407VET6 lower-level computer, and an analog-to-digital converter was used to convert the analog signals into digital signals. To ensure data accuracy and time synchronization, a timestamp synchronization module was set up at the acquisition end, using GPS timing signals to correct the sampling clock of each sensor, ensuring that the environmental conditions corresponding to various data at the same time are consistent. The sampling period was set to 20 ms to meet the dynamic change response requirements under complex sea conditions.

[0069] In the data processing flow, the STM32 performs low-pass filtering and noise suppression on the acquired raw signal, and then sends the preprocessed data to the host computer, Jetson Orin Nano, via serial port. The host computer runs a multi-threaded data management program under the ROS (Robot Operating System) framework, which performs time synchronization, formatting and caching of data from different sensors, and establishes a unified environmental parameter data structure.

[0070] Ultimately, the obtained environmental parameters include wind speed, wind direction, wave height, water surface velocity, ultraviolet radiation intensity, temperature, and humidity; the hull operating parameters include speed, heading, attitude angles (pitch and roll), thruster speed, and rudder position. All parameters form an environmental perception input set, which serves as the input source for subsequent multi-sensor fusion algorithms.

[0071] Through the above design, the fishery unmanned research vessel can conduct high-frequency and high-precision data acquisition in the target operating area under complex sea conditions, and realize real-time monitoring of the environmental conditions and the vessel's condition. Figure 2 and Figure 3 The structural layout design ensures the symmetry of the sensor installation positions and the unobstructed field of view, thereby effectively improving the perception accuracy and reliability of environmental parameters and providing high-quality data support for subsequent deep semantic segmentation and path planning modules.

[0072] In this embodiment, the unmanned research vessel for fisheries utilizes multi-source sensors installed on its top and sides to achieve multi-dimensional perception of environmental elements in the target operating waters and the vessel's own condition. Specifically, the multi-source sensors include:

[0073] The system includes wind speed and direction sensors, ultraviolet and air quality sensors, temperature and humidity sensors, gas concentration sensors, a GPS module, an inertial measurement unit (IMU), and sonar sensors mounted on the hull. The sampling frequency and accuracy of each sensor are configured differently. Environmental sensors are set to a sampling frequency of 1Hz to 5Hz for continuous monitoring of external environmental changes; motion state sensors (such as the IMU and GPS) are set to a sampling frequency of 20Hz to ensure real-time capture of speed and attitude changes.

[0074] To ensure data stability under complex sea conditions and high reflectivity, this embodiment utilizes the STM32F407VET6 chip for first-layer data preprocessing, including:

[0075] (1) Signal denoising and drift correction: random noise in sensor output is suppressed by moving mean filtering and adaptive Kalman filtering algorithms;

[0076] (2) Time synchronization and coordinate registration: GPS timestamps are used to align the output data of multiple sensors on a unified time axis and map all sensor data to the hull coordinate system.

[0077] Subsequently, the processed data is transmitted to the Jetson Orin Nano development board via the CAN bus, entering the second-level fusion processing stage. This stage employs a multi-sensor fusion mechanism based on weighted confidence: a weight factor is assigned to each type of sensor, and the weights are dynamically adjusted according to their historical stability and noise characteristics; the optimal estimate of the multi-source observation results is calculated through the data fusion engine to achieve unified modeling of the environmental state and the hull state; anomaly detection is performed on sudden signals, and when any sensor data exceeds a reasonable range, a redundancy correction mechanism is automatically triggered, with other sensor data replacing and compensating to ensure the continuity and robustness of the overall perception results.

[0078] Ultimately, the navigation status information set generated in this embodiment includes fused and corrected multi-dimensional parameters, including but not limited to:

[0079] Environmental parameters: wind speed, wind direction, wave height, water surface velocity, ultraviolet radiation intensity, temperature, humidity, air quality index;

[0080] Operating parameters: speed, heading angle, pitch angle, roll angle, longitudinal acceleration, rate of change of attitude angle, etc.

[0081] This navigation status information set is stored in a shared memory area using a unified data structure and is fully refreshed every 50ms sampling period. This information set is used not only as input to the subsequent obstacle detection model (S3) but also as input to the path planning model (S4), providing environmental context features and dynamic hull status.

[0082] Through the aforementioned fusion algorithm, this embodiment can achieve real-time perception of meteorological, fluid, and attitude changes under strong interference conditions on the sea surface, significantly improving data reliability and environmental perception accuracy, thus providing a high-precision input basis for subsequent intelligent navigation control decisions.

[0083] S2: Perform semantic segmentation on the navigation status information set and output obstacle detection results;

[0084] This involves semantic segmentation of the navigation status information set and outputting obstacle detection results, including but not limited to the following steps:

[0085] S21: Preprocess the navigation status information set to obtain an image sequence containing water surface areas and obstacle areas;

[0086] In this embodiment, a high-definition camera fixed to a gimbal above the hull is used to collect real-time image sequences of the operating water area. The frame rate is set to 30 frames per second, and the resolution is 1280×720 pixels.

[0087] The acquired raw images undergo initial processing by an STM32 lower-level computer, including time synchronization, white balance correction, and brightness equalization; they are then uploaded to a Jetson Orin Nano upper-level computer for further preprocessing.

[0088] Noise reduction: Gaussian filtering and adaptive median filtering are used to remove noise caused by wave reflection or sunlight highlights;

[0089] Color enhancement: The HSV spatial enhancement algorithm is used to improve the contrast between the water surface and obstacles in the image, avoiding edge blurring caused by changes in lighting;

[0090] Region cropping and labeling: Based on the GPS and heading angle information in the navigation status information set, the image region is projected onto the ship coordinate system, and the ROI region containing the main navigation field of view is selected.

[0091] The image sequence after the above preprocessing retains typical obstacle areas such as water surface, buoys, floating garbage, and aquaculture cages, which constitute the input samples for the subsequent semantic segmentation network.

[0092] S22: Construct a deep semantic segmentation network model that includes an encoding layer and a decoding layer;

[0093] S23: Perform semantic segmentation on image sequences using a deep semantic segmentation network model to generate semantic segmentation maps;

[0094] The encoding layer uses MobileNetV2 as the backbone network and extracts multi-scale contextual features by combining convolution and dilated convolution; the decoding layer uses a deconvolution structure to restore boundary information and reconstruct pixel classification to generate a semantic segmentation map.

[0095] In this embodiment, a water surface obstacle detection model is established using the DeeplabV3+ architecture, the core of which consists of an encoding layer, a feature fusion layer, and a decoding layer.

[0096] Encoder: Using MobileNetV2 as the backbone network, after inputting the image, it extracts texture and edge features at different scales through multi-layer convolution and batch normalization; it introduces dilated convolution structure with dilation rates set to 6, 12 and 18 respectively to expand the receptive field and retain multi-scale contextual information.

[0097] Feature Fusion Layer (ASPP Module): Performs hollow spatial pyramid pooling on the high-dimensional feature map of the encoded output to fuse global information at different scales, ensuring the ability to perceive obstacles of both large and small sizes.

[0098] Decoder: Employs a deconvolution and skip connection structure to fuse high-level semantic features with low-level detail features. It recovers boundary information through pixel-level upsampling, achieving refined segmentation output.

[0099] During model training, iterative training was performed using a manually annotated obstacle image dataset (containing samples from various scenes such as strong sunlight, high reflectivity, surge, and cloudy days). The loss function adopted was a joint loss function of cross-entropy and boundary smoothing, and the optimizer used was the Adam algorithm. After training, the model weight file was deployed on Jetson Orin Nano for real-time inference, outputting the semantic segmentation map corresponding to each frame.

[0100] This semantic segmentation map uses color coding to identify different categories of pixels, such as blue for water surfaces, red for floating debris, yellow for buoys, and green for aquaculture cages, thus visually distinguishing the spatial distribution of different types of obstacles.

[0101] S24: Based on the semantic segmentation map, use a spatial clustering algorithm to calculate the spatial location and distribution density of obstacles in the hull coordinate system to form obstacle detection results; among which, the obstacle detection results include the type, location and distribution information of obstacles on the water surface.

[0102] In this embodiment, the transformation matrix between the camera optical axis and the hull coordinate system is calculated using the attitude angle information (roll angle, pitch angle) provided by the IMU and the GPS position coordinates, thereby realizing the mapping from image pixel coordinates to actual water surface plane coordinates.

[0103] If the hull is equipped with a binocular camera or a structured light ranging module, the depth information of the obstacle can be calculated based on the parallax map; if there is no depth hardware, optical flow estimation is performed using the relative displacement and speed data of multiple frames of images to obtain the motion trend and distance change of the obstacle relative to the hull.

[0104] The spatially mapped set of obstacle pixels is input into a density clustering algorithm for cluster analysis.

[0105] After clustering, the following parameters are calculated for each obstacle: center position (centroid coordinates), with the hull as the origin representing the (x, y) position; geometric boundary, extracted using the minimum bounding rectangle algorithm; area weight, representing the obstacle size by pixel proportion; and confidence score, calculated by combining segmentation probability and time stability.

[0106] Simultaneously, a time sliding window mechanism is adopted to smooth and fuse the obstacle detection results of several consecutive frames, and filter out instantaneous noise targets (such as wave crest reflections) to obtain a stable obstacle distribution map.

[0107] After clustering, a two-dimensional obstacle density matrix is ​​constructed based on the spatial distribution of obstacles in the hull coordinate system, representing the obstacle distribution on the water surface in grid form. This matrix is ​​input into the path planning module and used as an important weighting factor in the obstacle avoidance cost function.

[0108] The final obstacle detection results include:

[0109] Obstacle types: Based on semantic tags, they are categorized into floating debris, buoys, fishing nets, islands, etc.

[0110] Spatial position: Output three-dimensional coordinates (x, y, z) in the hull coordinate system;

[0111] Distribution density and motion trend: Obstacle density surface and motion vector field are generated from the clustering results of consecutive frames for dynamic obstacle avoidance prediction.

[0112] S3: Based on the obstacle detection results, optimize the navigation path of the fishery unmanned research vessel using the improved depth deterministic policy gradient algorithm, and output the optimal navigation path.

[0113] Among them, reference Figure 4 As shown, based on obstacle detection results, the navigation path of the fishery unmanned research vessel is optimized using an improved depth-deterministic policy gradient algorithm, outputting the optimal navigation path, including:

[0114] S31: Based on the obstacle detection results and navigation status information set, generate a state-action mapping set; wherein, the state-action mapping set is used to describe the state space and action space of the correspondence between the hull operation state and the control action.

[0115] In this embodiment, obstacle detection results are fused with navigation status information to construct a complete state input space.

[0116] The state space includes: the hull's current position coordinates, heading angle, speed, attitude angle, surrounding obstacle distribution matrix, water current velocity vector, wind direction, wave height, and target point position. This state set describes the hull's comprehensive dynamic state in the surface water environment.

[0117] The Action Space includes thruster speed and rudder angle adjustments, both of which are continuously adjustable parameters used to control the forward speed and turning radius of the hull.

[0118] Accordingly, the action space is defined as a set of continuous control variables for the hull, including propulsion control variables (propeller speed change rate) and rudder angle control variables (steering angle change value). State vectors are paired with action vectors to generate a state-action mapping set. This set describes the hull's response strategies under different environmental conditions and serves as the input dataset for subsequent reinforcement learning models.

[0119] S32: Based on the state-action mapping set, construct a path planning model based on the improved deep deterministic policy gradient algorithm; the path planning model includes a policy network and a value network; the policy network is used to output continuous control actions to generate candidate paths, and the value network is used to receive state-action pair inputs, calculate the corresponding action value function and output the path value evaluation result;

[0120] In this embodiment, a path planning model is constructed based on an improved deep deterministic policy gradient algorithm, which mainly includes a policy network and a value network.

[0121] Specifically, the policy network is used to generate optimal continuous control actions from the current environmental state. Its input is a navigation state vector, including high-dimensional information such as hull coordinates, heading, speed, attitude angles, wind direction, water current speed, and obstacle distribution matrix;

[0122] The policy network adopts a three-layer structure design:

[0123] The first layer is the feature extraction layer, which uses multidimensional linear mapping and normalization operations to unify the input environmental parameters into a standard feature space.

[0124] The second layer is the decision layer, which uses a nonlinear activation function to enhance the ability to model complex relationships between states, especially to perform partitioned responses to dynamic disturbances from multiple obstacles.

[0125] The third layer is the output layer, which restricts the generated control actions to a physically safe range to ensure that propulsion and steering are both within the hull's executable range.

[0126] The policy network explores by introducing random noise to avoid getting trapped in local optima. During the training phase, the policy network continuously adjusts its parameters so that the output actions can maximize the overall benefits of navigation (e.g., low energy consumption, short distance, and smooth path) under specific conditions.

[0127] Specifically, the role of the value network is to evaluate the actions output by the policy network. Its input consists of a state vector and the corresponding action, and its output is the comprehensive value of that action. The value network employs a two-stream input structure:

[0128] The state input stream focuses on extracting static features such as hull attitude, weather conditions, and obstacle distribution;

[0129] The motion input stream reflects the dynamic changing trends of continuous control variables such as the thruster and rudder angle.

[0130] After the two features are cross-referenced by the fusion layer, they are fed into the multi-layer fully connected module for comprehensive evaluation.

[0131] Value networks gradually master the mapping relationship between "state-action-result" through continuous training, thereby accurately predicting the navigation cost and safety of different action plans and providing a basis for strategy optimization.

[0132] S33: During the training process of the path planning model, a priority experience replay mechanism based on path importance scoring is adopted to update the experience samples with weights and obtain a priority experience sample set.

[0133] In this embodiment, an experience pool is established during the training process to store samples after each interaction between the submarine and the environment. Each sample contains five types of information: current state, action taken, navigation reward obtained, execution result status, and environmental feedback timestamp.

[0134] The experience pool adopts a dynamic rolling structure with a capacity of approximately 50,000 samples. When a new sample arrives, it automatically replaces the oldest low-value sample to maintain the timeliness and representativeness of the samples.

[0135] Different navigation paths have varying importance to model learning; therefore, a path importance scoring mechanism is introduced. A comprehensive importance score is calculated for each path sample, based on the following factors:

[0136] Mission relevance: Whether the sample is located in a critical segment of flight close to the target point or obstacle avoidance area;

[0137] Environmental complexity: obstacle density, flow rate change rate, and wind and wave interference level of the corresponding frame;

[0138] Strategy performance: The average reward level obtained by this route segment during the current voyage.

[0139] The scoring results are mapped to weight values ​​for subsequent sample extraction. Samples with higher importance receive a higher sampling probability, allowing the model to prioritize learning scenarios that are difficult to decide, high-risk, or have significant strategy differences.

[0140] In each training round, several batches of samples are drawn from the experience pool according to their importance weights for replay. Unlike traditional random sampling, priority experience replay significantly improves sample utilization, allowing the network to focus on learning the most decision-making scenarios, such as obstacle avoidance in narrow waterways or areas with sharp turns, in the early stages. Simultaneously, to prevent the model from overly favoring high-weight samples, the gradient update amount for each sample is reverse-balanced after sampling, ensuring the overall sample distribution remains diverse. As training progresses, the experience pool is periodically recalculated to ensure the model always focuses on the key learning areas of the current stage. The resulting priority experience sample set contains various typical environments, complex obstacle scenarios, and high-reward paths, providing high-value data support for subsequent attention mechanisms and policy optimization.

[0141] S34: Based on the priority experience sample set, an attention mechanism is introduced to perform weighted fusion of state features and action features to generate a fused feature tensor;

[0142] In this embodiment, an attention mechanism is introduced into the improved deep deterministic policy gradient model to enhance the network's feature selection and critical state response capabilities in complex water environments.

[0143] During the fusion phase, this embodiment uses a multi-head attention structure to adaptively weight the state features and action features.

[0144] Each "attention head" independently focuses on different semantic levels: for example, one attention head focuses on the spatiotemporal changes of a local obstacle region, while another focuses on the impact of speed changes on energy consumption.

[0145] The network automatically adjusts the weight coefficients of each attention point according to the current task stage (obstacle avoidance, constant speed cruise, or acceleration segment), so that the model can focus on the most influential feature combinations in different scenarios.

[0146] For example, when the unmanned research vessel is in an area with dense obstacles, the weight of visual channel features is dynamically increased, making the model pay more attention to the safety of local space; while in the open sea and when there is greater environmental disturbance, the weight of environmental features and speed features is increased to ensure the stability of path planning and energy consumption optimization.

[0147] The feature output after multi-head attention fusion is integrated into a high-dimensional fusion feature tensor, which includes multi-dimensional features such as environmental situation, obstacle distribution, hull attitude, and control action intention.

[0148] This tensor is not only used for the action value evaluation of the value network, but also fed back to the policy update stage of the policy network, realizing the adaptive propagation of features in the "perception-decision-feedback" closed loop.

[0149] Therefore, the path planning model can gradually enhance its response to key states (such as "nearby obstacles" and "sudden changes in wind and waves") during the training process, thereby improving the agent's policy robustness and convergence stability in dynamic water environments.

[0150] S35: Based on the fused feature tensor, a reward function is constructed, and the path planning model is iteratively optimized through a deterministic policy gradient update rule to obtain the optimal policy parameters; the reward function includes a range cost term and an energy consumption cost term, which are used to comprehensively evaluate the travel distance and propulsion energy consumption;

[0151] In this embodiment, a dynamic reward mechanism based on fused feature tensors is used to continuously learn and self-correct the path planning strategy, ultimately achieving a balanced control that minimizes flight distance, energy consumption, and safety.

[0152] The reward mechanism is no longer limited to static rewards with fixed weights, but adopts a dynamic multi-factor hierarchical design, which enables the model to adaptively adjust the target weights at different stages of the operation.

[0153] The reward function mainly consists of three types of sub-items:

[0154] Voyage cost: Used to encourage the vessel to reach the target point as quickly as possible and avoid ineffective navigation or repeated routes;

[0155] Energy consumption cost: Dynamically evaluate energy consumption per unit distance by combining propeller power and sailing resistance characteristics;

[0156] Safety constraint: Calculate the minimum safe distance between the hull and the nearest obstacle based on obstacle detection results. When the distance is lower than the set threshold, the penalty weight will be automatically increased.

[0157] During training, the relative weights of the three sub-items are dynamically adjusted according to different stages of the mission (takeoff, cruising, obstacle avoidance, and return). For example, during the obstacle avoidance phase, the weight of the safety constraint item is automatically increased; during the long-distance cruising phase, the weight of the energy consumption item is increased to improve range economy.

[0158] In each round of interaction, the state change, control action, reward value and the next state are stored in the experience pool, and multiple time series are randomly sampled and replayed each time update.

[0159] To accelerate model convergence and improve stability, a rolling weighting mechanism is introduced during reward updates. This mechanism assigns higher update weights to recently high-performing path samples, while reducing the influence of samples with unstable performance or high costs. This ensures that the model can continuously focus on learning high-quality path segments during long-term training, thereby reducing training fluctuations and improving the reliability of the final policy.

[0160] After multiple rounds of interactive training and dynamic reward updates, the model gradually forms a robust policy mapping relationship: outputting an appropriate combination of propulsion and rudder angle under specific conditions. During training, the policy convergence is continuously monitored, including metrics such as cumulative reward stability, path smoothness, and action continuity. When all metrics reach preset thresholds, the model stops updating and outputs the final optimal policy parameters.

[0161] S36: Generate the optimal navigation path based on the path planning model corresponding to the optimal strategy parameters.

[0162] S4: Based on the optimal navigation path, use model predictive control algorithm to decouple speed and heading control, and output speed control command and heading control command;

[0163] Among them, reference Figure 5 As shown, based on the optimal navigation path, a model predictive control algorithm is used to decouple speed and heading control, outputting speed control commands and heading control commands, including but not limited to the following steps:

[0164] S41: Based on the optimal navigation path, construct a hull optimization model based on model predictive control algorithm; the hull optimization model uses the ship dynamics equations and performs rolling calculations on the six degrees of freedom of the ship using the fourth-order Runge-Kutta method, and forms a set of predicted hull motion states based on the rolling calculation results; the six degrees of freedom states include hull position, velocity, acceleration, bow angle, angular velocity and angular acceleration;

[0165] In this embodiment, the optimal navigation path is discretized into several path nodes, each containing spatial coordinates, target speed, and desired heading angle. Subsequently, a hull optimization model is constructed based on the hull's dynamic characteristics to describe the hull's dynamic response relationship under six degrees of freedom.

[0166] The six degrees of freedom include: position parameters (longitudinal, lateral, and vertical displacement), velocity parameters (ship speed, lateral velocity, and heave rate), angle parameters (bow angle, roll angle, and pitch angle), and attitude changes such as angular velocity and angular acceleration.

[0167] The hull optimization model uses the fourth-order Runge-Kutta method to perform predictive calculations within a rolling time window, progressively updating the hull's future motion state. The prediction period is set to 1-2 seconds, and the sampling period is 0.05 seconds, enabling it to assess the state evolution trend several steps in advance.

[0168] The prediction results form a state set for the predicted motion of the hull, providing dynamic constraint boundaries for subsequent speed and heading control. This state set not only reflects the inertial and damping characteristics of the hull, but also includes the coupled effects of external disturbances (wind, waves) on attitude and velocity, providing a physical basis for decoupled control.

[0169] Furthermore, the optimal navigation path is discretized as follows:

[0170] ;

[0171] Where u represents the longitudinal velocity of the ship; v represents the lateral velocity of the ship; r represents the angular velocity of the ship's bow turn; x represents the position of the ship on the X-axis in the Earth's fixed coordinate system; y represents the position of the ship on the Y-axis in the Earth's fixed coordinate system; ψ represents the heading angle of the ship; k represents the value at the current time k; and X represents the longitudinal force acting on the hull. This represents the longitudinal force generated by the propeller. This is the direct input for speed control; This represents the longitudinal force component generated by the rudder; Y represents the longitudinal force (including resistance) generated by the interaction between the hull and the water flow; Y represents the lateral force acting on the hull. This represents the lateral force component generated by the propeller; This represents the lateral force generated by the rudder. This is the primary source of lateral force for heading control; This represents the lateral force generated by the hull; N represents the torque acting on the hull about the z-axis. This indicates the pitching torque generated by the propeller; This represents the pitching torque generated by the rudder. This is the direct input for heading control; The torque generated by the hull during bowing is represented by m; m represents the mass of the ship. This indicates the ship's additional mass in the x-axis direction (longitudinal direction); This indicates the ship's additional mass in the y-axis direction (lateral direction); This represents the moment of inertia of the ship about the z-axis (vertical axis); This represents the ship's additional moment of inertia about the z-axis; A discretized approximation of longitudinal acceleration; A discretized approximation of lateral acceleration; A discretized approximation of the head-turn angular acceleration; Represents the velocity in the Earth coordinate system along the X-axis; Represents the velocity in the Earth coordinate system along the Y-axis; Indicates the rate of change of the heading angle; This indicates the speed of the hull in the longitudinal direction, that is, the propulsion speed along the bow direction of the hull; This indicates the hull's velocity in the lateral direction, which is usually caused by external flow fields, sway disturbances, and the action of the rudder surfaces, and is used to describe the hull's lateral motion tendency. This indicates the bow angular velocity of the hull, which is the angular velocity of rotation about the vertical axis.

[0172] The predicted state for the next moment is:

[0173] ;

[0174] in, and The coordinates of the center of mass of the unmanned surface vessel (USV) in the geographic coordinate system represent the lateral and longitudinal positions of the USV's center of mass, which are used to describe the actual spatial position of the USV in the operating waters. The heading angle represents the angle of the hull's bow relative to geographic north, used to determine the direction of navigation and attitude. This represents the time interval between each update during the forecasting and control process, typically ranging from tens to hundreds of milliseconds, and is used to determine the time resolution of the rolling forecast. It represents the longitudinal acceleration function, reflecting the combined effect of propulsion, propeller load, and fluid resistance on the change in the forward speed of the hull; The lateral acceleration function is mainly caused by rudder deflection, lateral hydrodynamics, and wave disturbance, and is used to predict the hull's sway motion. This represents the bow angular acceleration function, which indicates the influence of the rudder angle torque and propulsion asymmetric torque on the bow rotation motion of the hull; and These represent the velocity variation functions of the hull along the X and Y axes in the geographic coordinate system, respectively, used to calculate the displacement update of the hull's center of mass on the horizontal plane; This represents the rate of change function of the heading angle, used to predict the trend of the bow angle change of the hull at the next moment; Indicates at time Given the known state and input conditions, for the next control sampling time... The predicted longitudinal velocity; Indicates at time Under the conditions The predicted value of lateral velocity; Indicates at time Under the conditions Predicted value of bow angular velocity (rotational angular velocity about the vertical axis); Indicates at time Under the conditions Predicted X-coordinate of the location; Indicates at time Under the conditions Predicted Y-coordinate of the location; Indicates at time under conditions The predicted heading angle.

[0175] S42: Based on the hull optimization model, construct a speed prediction sub-model and a heading prediction sub-model; the speed prediction sub-model is used to predict the propulsion-speed response relationship in the longitudinal direction, and combined with propulsion power constraints and propeller speed variation constraints, the speed control sequence is obtained by rolling solution; the heading prediction sub-model is used to predict the bow angle-rudder angle response relationship in the lateral direction, and combined with rudder angle constraints and lateral deviation constraints, the heading correction sequence is obtained by rolling solution.

[0176] In this embodiment, after the hull optimization model is established, the overall control problem is decomposed into two relatively independent sub-control problems, which are used for longitudinal propulsion and lateral steering control, respectively, thereby realizing hierarchical prediction and optimization of speed and heading.

[0177] (1) Speed ​​prediction sub-model (longitudinal control):

[0178] The speed prediction sub-model is used to describe the response relationship between propulsion and hull speed. Its core objective is to predict the speed change trend at several future moments based on the speed command of the optimal path and generate the propulsion control sequence with optimal energy consumption.

[0179] The speed prediction sub-model first calculates the future speed change trend based on the hull's current speed, target speed, and external disturbances (such as downstream or upstream currents). In this process, it takes into account factors such as the propeller's power curve, speed regulation delay, propulsion efficiency, and hull resistance changes.

[0180] For example, when the hull is in an area with uphill currents or wind and waves, the model will automatically increase the upper limit of propulsion power to maintain speed; when the hull is in a downstream or tailwind area, the model will automatically reduce propulsion power to prioritize maintaining the steady speed with the lowest energy consumption.

[0181] Furthermore, the speed prediction sub-model incorporates a thruster speed change rate constraint to limit frequent propeller acceleration and deceleration, thereby reducing mechanical wear and energy loss. The speed control sequence for several future moments is solved in a rolling fashion within the prediction window and updated in real-time during each sampling period to ensure smooth speed transitions and continuously controllable acceleration. This mechanism guarantees that the unmanned research vessel can achieve target speed tracking with optimal energy consumption under various weather conditions, avoiding unstable operations such as "over-acceleration" or "frequent deceleration."

[0182] (2) Forecasting sub-model (lateral control):

[0183] The heading prediction sub-model is mainly used to predict the response relationship between the bow angle and rudder angle of the hull, which is a key link in controlling the stability of the hull's trajectory.

[0184] The heading prediction sub-model integrates the hydrodynamic characteristics of the rudder surface, the rudder angle input lag, and the influence of hull inertia on steering. Within each roll cycle, based on the current heading angle, the target heading angle, and obstacle distribution information, it predicts the heading change trend at several future moments and calculates the rudder angle correction sequence in real time.

[0185] To prevent the hull from rolling or becoming unstable due to excessively rapid changes in rudder angle, the model is equipped with an upper limit on the rate of change of rudder angle and a maximum yaw angle constraint to ensure smooth and controllable servo motor actions.

[0186] Meanwhile, during the prediction phase, the lateral deviation (the difference in lateral distance between the hull track and the optimal path) is dynamically evaluated, and a penalty weight is applied to the deviation, so that the rudder angle control is always optimized with the goal of minimizing the lateral drift error.

[0187] For example, when the hull deviates from the predetermined path, the rudder angle correction force is automatically increased to quickly return to the center track; while when the hull track is stable, the rudder angle output is gradually reduced to reduce heading fluctuations and improve energy efficiency.

[0188] Through the hierarchical control design of the two prediction sub-models (longitudinal and lateral), independent prediction and constraint solving of speed and heading are achieved. Each sub-model is optimized in the future time domain, and the prediction results are fed back to the main controller in real time, thereby achieving multi-objective balance in dynamic environments: ensuring accurate speed tracking, maintaining stable heading, and minimizing energy consumption.

[0189] S43: In the joint control phase of the speed prediction sub-model and the speed prediction sub-model, a speed-heading decoupled control strategy is adopted to coordinate and optimize longitudinal propulsion and lateral maneuvering, and generate a control increment sequence.

[0190] In this embodiment, after entering the core execution layer of model predictive control, in order to ensure that speed and heading do not interfere with each other under dynamic sea conditions, a speed-heading decoupling control strategy based on hierarchical distributed optimization is adopted.

[0191] Specifically, the upper-level controller is responsible for global path tracking and speed control, while the lower-level controller is responsible for rudder angle control and attitude maintenance. Based on the optimal path node set provided by the path planning module, the upper-level controller calculates the target speed and attitude change trends of the hull over multiple sampling periods. The lower-level controller uses the reference speed and heading angle output from the upper-level controller as input to adjust the thruster speed and rudder angle response in real time, achieving rapid tracking. This two-layer structure ensures that longitudinal propulsion and lateral maneuvering commands do not compete with each other in the time domain, and the control output has feedforward characteristics, allowing for correction before disturbances occur.

[0192] In actual navigation, changes in the propulsion force of the hull often affect heading stability. For example, when the propeller accelerates, slight yaw may occur due to fluid asymmetry; similarly, rudder angle adjustments may cause a sudden drop in speed. To address these coupling issues, this embodiment introduces a "cross-sensitivity compensation module" at the control solution layer. This module automatically calculates the degree of mutual influence between the propulsion system and the servo system based on the state changes of the previous cycle and incorporates compensation terms during optimization. This allows for smoother heading corrections while maintaining stable speed changes, thereby avoiding over-response or control oscillations.

[0193] Within each prediction cycle, instead of directly outputting absolute control commands, a sequence of control increments (i.e., the change in control input between the current and previous moments) is calculated. The introduction of control increments makes the entire control process exhibit a "gradual convergence" characteristic: when the hull state deviates significantly, the control increment is larger to ensure rapid correction; when approaching the target state, the control increment gradually decreases to achieve a smooth transition.

[0194] This mechanism not only improves the stability of the control system, but also significantly reduces the mechanical wear of actuators (such as thrusters and servo motors), thus extending the equipment's lifespan.

[0195] Under multiple disturbance conditions (such as periodic wave disturbances or sudden wind speed changes), the longitudinal and lateral control weights are dynamically adjusted.

[0196] When the lateral yaw rate exceeds the set threshold, the priority of heading control is temporarily increased to ensure the stability of the hull track; when the wind speed suddenly increases or the countercurrent intensifies, the longitudinal propulsion weight is increased to maintain the target speed.

[0197] This real-time weight redistribution mechanism enables unmanned surface vessels to dynamically maintain their optimal state according to environmental changes, achieving true collaborative optimization and adaptive navigation.

[0198] Furthermore, the specific formulas for speed control and heading control are as follows:

[0199] (1) Speed ​​control:

[0200] To reduce algorithm complexity, the speed control is set as the control of the longitudinal velocity u. The Euler iteration method is used to predict the time domain. The longitudinal speed of the ship inside is:

[0201] ;

[0202] in, Indicates at time Under the information conditions, for the future The predicted longitudinal velocity of the hull for each prediction step is used to describe the trend of the forward velocity of the unmanned research vessel in the prediction time domain. Indicates the current moment Under the condition of, for the first The longitudinal acceleration function value of the prediction step, i.e., the hull in the future step. The acceleration is estimated under the combined effects of propulsion, drag, and external disturbances (such as waves and wind); i represents the summation index of the acceleration term, which is the prediction result of each step continuously accumulated from the current moment in the prediction window; j represents the prediction step size index, which is used to indicate the position of the current prediction value in the future time series. This indicates the length of the prediction time domain.

[0203] To achieve energy conservation and reduce losses from frequent propeller speed adjustments, propeller performance also needs to be considered. Therefore, taking into account both reducing speed error and controlling the speed input, the following speed optimization function is constructed:

[0204] ;

[0205] in, This indicates that, within the rolling optimization framework of model predictive control, the optimization objectives are to minimize the longitudinal velocity tracking error of the hull and to smooth the changes in propulsion control, considering the control input at the current moment and several future steps. Its change Simultaneously solving the problem, the cost function reaches its minimum value; Indicates at time Under these conditions, for the future... The speed error vector for each prediction step, i.e., the deviation between the expected speed and the predicted speed; The penalty weight represents the speed error; Indicates the current moment Next, for the first The control input increment for each control step is the change in thruster speed between two adjacent sampling times. The penalty weight representing the propeller speed; This indicates the control time domain (the number of steps to optimize the control input).

[0206] (2) Heading control:

[0207] Predicting the time domain based on the Euler iteration method The predicted bow angle of the ship inside is:

[0208] ;

[0209] in, Indicates at time Under these conditions, for the future... The predicted heading angle for each prediction step; Indicates at time Under the condition of, for the first The rate of change function of the heading angle for each prediction step, i.e., the predicted value of the heading angular velocity.

[0210] A smaller rudder angle is more conducive to ship safety. Therefore, taking into account both rudder angle performance and heading tracking performance, the following optimization function is constructed:

[0211] ;

[0212] in, Indicates the current moment Next, by adjusting the rudder angle input sequence and its change To minimize the energy consumption of heading deviation and rudder angle change in the future prediction time domain, and achieve the comprehensive optimization of heading tracking accuracy and control smoothness; Indicates the current moment Next, regarding the future The heading angle error vector for each prediction step, i.e., the deviation between the expected heading angle and the predicted heading angle; Indicates the penalty weight for heading error; Indicates the current time Next The change in rudder angle per control step; This indicates the penalty weight for the rudder angle.

[0213] S44: Based on the control increment sequence, perform rolling correction on the propeller speed and rudder angle input, and output speed control command and heading control command.

[0214] In this embodiment, the propulsion and rudder angle inputs are adjusted in real time based on the control increment sequence generated by the decoupling control strategy to achieve dynamic correction of the navigation attitude.

[0215] Propeller control section: The control system finely adjusts the propeller speed according to the speed control command. When water resistance or wind speed increase is detected, the system automatically increases the propeller speed; when environmental resistance decreases, the propeller speed smoothly decreases to maintain optimal energy consumption.

[0216] Rudder angle control section: The servo system dynamically corrects the rudder angle based on the heading control command, and adjusts the rate of angle change in real time in conjunction with the feedback signal from the angle sensor to prevent oversteering or rudder angle oscillation. The system adopts a closed-loop feedback mechanism. When the actual heading deviation is detected to exceed the threshold, the target rudder angle value is immediately updated and the future heading is re-predicted.

[0217] This rolling correction process is updated in real time within each control cycle (approximately 50 milliseconds) to ensure that the hull always sails near the desired track.

[0218] In addition, to improve the system's anti-disturbance capability, the control module is equipped with an adaptive adjustment mechanism: when a sudden wind or wave impact is detected, the prediction time domain length will be temporarily adjusted and the control sequence will be recalculated to achieve dynamic recovery and attitude self-stabilization.

[0219] The final output speed and heading control commands are sent from the host computer to the actuators, which work together through the propulsion system and servo system to achieve high-precision control of the speed and direction. Field tests showed that this control scheme maintained a heading error within ±1.2° and a speed deviation of less than ±3% under Force 3 wind and wave conditions, significantly improving the navigation stability and energy efficiency of the unmanned research vessel.

[0220] S5: Based on the speed control command and the heading control command, output the real-time navigation control results of the fishery unmanned scientific research vessel.

[0221] Please see Figure 6 This application also provides an intelligent navigation control system based on a fishery unmanned research vessel, the system comprising:

[0222] The data acquisition module is used to obtain environmental parameters and vessel operating status parameters of the target operating area;

[0223] The data fusion processing module is used to perform real-time perception and feature extraction of environmental parameters and hull operating status based on multi-sensor fusion algorithms, and generate a navigation status information set.

[0224] The surface obstacle detection module is used to establish a surface obstacle detection model using a deep semantic segmentation algorithm, perform semantic segmentation on the navigation state information set, and output obstacle detection results.

[0225] The path planning module is used to establish a path planning model based on the obstacle detection results and navigation status information set, and to optimize the navigation path of the hull and output the optimal navigation path.

[0226] The navigation control module is used to decouple speed and heading based on the optimal navigation path using model predictive control algorithms, and output speed control commands and heading control commands.

[0227] The execution control module is used to output the real-time navigation control status of the fishery unmanned research vessel based on the speed control command and the heading control command.

[0228] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0229] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0230] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0231] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0232] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0233] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0234] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0235] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0236] This application provides an intelligent navigation control method and system for an unmanned research vessel used in fisheries. By constructing an intelligent navigation control system based on multi-sensor fusion, it achieves autonomous perception and stable control of the unmanned research vessel in complex aquatic environments. By fusing environmental parameters and vessel operating status information, it can comprehensively acquire multi-source data such as wind speed, wave height, water flow speed, and vessel attitude, significantly improving environmental perception accuracy and dynamic response capabilities. A surface obstacle detection model established using a deep semantic segmentation algorithm can accurately identify and segment various types of obstacles, improving the real-time performance and reliability of obstacle avoidance and avoiding recognition errors caused by visual noise or changes in lighting. A path planning model based on the improved Deep Deterministic Policy Gradient (DDPG) algorithm, combined with navigation status information and obstacle detection results, achieves global policy optimization of the vessel's navigation path, resulting in an optimal balance between safety, energy consumption, and travel time. Meanwhile, the introduction of model predictive control (MPC) algorithm to decouple speed and heading control, through propeller speed adjustment and dynamic rudder angle correction, effectively suppresses track deviation and attitude oscillation caused by coupling interference, and significantly improves navigation stability and control accuracy.

[0237] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0238] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0239] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0240] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0241] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0242] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. An intelligent navigation control method based on a fishery unmanned scientific research boat, characterized in that, The method includes the following steps: Based on a multi-sensor fusion algorithm, the environmental parameters and hull operation status parameters of the fishery unmanned scientific research vessel in the target operating waters are perceived and features are extracted in real time to generate a navigation status information set. The navigation status information set is semantically segmented, and obstacle detection results are output. Based on the obstacle detection results, the navigation path of the fishery unmanned research vessel is optimized using an improved depth deterministic strategy gradient algorithm, and the optimal navigation path is output. Based on the optimal navigation path, a model predictive control algorithm is used to decouple the speed and heading control, and output speed control commands and heading control commands. The real-time navigation status of the fishery unmanned research vessel is controlled according to the speed control command and the heading control command. The step of optimizing the navigation path of the fishery unmanned research vessel based on the obstacle detection results and using an improved depth deterministic policy gradient algorithm to output the optimal navigation path includes: Based on the obstacle detection results and the navigation status information set, a state-action mapping set is generated; the state-action mapping set is used to describe the state space and action space that correspond to the hull operation state and control actions. Based on the state-action mapping set, a path planning model based on the improved deep deterministic policy gradient algorithm is constructed. During the training process of the path planning model, a priority experience replay mechanism based on path importance scoring is adopted to update the experience samples with weights, thereby obtaining a priority experience sample set. An attention mechanism is introduced based on the priority experience sample set to weight and fuse state features and action features to generate a fused feature tensor. Based on the fused feature tensor, a reward function is constructed, and the path planning model is iteratively optimized using a deterministic policy gradient update rule to obtain the optimal policy parameters. The reward function includes a range cost term and an energy consumption cost term, which are used to comprehensively evaluate the travel distance and propulsion energy consumption. The optimal navigation path is generated based on the path planning model corresponding to the optimal strategy parameters.

2. The method according to claim 1, characterized in that, The navigation status information set includes wind speed, wind direction, wave height, surface current parameters, ultraviolet radiation intensity, temperature and humidity, speed information, heading information, and hull attitude angle information.

3. The method according to claim 1, characterized in that, The semantic segmentation of the navigation state information set and the output of obstacle detection results include: The navigation status information set is preprocessed to obtain an image sequence containing water surface areas and obstacle areas; Construct a deep semantic segmentation network model that includes encoding and decoding layers; The image sequence is semantically segmented using a deep semantic segmentation network model to generate a semantic segmentation map. Based on the semantic segmentation map, the spatial location and distribution density of obstacles in the hull coordinate system are calculated using a spatial clustering algorithm to form obstacle detection results; the obstacle detection results include the type, location and distribution information of obstacles on the water surface.

4. The method according to claim 3, characterized in that, The encoding layer uses MobileNetV2 as the backbone network and extracts multi-scale contextual features by combining convolution and dilated convolution; the decoding layer performs boundary information recovery and pixel classification reconstruction based on the deconvolution structure to generate a semantic segmentation map.

5. The method according to claim 1, characterized in that, The path planning model includes a policy network and a value network; the policy network is used to output continuous control actions to generate candidate paths, and the value network is used to receive state-action pair inputs, calculate the corresponding action value function, and output the path value evaluation result.

6. The method according to claim 1, characterized in that, Based on the optimal navigation path, a model predictive control algorithm is used to decouple speed and heading control, outputting speed control commands and heading control commands, including: Based on the optimal navigation path, a hull optimization model based on the model predictive control algorithm is constructed; Based on the aforementioned hull optimization model, a speed prediction sub-model and a heading prediction sub-model are constructed. In the joint control phase of the speed prediction sub-model and the speed prediction sub-model, a speed-heading decoupled control strategy is adopted to coordinate and optimize longitudinal propulsion and lateral maneuvering, and generate a control increment sequence. Based on the control increment sequence, the propeller speed and rudder angle input are rolled and corrected, and the speed control command and heading control command are output.

7. The method according to claim 6, characterized in that, The hull optimization model utilizes the ship dynamics equations and performs rolling calculations on the ship's six degrees of freedom states using the fourth-order Runge-Kutta method. Based on the rolling calculation results, a set of predicted hull motion states is formed. The six degrees of freedom states include hull position, velocity, acceleration, bow angle, angular velocity, and angular acceleration.

8. The method according to claim 6, characterized in that, The speed prediction sub-model is used to predict the thrust-speed response relationship in the longitudinal direction. Combined with the thrust power constraint and the propeller speed change constraint, the speed control sequence is obtained by rolling solution. The heading prediction sub-model is used to predict the heading angle-rudder angle response relationship in the lateral direction. Combining rudder angle constraints and lateral deviation constraints, the heading correction sequence is obtained by rolling solution.

9. An intelligent navigation control system based on an unmanned research vessel for fisheries, characterized in that, The system is applied to the method as described in any one of claims 1-8, comprising: The data acquisition module is used to obtain environmental parameters and vessel operating status parameters of the target operating area; The data fusion processing module is used to perform real-time perception and feature extraction of environmental parameters and hull operating status based on multi-sensor fusion algorithms, and generate a navigation status information set. The surface obstacle detection module is used to establish a surface obstacle detection model using a deep semantic segmentation algorithm, perform semantic segmentation on the navigation state information set, and output obstacle detection results. The path planning module is used to establish a path planning model based on the obstacle detection results and the navigation state information set, optimize the navigation path of the hull, and output the optimal navigation path. The navigation control module is used to decouple the speed and heading based on the optimal navigation path using a model predictive control algorithm, and output speed control commands and heading control commands. The execution control module is used to output the real-time navigation control status of the fishery unmanned research vessel according to the speed control command and the heading control command.

Images