Unmanned ship measurement and control system and method based on Beidou artificial intelligence

Abstract

The invention discloses an unmanned ship measurement and control system and method based on Beidou artificial intelligence, and relates to the technical field of unmanned ship measurement and control. According to the Beidou-based artificial intelligence unmanned ship measurement and control method, Beidou positioning information, navigation state parameters and water area image data are obtained in real time in the cruising process of an unmanned ship; inputting the image data into an AI visual recognition model to judge whether an obstacle exists or not; if there is an obstacle, extracting the features of the obstacle and constructing an obstacle spatial distribution map; generating a plurality of avoidance paths based on the distribution map and the navigation state parameters, and selecting an optimal path; according to the method, the image of the water area in front of the ship body is input into the AI visual model, feature extraction, target detection and depth estimation are combined, the spatial position and boundary information of the obstacle are extracted, and the spatial distribution diagram is constructed under the ship body coordinate system; and the obstacle identification precision and the positioning modeling capability are improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned vessel measurement and control technology, specifically to a Beidou-based artificial intelligence unmanned vessel measurement and control system and method. Background Technology

[0002] In recent years, with the rapid development of artificial intelligence, satellite navigation, computer vision, and unmanned systems technologies, unmanned surface platforms have been widely used in waterway inspection, water quality monitoring, underwater mapping, and surface security. The BeiDou Navigation Satellite System, my country's independently developed global satellite navigation system, possesses all-weather, high-precision, and globally covered spatiotemporal positioning capabilities, providing unmanned vessels with precise positioning, heading tracking, and path planning support. Simultaneously, artificial intelligence visual recognition technology can efficiently process surface image data, enabling environmental perception tasks such as obstacle recognition and surface target detection. Combining multi-source information such as water quality sensors, lidar, and depth estimation, intelligent unmanned surface vessel systems with autonomous navigation and real-time monitoring capabilities have become a key equipment direction for future waterway operations and intelligent inspection.

[0003] The limitations of existing technologies include at least the following problems: when performing pre-set route inspection tasks, unmanned vessels lack a dynamic obstacle avoidance and path correction mechanism based on visual recognition, making it difficult for them to respond effectively to sudden obstacles in real aquatic environments.

[0004] Currently, most automated inspection systems rely solely on preset routes and navigation positioning data for path control, lacking the ability to identify and respond to potential obstacles in the waters ahead of the vessel in real time. This is especially problematic when non-preset targets such as floating debris, weeds, or other small vessels are present, easily leading to collisions or course deviations due to a lack of effective perception and processing capabilities. Furthermore, even when existing solutions are equipped with some sensors, they often fail to integrate with the path planning algorithm, making it difficult to simultaneously generate avoidance paths and correct navigation parameters after obstacle detection. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a BeiDou-based AI-powered unmanned vessel measurement and control system and method, which solves the problem that existing technologies lack a closed-loop, intelligent path adjustment mechanism when facing complex and variable aquatic environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a BeiDou-based AI-powered unmanned vessel measurement and control method, comprising the following steps: when the unmanned vessel is cruising in the target inspection waters along the target route, the BeiDou positioning information, current navigation status parameters, and water area image data in front of the vessel are acquired in real time; the water area image data is input into a pre-trained AI visual recognition model to determine whether there are obstacle targets; if obstacle targets are identified, the obstacle target feature set is extracted, and an obstacle spatial distribution map is constructed; based on the obstacle spatial distribution map and the current navigation status parameters, several avoidance paths are analyzed and generated, and the optimal avoidance path is selected; based on the optimal avoidance path, the current navigation status parameters are adjusted, and the target route is synchronously corrected in conjunction with the BeiDou positioning information.

[0007] Furthermore, the water area image data includes several water area pixels, and each water area pixel corresponds to a water area pixel value and a three-dimensional coordinate.

[0008] Furthermore, the obstacle target feature set includes the spatial location, size information, and boundary range of several obstacles, and the AI ​​visual recognition model includes an image feature extraction network, an obstacle target detection network, a depth estimation network, and a feature fusion network.

[0009] Furthermore, the specific steps for the AI ​​visual recognition model to identify and extract features of obstacle targets are as follows: In the image feature extraction network of the AI ​​visual recognition model, convolutional feature extraction and multi-scale semantic enhancement processing are performed on the input water image data to generate image semantic features; in the obstacle target detection network of the AI ​​visual recognition model, obstacle target regions in the water image are identified based on image semantic features, and the corresponding two-dimensional boundary range is output; in the depth estimation network of the AI ​​visual recognition model, the spatial location and size information of the obstacle target are analyzed based on the obstacle target region and image semantic features; in the feature fusion network of the AI ​​visual recognition model, the boundary range, spatial location, and size information of the obstacle target are fused to generate an obstacle target feature set.

[0010] Furthermore, the specific steps for constructing the obstacle spatial distribution map are as follows: read the spatial location, size information and boundary range of each obstacle; project the spatial location of each obstacle onto a three-dimensional spatial coordinate system with the center of the unmanned vessel as the origin; divide the space into spatial grid cells within a preset range in front of the vessel, and mark the spatial location and boundary range of each obstacle in the corresponding grid cells; generate the obstacle spatial distribution map based on the marking results of the grid cells.

[0011] Furthermore, the current navigation status parameters include the ship's position, speed, heading angle, and target route direction. The specific steps for analyzing and generating several avoidance paths are as follows: First, using the ship's position and heading angle from the current navigation status parameters as the starting direction, construct a candidate path search area in the waters ahead of the ship. Second, project the spatial positions and boundary ranges of each obstacle in the obstacle spatial distribution map in the three-dimensional coordinate system onto the candidate path search area. Third, set the path search step size and angle threshold based on the speed and heading angle from the current navigation status parameters, and iteratively generate multiple heading offset paths in the search area. Fourth, determine whether each heading offset path spatially overlaps with obstacles in the obstacle spatial distribution map, and filter out paths with collision risk. Fifth, set the remaining collision-free paths as avoidance paths.

[0012] Furthermore, the specific steps for selecting the optimal avoidance path are as follows: Based on the ship's position, speed, and heading angle in the current navigation state parameters, analyze the navigation offset and target route deviation angle corresponding to each avoidance path; extract the minimum safe distance and path length between each avoidance path and the obstacle from the obstacle spatial distribution map; and analyze the optimal avoidance path based on the navigation offset, target route deviation angle, minimum safe distance, and path length.

[0013] Furthermore, the specific steps for analyzing the optimal avoidance path are as follows: weighted analysis is performed on the navigation offset, target route deviation angle, minimum safe distance, and path length to obtain the path score for each avoidance path; the path scores of all avoidance paths are compared and analyzed, and the avoidance path with the highest path score is selected as the optimal avoidance path.

[0014] Furthermore, during the unmanned vessel inspection process, water quality testing data is collected and uploaded in real time, including the following steps: during the unmanned vessel inspection process, water quality testing data of the inspection points is acquired in real time and water quality analysis is performed. The water quality testing data includes pH value, conductivity, dissolved oxygen concentration, and turbidity; the water quality testing data and water quality analysis results are uploaded to a preset remote data center in real time.

[0015] A BeiDou-based AI-powered unmanned vessel tracking and control system includes: a cruise perception unit, used to acquire in real time the unmanned vessel's BeiDou positioning information, current navigation status parameters, and water area image data in front of the vessel while it cruises along a target route in the target inspection area; an obstacle identification unit, used to input the water area image data into a pre-trained AI visual recognition model to determine whether there are obstacle targets; an obstacle modeling unit, used to extract obstacle target feature sets and construct an obstacle spatial distribution map when obstacle targets are identified; an obstacle avoidance path unit, used to analyze and generate several avoidance paths based on the obstacle spatial distribution map and current navigation status parameters, and select the optimal avoidance path; and a trajectory control unit, used to adjust the current navigation status parameters based on the optimal avoidance path and synchronously correct the target route in conjunction with BeiDou positioning information.

[0016] The present invention has the following beneficial effects:

[0017] (1) The Beidou-based AI-based unmanned vessel measurement and control method inputs the water area image data collected in front of the vessel into a pre-trained AI visual recognition model. Combined with multi-level network processing steps such as image feature extraction, target detection, depth estimation and feature fusion, it can not only accurately identify obstacles on the water surface, but also effectively extract their spatial location, size information and boundary range, and construct a spatial distribution map in the vessel coordinate system. By introducing a three-dimensional coordinate mapping and spatial grid division mechanism, the system can more intuitively present the distribution characteristics of obstacles in the current navigation environment. Compared with the traditional single-frame image or two-dimensional recognition method, the recognition accuracy and modeling depth are significantly improved, which significantly enhances the unmanned vessel's ability to identify and locate static or dynamic obstacles in complex waters.

[0018] (2) The Beidou-based AI-based unmanned vessel measurement and control method constructs a search area based on the real-time position and heading status of the vessel. By setting the path search step size, angle threshold, and spatial overlap detection mechanism, the system dynamically analyzes and filters the relationship between candidate paths and obstacles. After eliminating paths with collision risks, the system further combines factors such as navigation offset, target route deviation angle, minimum safe distance, and path length to comprehensively score all feasible paths and select the current optimal path based on the scoring results. Compared with the traditional scheme that relies on static preset paths, this method has stronger environmental adaptability and real-time response capability. It can flexibly adjust the navigation strategy according to the actual water conditions and supports the adjustment of preference weights based on mission objectives, thereby ensuring navigation safety while taking into account speed efficiency and path optimization requirements.

[0019] (3) The Beidou-based AI-powered unmanned vessel measurement and control method interpolates the path offset angle with the current speed and heading of the vessel and generates corresponding rudder angle and propulsion control commands in combination with the vessel kinematic model. The system can achieve precise control of the actual navigation attitude. At the same time, it uses the Beidou positioning system to obtain the actual position data of the vessel in real time and compares and corrects it with the target route to form a closed-loop control process of continuous track execution, positioning feedback and dynamic adjustment. In complex water environments, this mechanism can effectively deal with the yaw problem caused by navigation errors, water flow disturbances or obstacle changes, and ensure that the unmanned vessel always sails stably along the preset route. In long-distance inspection or water quality monitoring tasks, this method can continuously adjust the track to ensure complete coverage of the target water area, while improving the coherence and regional accuracy of the collected water quality data.

[0020] (4) The Beidou-based AI-powered unmanned vessel measurement and control system is composed of a combination of a cruise perception unit, an obstacle identification unit, an obstacle modeling unit, an avoidance path unit, and a trajectory control unit. Each module has a clear functional boundary and standardized data interface, and has high configurability and replaceability. For example, the obstacle identification unit can flexibly replace different types of AI vision models to adapt to different application scenarios such as sea, lake, or inland waterway. The control logic in the trajectory control unit can also be parameterized and adapted according to the power and steering control characteristics of different ship types. This modular design concept significantly improves the system's migration capability and subsequent functional expansion capability between different platforms, which is conducive to enterprises to quickly deploy the system on different types of unmanned vessels. It is also convenient to integrate additional functions such as water quality detection, shore-based collaborative control, and satellite link communication. It has good engineering implementation value and commercial promotion potential. Compared with the traditional customized measurement and control scheme with tight functional coupling and difficult upgrades, this system has a more flexible structure, lower maintenance cost, and higher update efficiency.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] Figure 1 This is a flowchart of a Beidou-based artificial intelligence-based unmanned vessel measurement and control method according to the present invention.

[0023] Figure 2 This is a flowchart illustrating the specific steps involved in constructing a spatial distribution map of obstacles in a BeiDou-based artificial intelligence-driven unmanned vessel measurement and control method according to the present invention.

[0024] Figure 3 This is a block diagram of a Beidou-based artificial intelligence-based unmanned ship measurement and control system according to the present invention. Detailed Implementation

[0025] Please see Figure 1This invention provides a technical solution: a BeiDou-based AI-powered unmanned vessel measurement and control method, comprising the following steps: when the unmanned vessel is cruising along a target route in a target inspection area, the BeiDou positioning information, current navigation status parameters, and water area image data in front of the vessel are acquired in real time. The water area image data is collected in real time by a high-definition binocular camera or lidar installed at the front of the vessel, with an image frame rate of not less than 30fps and an image resolution of not less than 1080p, used to construct real-time forward environmental perception input; the water area image data is input into a pre-trained AI visual recognition model for judgment. The system determines whether an obstacle target exists. If an obstacle target is identified, its feature set is extracted, and a spatial distribution map of the obstacle is constructed. Based on the spatial distribution map of the obstacle and the current navigation state parameters, several avoidance paths are generated, and the optimal avoidance path is selected. Based on the optimal avoidance path, the current navigation state parameters are adjusted, and the target route is synchronously corrected in conjunction with BeiDou positioning information. Specifically, the heading angle and speed of the optimal path are controlled by difference with the current navigation state, and control commands (rudder angle, throttle, etc.) are generated in conjunction with the ship's kinematics model. The route deviation is then corrected in real time using BeiDou positioning data.

[0026] Specifically, the water area image data includes several water area pixels, and each water area pixel corresponds to a water area pixel value and a three-dimensional coordinate.

[0027] The three-dimensional coordinates are spatial position data obtained from the depth perception of image pixels by the binocular cameras, lidar, or depth estimation models mounted on the hull. A three-dimensional spatial coordinate system is established with the center of the unmanned hull as the origin, where the X-axis represents the forward direction of the hull, the Y-axis represents the lateral (left / right) direction of the hull, and the Z-axis represents the vertical (up / down) direction. Each water pixel is mapped to the hull coordinate system after camera intrinsic parameter transformation and depth fusion processing, and the spatial position of the pixel in the hull reference system is represented by a (X,Y,Z) triple.

[0028] The obstacle target feature set includes the spatial location, size information, and boundary range of several obstacles. The AI ​​visual recognition model includes an image feature extraction network, an obstacle target detection network, a depth estimation network, and a feature fusion network.

[0029] The pre-training steps for the AI ​​visual recognition model are as follows:

[0030] Collect image data containing various real water scenes, covering different weather and lighting conditions such as sunny days, cloudy days, and nighttime;

[0031] The images are manually annotated, including: the two-dimensional bounding box of the obstacle target, the corresponding spatial location and size information (which can be obtained with the assistance of LiDAR or binocular camera), as well as camera parameters and ship coordinate system transformation information.

[0032] First, a general image dataset is used to train the basic network to enable it to have basic image feature recognition capabilities.

[0033] Then, the model was further trained on a water image dataset to adapt to special background conditions such as water surface reflection, floating objects, and wave interference.

[0034] Using water area images and labeled obstacle bounding boxes, an object detection network is trained to accurately identify various obstacles (such as driftwood, garbage bags, small boats, etc.) in complex water environments.

[0035] Enhance the detection capabilities for small targets, occluded objects, and low-contrast targets during training.

[0036] The depth estimation network is trained using image samples with known depth information, enabling it to estimate the approximate spatial location and size of each obstacle target from a single frame image.

[0037] For images without true depth labels, pseudo-labels can be introduced or semi-supervised learning can be performed to improve the robustness and generalization ability of the model.

[0038] The detected boundary range information is fused with the estimated spatial location and size information for training, thereby optimizing the fusion network's ability to understand and represent multi-source features.

[0039] The goal is to enable the network to stably output a structured set of obstacle target features, supporting the analysis of subsequent path avoidance and control strategies.

[0040] The above four sub-networks are integrated into a unified AI visual recognition model framework for overall training and optimization, enabling the networks to work together and improving the overall recognition accuracy and stability.

[0041] During training, model parameters are continuously adjusted through cross-validation to ensure the final model's effectiveness in real-world applications in complex aquatic scenarios.

[0042] The trained AI model is compressed and optimized, including model pruning, quantization, and lightweighting, to ensure that the model can be deployed on the edge computing device of the unmanned ship and meet the requirements for real-time operation.

[0043] The specific steps of AI visual recognition models in identifying and extracting features from obstacle targets are as follows:

[0044] In the image feature extraction network of the AI ​​visual recognition model, convolutional feature extraction and multi-scale semantic enhancement processing are performed on the input water image data to generate image semantic features, specifically as follows:

[0045] Shallow convolutional layers are used to extract low-level edge and texture features, and ResNet, MobiNet and other backbone networks are used to extract multi-level features from the image.

[0046] Feature maps at multiple scales are input into the Feature Pyramid Network (FPN) or Atrous Spatial Pyramid Pooling (ASPP) module for multi-scale semantic enhancement;

[0047] By fusing deep semantic features and shallow spatial detail features through upsampling and lateral connection, a semantically rich and moderately resolution image semantic feature map is obtained.

[0048] In the obstacle target detection network of the AI ​​visual recognition model, obstacle target regions in water images are identified based on image semantic features, and the corresponding two-dimensional boundary range is output, specifically as follows:

[0049] Input image semantic features into YOLO, Faster-RCNN, or CenterNet object detection networks;

[0050] The two-dimensional bounding box coordinates of obstacle targets are predicted using either the anchor box mechanism or the center point mechanism. The output includes the center point position, width, height, and detection confidence of each target.

[0051] Redundant bounding boxes are removed using the non-maximum suppression (NMS) method, retaining only obstacle target regions with high confidence.

[0052] In the depth estimation network of the AI ​​visual recognition model, the spatial location and size information of the obstacle target are analyzed based on the obstacle target region and image semantic features. Specifically:

[0053] Input the obstacle target region and image semantic features into a depth estimation network (such as MonoDepth2, AnyNet, or a Transformer-based depth regression network);

[0054] Perform monocular depth estimation on pixels within each obstacle target area and output the depth value of each pixel;

[0055] By combining the camera's intrinsic parameter matrix and camera coordinate system calibration parameters, the pixel coordinates and depth values ​​of the two-dimensional image are calculated into three-dimensional spatial coordinates (X,Y,Z) through back projection.

[0056] Based on the clipping boundary of the 3D point cloud or depth map, calculate the actual size (e.g., length, width, height, or bounding box volume) of the obstacle target;

[0057] The output is the three-dimensional center coordinates (X,Y,Z) of each obstacle target in the ship's reference coordinate system, as well as the corresponding length, width and height dimensions, forming a spatial entity representation.

[0058] In the feature fusion network of the AI ​​visual recognition model, the boundary range, spatial location, and size information of the obstacle target are fused to generate the obstacle target feature set, which is as follows:

[0059] The 2D bounding box results from the object detection network are indexed, aligned, and feature-stitched with the 3D spatial coordinates and size information from the depth estimation network.

[0060] Use attention mechanisms or Transformer structures to enhance the contextual relationships between different pieces of information and improve the ability to express fused information;

[0061] Output a structured set of obstacle target features, including: the spatial position of the obstacle in the ship's coordinate system (3D coordinates), size information (length, width, and height), and boundary range (2D bounding box coordinates).

[0062] In this implementation scheme, a three-dimensional spatial perception mechanism with the unmanned vessel as the reference frame is constructed, enabling the system to map water area image pixels into spatial coordinate points with depth information. This accurately represents the relative position, size, and boundaries of obstacles in the water area. The method not only uses multi-scale convolutional networks to improve image semantic understanding capabilities but also combines depth estimation networks to restore spatial information of single-frame images, solving the obstacle perception problem under conditions such as water surface reflection and complex lighting. In the obstacle feature construction process, a complete target representation is formed by fusing two-dimensional bounding boxes and three-dimensional size data, further improving the accuracy and stability of subsequent path analysis. In addition, the system decouples and optimizes image feature extraction, obstacle detection, depth estimation, and feature fusion through a modular structure, making it easy to deploy on edge devices with limited computing resources for real-time processing. This enables the unmanned vessel to have stable and accurate perception and judgment capabilities in dynamic and complex water areas, providing highly reliable input for path planning and trajectory control.

[0063] Specifically, such as Figure 2 As shown, the specific steps for constructing the obstacle spatial distribution map are as follows: Read the spatial location, size information and boundary range of each obstacle. Specifically, read the obstacle target feature set output by the AI ​​visual recognition model, and extract the spatial location (three-dimensional center coordinate point), size information (width, height and depth) and two-dimensional bounding box range (coordinates of the upper left and lower right corners in the image plane) of each obstacle in the three-dimensional spatial coordinate system.

[0064] The spatial positions of each obstacle are projected into a three-dimensional spatial coordinate system with the center of the unmanned vessel as the origin. Specifically, the current position coordinates of the unmanned vessel are used as the origin, and a local vessel coordinate system is established in combination with its heading angle. The world coordinates of the obstacles are then transformed into the local three-dimensional coordinate system of the vessel using coordinate transformation methods to complete the position projection mapping.

[0065] Within a preset range in front of the hull, spatial grid cells are divided, and the spatial position and boundary range of each obstacle are marked in the corresponding grid cells. Specifically, three-dimensional grid volume cells are constructed within the inspection range set in front of the hull according to a fixed size (e.g., 1m×1m×1m). Boundary bounding box fitting is performed on each obstacle, and the grid cell into which its projection area falls is marked as the "obstacle area". At the same time, the corresponding obstacle number is recorded.

[0066] Based on the marking results of the grid cells, an obstacle spatial distribution map is generated. Specifically, a structured three-dimensional raster map data structure is constructed based on the obstacle marking information of each cell in the grid, and the occupancy status of each raster cell is output to represent the distribution of obstacles in space, serving as an input reference for subsequent path avoidance.

[0067] In this implementation scheme, by accurately mapping the identified obstacle features to a three-dimensional coordinate system under the unmanned vessel's hull reference frame, an effective transformation from image perception to spatial modeling is achieved. This method can structurally represent the spatial position, size information, and boundary range of obstacles in preset three-dimensional grid cells, and establish a dynamic coordinate reference by combining the current position and heading of the hull, so that the obstacle information always maintains spatial consistency during actual navigation. By dividing the space in front of the hull into grids and marking the grids of the obstacle projection area, a structured, queryable, and updatable three-dimensional obstacle distribution map can be quickly generated. This distribution map not only improves the system's perception accuracy of complex obstacle layouts, but also provides an intuitive and computable input basis for subsequent path planning, collision avoidance, and trajectory correction. It has high real-time performance and high scalability, and is particularly suitable for multi-obstacle and dynamically changing aquatic environments.

[0068] Specifically, the current navigation status parameters include the ship's position, speed, heading angle, and target route direction. The specific steps for analyzing and generating several avoidance paths are as follows: Set the ship's position and heading angle in the current navigation status parameters as the starting direction, and construct a candidate path search area in the waters in front of the ship. Specifically, with the current ship's position as the starting point and the current heading angle as the center direction, set a certain angle range (such as ±45°) and distance range (such as 100 meters) in front to form a fan-shaped or rectangular path search area, which serves as the spatial boundary for generating candidate paths.

[0069] The spatial position and boundary range of each obstacle in the obstacle spatial distribution map in the three-dimensional coordinate system are projected to the candidate path search area. Specifically, the coordinates and boundaries of each obstacle in the obstacle spatial distribution map are clipped in the ship's local coordinate system, retaining only the obstacle information located in the search area, and then reprojected onto the two-dimensional plane of the path search area for path collision detection.

[0070] Based on the speed and heading angle in the current navigation status parameters, the path search step size and angle threshold are set, and multiple heading offset paths are iteratively generated in the search area. Specifically, the travel distance within the time interval is calculated based on the current ship speed as the path step size, the heading offset angle is set (e.g., every 5°), and multiple path segments with different heading offsets starting from the current position are gradually generated within the set range to form an avoidance path candidate set.

[0071] Determine whether each heading offset path spatially overlaps with obstacles in the obstacle spatial distribution map, and filter out paths with collision risk. Specifically, perform spatial overlap detection between each offset path and the grid area where the obstacle is located. If the point passed by the path overlaps with any obstacle grid cell, the path is marked as "collision risk" and removed; otherwise, it is marked as "collision-free path".

[0072] Set the remaining set of collision-free paths as avoidance paths.

[0073] In this implementation scheme, a dynamic, refined, and targeted path planning mechanism is achieved by introducing collaborative analysis of current navigation state parameters and spatial obstacle distribution maps. A forward path search area is constructed centered on the ship's current position and heading angle, ensuring that all candidate paths are within a reasonable navigation range. By cropping and mapping the 3D information from the obstacle spatial distribution map onto a 2D plane, potential risk areas within the forward navigation area can be accurately identified, enabling efficient collision detection with the path. Simultaneously, by combining ship speed and heading angle to set the path step size and offset angle, multiple heading offset paths are generated iteratively, constructing a path candidate set covering various avoidance directions. This set possesses good spatial coverage and maneuverability. Through overlap detection of each path with the obstacle grid, paths with collision risks can be eliminated, selecting the optimal set of collision-free avoidance paths. This provides highly reliable input for subsequent trajectory correction of the unmanned vessel. This method is logically clear, highly adaptable, and capable of handling complex aquatic environments with multiple obstacles and changing headings, effectively improving the navigation stability and obstacle avoidance success rate of unmanned vessels in dynamic environments.

[0074] Specifically, the steps for selecting the optimal avoidance path are as follows: Based on the ship's position, speed and heading angle in the current navigation state parameters, analyze the navigation offset and the deviation angle of the target route corresponding to each avoidance path. Specifically, for each avoidance path, calculate the angle between its endpoint and the current heading as the deviation angle, and at the same time calculate the Euclidean distance from the ship's current position to the endpoint of the path as the offset, and compare it with the direction of the target route to form a heading deviation index.

[0075] From the obstacle spatial distribution map, extract the minimum safe distance and path length of each avoidance path from the obstacles. Specifically, in the obstacle distribution map, traverse the spatial distance between each avoidance path and all obstacles, record the minimum value as the minimum safe distance; at the same time, calculate the path length based on the sum of the distances between all points on the path.

[0076] Based on the navigation offset, the target route deviation angle, the minimum safe distance, and the path length, the optimal avoidance path is analyzed.

[0077] In this implementation plan, a multi-factor quantitative evaluation mechanism is used to improve the scientific nature and safety of unmanned vessels' path avoidance decisions in complex water environments. The system first combines the current speed, position, and heading angle of the vessel to quantify the deviation between the endpoint of each avoidance path and the target route, and calculates the deviation angle and offset to ensure that the path avoids obstacles while staying as close as possible to the original route, reducing path redundancy caused by deviation. Secondly, by analyzing the minimum safe distance between each path and obstacle and the total path length, the system evaluates the safety margin and navigation efficiency of the path. The deviation angle, offset, minimum safe distance, and path length are uniformly incorporated into the path evaluation index system to effectively screen out the optimal avoidance path under the current environmental conditions. This method fully considers various physical characteristics and spatial relationships, and significantly improves the intelligent obstacle avoidance and autonomous path correction capabilities of unmanned vessels.

[0078] Specifically, the steps for analyzing the optimal avoidance path are as follows: weighted analysis is performed on the navigation offset, target route deviation angle, minimum safe distance, and path length to obtain a path score for each avoidance path; the path scores of all avoidance paths are compared and analyzed, and the avoidance path with the highest path score is selected as the optimal avoidance path.

[0079] In this implementation plan, by uniformly weighting the evaluation of navigation offset, target route deviation angle, minimum safe distance, and path length, a quantitative comparison and intelligent selection of all avoidance paths are achieved. This enables the system to adaptively select the current optimal navigation path in a multi-objective trade-off. Compared with traditional binary judgment (such as collision detection) methods, this scoring mechanism is more flexible and intelligent. It can select an ideal path that is both safe and close to the mission route from multiple feasible paths. It is especially suitable for complex environments with dense obstacles or limited path selection space. At the same time, this mechanism provides a clear basis for subsequent path execution and control command generation, ensuring the continuous and stable operation of the unmanned vessel in a dynamic water environment.

[0080] Specifically, the unmanned vessel collects and uploads water quality testing data in real time during its inspection process, including the following steps: During the unmanned vessel's inspection, water quality testing data at the inspection points is acquired in real time and analyzed. The water quality testing data includes pH value, conductivity, dissolved oxygen concentration, and turbidity. Specifically, as the unmanned vessel cruises to each inspection point, it acquires the pH value, conductivity, dissolved oxygen concentration, and turbidity of the target water body in real time through its onboard water quality sensor module; the various test data are normalized (units removed), and then weighted and summed according to set weighting coefficients (e.g., pH value weight 0.25, conductivity 0.25, dissolved oxygen concentration 0.30, turbidity 0.20) to generate the corresponding water quality score for each point.

[0081] Water quality testing data and water quality analysis results are uploaded to a pre-set remote data center in real time. Specifically, after data collection and water quality score calculation are completed at each inspection point, the raw testing data and corresponding score results are packaged in a unified data format through a shipborne communication module (such as 4G / 5G or satellite communication) and uploaded to the set remote data center server through a data transmission protocol for subsequent water quality trend analysis or early warning.

[0082] In this implementation plan, real-time data collection, standardized analysis, and remote synchronous transmission of water quality data are achieved during unmanned vessel inspections, significantly improving the intelligence and response efficiency of water environment monitoring. By equipping the system with multi-parameter water quality sensors (such as pH, conductivity, dissolved oxygen, and turbidity), the system can automatically complete on-site sampling and testing as the vessel cruises to each inspection point. It can also quickly generate water quality scores through normalization and weighted models to quantify the water quality status at each point. This structured scoring mechanism takes into account multiple indicator characteristics, facilitating intuitive evaluation and trend comparison. At the same time, the communication module based on 4G / 5G or satellite links supports stable remote data upload, enabling simultaneous measurement and transmission, eliminating the need for manual data collection and delayed processing. This makes it particularly suitable for real-time monitoring of large water areas, unmanned areas, or environmentally sensitive areas.

[0083] Please see Figure 3This invention provides a technical solution: a BeiDou-based AI-powered unmanned vessel tracking and control system, comprising: a cruise perception unit, used to acquire in real time the BeiDou positioning information, current navigation status parameters, and water area image data in front of the vessel when the unmanned vessel is cruising in the target inspection water area according to the target route; an obstacle identification unit, used to input the water area image data into a pre-trained AI visual recognition model to determine whether there are obstacle targets; an obstacle modeling unit, used to extract the obstacle target feature set and construct an obstacle spatial distribution map when an obstacle target is identified; an obstacle avoidance path unit, used to analyze and generate several avoidance paths based on the obstacle spatial distribution map and the current navigation status parameters, and select the optimal avoidance path; and a trajectory control unit, used to adjust the current navigation status parameters based on the optimal avoidance path and synchronously correct the target route in conjunction with the BeiDou positioning information.

[0084] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0085] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for telemetry and control of unmanned vessels based on BeiDou artificial intelligence, characterized in that, Includes the following steps: When the unmanned ship is cruising in the target inspection area along the target route, it acquires the Beidou positioning information, current navigation status parameters and water image data of the water area in front of the ship in real time. Water area image data is input into a pre-trained AI visual recognition model to determine whether there are obstacle targets. If an obstacle target is identified, the feature set of the obstacle target is extracted, and a spatial distribution map of the obstacle is constructed; Based on the obstacle spatial distribution map and current navigation status parameters, several avoidance paths are generated, and the optimal avoidance path is selected. Based on the optimal avoidance path, the current navigation status parameters are adjusted, and the target route is synchronously corrected in conjunction with BeiDou positioning information.

2. The BeiDou-based AI-powered unmanned vessel telemetry and control method according to claim 1, characterized in that, The water area image data includes several water area pixels, and each water area pixel corresponds to a water area pixel value and a three-dimensional coordinate.

3. The BeiDou-based AI-powered unmanned vessel telemetry and control method according to claim 1, characterized in that, The obstacle target feature set includes the spatial location, size information, and boundary range of several obstacles. The AI ​​visual recognition model includes an image feature extraction network, an obstacle target detection network, a depth estimation network, and a feature fusion network.

4. The BeiDou-based AI-powered unmanned vessel telemetry and control method according to claim 3, characterized in that, The specific steps of AI visual recognition models in identifying and extracting features from obstacle targets are as follows: In the image feature extraction network of the AI ​​visual recognition model, convolutional feature extraction and multi-scale semantic enhancement processing are performed on the input water image data to generate image semantic features; In the obstacle target detection network of the AI ​​visual recognition model, obstacle target regions in water images are identified based on image semantic features, and the corresponding two-dimensional boundary range is output. In the depth estimation network of the AI ​​visual recognition model, the spatial location and size information of the obstacle target are analyzed based on the obstacle target region and image semantic features; In the feature fusion network of the AI ​​visual recognition model, the boundary range, spatial location and size information of the obstacle target are fused to generate the obstacle target feature set.

5. The BeiDou-based AI-powered unmanned vessel telemetry and control method according to claim 3, characterized in that, The specific steps for constructing an obstacle spatial distribution map are as follows: Read the spatial location, size information, and boundary range of each obstacle; The spatial positions of each obstacle are projected onto a three-dimensional spatial coordinate system with the center of the unmanned vessel as the origin; Divide the space into spatial grid cells within a preset area in front of the hull, and mark the spatial position and boundary range of each obstacle in the corresponding grid cell; An obstacle spatial distribution map is generated based on the marking results of the grid cells.

6. The BeiDou-based AI-powered unmanned vessel telemetry and control method according to claim 5, characterized in that, The current navigation status parameters include the ship's position, speed, heading angle, and target course direction. The specific steps for analyzing and generating several avoidance paths are as follows: The candidate path search area is constructed in the waters in front of the ship, with the ship's position and heading angle in the current navigation status parameters as the starting direction. Project the spatial location and boundary range of each obstacle in the obstacle spatial distribution map into the candidate path search area in the three-dimensional coordinate system; Based on the speed and heading angle in the current navigation state parameters, set the path search step size and angle threshold, and iteratively generate multiple heading offset paths in the search area; Determine whether each heading deviation path spatially overlaps with obstacles in the obstacle spatial distribution map, and filter out paths with collision risk; Set the remaining set of collision-free paths as avoidance paths.

7. The BeiDou-based AI-powered unmanned vessel telemetry and control method according to claim 6, characterized in that, The specific steps for selecting the optimal avoidance path are as follows: Based on the ship's position, speed and heading angle in the current navigation status parameters, analyze the navigation offset and target course deviation angle corresponding to each avoidance path; From the obstacle spatial distribution map, extract the minimum safe distance and path length between each avoidance path and the obstacle; Based on the navigation offset, the target route deviation angle, the minimum safe distance, and the path length, the optimal avoidance path is analyzed.

8. The BeiDou-based AI-powered unmanned vessel telemetry and control method according to claim 7, characterized in that, The specific steps for analyzing the optimal avoidance path are as follows: A weighted analysis of navigation offset, target route deviation angle, minimum safe distance, and path length is performed to obtain a path score for each avoidance path; The path scores of all avoidance paths are compared and analyzed, and the avoidance path with the highest path score is selected as the optimal avoidance path.

9. The BeiDou-based AI-powered unmanned vessel telemetry and control method according to claim 1, characterized in that, The process of collecting and uploading water quality testing data in real time during unmanned vessel inspections includes the following steps: During the unmanned vessel's inspection, water quality data at the inspection points are acquired in real time and analyzed. The water quality data includes pH value, conductivity, dissolved oxygen concentration, and turbidity. Water quality testing data and water quality analysis results are uploaded to a pre-set remote data center in real time.

10. A BeiDou-based AI-powered unmanned vessel telemetry and control system, employing the BeiDou-based AI-powered unmanned vessel telemetry and control method as described in any one of claims 1-9, characterized in that... include: The cruise perception unit is used to acquire the BeiDou positioning information, current navigation status parameters and water image data of the water in front of the ship in real time when the unmanned ship is cruising in the target inspection water area according to the target route. The obstacle recognition unit is used to input water area image data into a pre-trained AI visual recognition model to determine whether there are obstacle targets. The obstacle modeling unit is used to extract the feature set of obstacle targets and construct a spatial distribution map of obstacles when they are identified. The avoidance path unit is used to analyze and generate several avoidance paths based on the obstacle spatial distribution map and the current navigation status parameters, and select the optimal avoidance path. The trajectory control unit is used to adjust the current navigation status parameters based on the optimal avoidance path and to synchronously correct the target route by combining BeiDou positioning information.

Images