An unmanned ship and a method for cooperative operation thereof

Abstract

The application provides a UAV and unmanned ship cooperative operation system and method, and relates to the technical field of UAV and unmanned ship cooperative operation.The system comprises a UAV, an unmanned ship and a ground station;the UAV comprises blades for providing power for the UAV, which are fixed on the UAV arms through screws;the UAV is provided with a wireless communication module;the UAV is provided with a depth vision camera for collecting obstacle information in a scene;the unmanned ship comprises a wireless communication module;the unmanned ship is provided with a propeller for providing power for the unmanned ship;the ground station comprises a wireless communication module;and the ground station is provided with a display screen module for displaying a digital map transmitted by the UAV.Through the cooperative control of the UAV, the unmanned ship and the ground station, and in combination with deep learning and intelligent algorithms, the application realizes efficient cooperative operation of polar sea ice monitoring and unmanned platforms.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle and unmanned surface vehicle cooperative operation, and particularly relates to an unmanned aerial vehicle and unmanned surface vehicle cooperative operation system and a method thereof. BACKGROUND

[0002] In the development of polar operation, such as ocean resource exploration, environmental monitoring and other wide sea monitoring fields, unmanned systems play an increasingly important role. At present, most of the tasks are executed by using a single unmanned system, however, this scheme has some drawbacks: on the one hand, the single platform has limited load capacity, and the battery efficiency drops sharply under the condition of extremely low temperature, making it difficult to complete the multi-dimensional data collection simultaneously; on the other hand, the navigation accuracy decreases due to the complex polar ice conditions, and the reliability of the single obstacle avoidance system is insufficient.

[0003] In order to promote the intelligent and efficient development of polar operation, there is an urgent need for an efficient and flexible unmanned aerial vehicle and unmanned surface vehicle cooperative operation system and the corresponding cooperative operation method which can meet the base operation requirements. SUMMARY

[0004] In view of the technical problems of limited load capacity of a single platform, difficulty in completing multi-data collection, decreased navigation accuracy and difficulty in guaranteeing system obstacle avoidance in the prior art, the present application provides an unmanned aerial vehicle and unmanned surface vehicle cooperative operation system and a method thereof.

[0005] To achieve the above purpose, the technical solution adopted by the present application is as follows: an unmanned aerial vehicle and unmanned surface vehicle cooperative operation system, the system comprising: an unmanned aerial vehicle, an unmanned surface vehicle and a ground station;

[0006] The unmanned aerial vehicle comprises: a paddle for providing power to the unmanned aerial vehicle, fixed to the unmanned aerial vehicle arm by a screw; an unmanned aerial vehicle wireless communication module; and a depth vision camera for collecting obstacle information in the scene.

[0007] The unmanned surface vehicle comprises: an unmanned surface vehicle wireless communication module; and an unmanned surface vehicle propeller for providing power to the unmanned surface vehicle.

[0008] The ground station comprises: a ground station wireless communication module; and a display screen module for displaying a digital map transmitted by the unmanned aerial vehicle.

[0009] The unmanned aerial vehicle is provided with an A platform, the unmanned surface vehicle is provided with a B platform, and the ground station is provided with a C platform; the A platform constructs a digital map and identifies a sea ice area during operation, and after completion, the data are transmitted to the C platform and standby; the C platform performs path planning after manually setting the target position of the B platform, and sends a control instruction; the B platform receives the control instruction, moves to the target position according to the planned path, and then stands by; wherein the path planning adopts a deep Q network (DQN) algorithm, which comprises:

[0010] Initialize the parameters of the UAV system, the USV system, the main network, the target network, the experience replay buffer; set the hyperparameters and reward-related parameters; reset the environment and obtain the initial state;

[0011] Then, training is performed, specifically including:

[0012] Loop each training episode: reset the environment, initialize the positions of the UAV and the USV; set the total reward and the step count to zero; obtain the initial sea ice distribution and the navigable area information;

[0013] Loop each step: select the moving action of the USV according to the ε-greedy policy; the UAV performs the sea ice identification and map construction tasks; execute the moving action of the USV to obtain the next state and the two key information of whether to end; calculate the reward; store the experience in the experience replay buffer;

[0014] If the experience replay buffer sample is sufficient: extract the sample; calculate the target action value function Q value and the current action value function Q value; calculate the loss and update the main network;

[0015] If the update frequency is met, update the target network parameters, which means that the number of training steps reaches a specified number;

[0016] Update the states of the UAV and the USV, increment the step count, and accumulate the reward;

[0017] If it is finished, record the information and exit the loop;

[0018] During the entire reinforcement learning training loop, each step of the USV produces an immediate reward, which is determined by the reward function; the immediate reward is accumulated in each training episode to form a cumulative reward, and the goal of the algorithm is to maximize the cumulative reward by continuously adjusting the policy, so that the USV learns to complete the task efficiently and safely;

[0019] Reward function for approaching the target The design is as follows:

[0020]

[0021] In the formula, is the distance between the current USV and the target, is the scaling coefficient;

[0022] The USV is also rewarded according to the change in its heading, with a positive reward if the heading change is small, and a negative reward otherwise. The reward expression is as follows:

[0023]

[0024] In the formula, an angle of change in the heading of the unmanned ship, a threshold value of the change in the heading, a maximum angle of change in the heading.

[0025] In another aspect, the present application provides a method for the unmanned ship cooperative operation system described above, the method comprising:

[0026] Step 1, system initialization, completing the parameter setting and function initialization of the unmanned aerial vehicle and the unmanned ship;

[0027] Step 2, reading the initial position information according to the set initial position information;

[0028] Step 3, the unmanned aerial vehicle reads the target position information set by the ground station;

[0029] Step 4, the unmanned aerial vehicle reaches the target position through a motion control algorithm;

[0030] Step 5, judging whether the target position is the set target position, if the unmanned aerial vehicle reaches the set target position, entering Step 6, if not, turning to Step 3;

[0031] Step 6, the unmanned aerial vehicle reaches the set target position, and the unmanned aerial vehicle hovers at a constant height;

[0032] Step 7, the unmanned aerial vehicle carries out constant-speed rotation with a depth vision camera, and the camera collects the object image of the current area;

[0033] Step 8, constructing a digital map through the depth vision camera;

[0034] Step 9, identifying sea ice in the image through a deep learning algorithm;

[0035] Step 10, marking the navigable area on the constructed digital map, and after completion, entering Step 11;

[0036] Step 11, transmitting the map information from the unmanned aerial vehicle end to the ground station through wireless communication technology, and after completion, entering Step 18;

[0037] Step 12, judging whether the unmanned aerial vehicle data transmission is completed, if completed, entering Step 13, if not, entering Step 11;

[0038] Step 13, the unmanned aerial vehicle standby;

[0039] Step 14, setting the unmanned ship moving end position through the ground station;

[0040] Step 15, setting the moving path of the unmanned ship through a path planning algorithm;

[0041] Step 16, send ground station control instructions to unmanned boat through wireless communication technology;

[0042] Step 17, unmanned boat executes control instructions and moves to designated location;

[0043] Step 18, judge whether unmanned boat reaches designated location, if yes, enter step 19, if not, enter step 16;

[0044] Step 19, system enters standby state, waits for new task instructions, and the process ends.

[0045] Compared with the prior art, the beneficial effects of the present application are:

[0046] 1. The present application breaks through the limitation of single platform capacity through multi-platform cooperation:

[0047] In view of the fact that traditional single unmanned aerial vehicle or single unmanned boat is limited by load capacity, endurance time or sensor coverage range, it is difficult to independently complete the multi-data collection and obstacle avoidance task in complex sea area. The unmanned aerial vehicle of the present application is equipped with a depth camera and a wireless communication module, responsible for high-altitude sea ice image collection and digital map construction; the unmanned boat serves as an execution platform, responsible for low-altitude or water surface navigation area operation; the ground station serves as the hub, overall plans the path planning and control instruction issuing, so as to realize the multi-platform cooperation.

[0048] 2. Sea ice identification based on deep learning improves monitoring accuracy:

[0049] In view of the fact that traditional sea ice monitoring relies on manual visual interpretation or low-precision sensors, it is difficult to cope with complex light, climate and sea ice shape changes. The present application adopts YOLO target detection algorithm model, optimizes network parameters through training set, realizes rapid and accurate identification of sea ice target (detection accuracy and recall rate are significantly improved). After model deployment, it supports automatic identification of newly collected images, and the inference speed meets the real-time monitoring demand.

[0050] 3. Intelligent path planning and obstacle avoidance ensure operation safety:

[0051] In view of the fact that the polar environment is complex, and the traditional path planning algorithm is difficult to dynamically adapt to the distribution of obstacles and the change of speed. The present application adopts three-dimensional coordinate fusion: combining the three-dimensional target position provided by the depth camera of the unmanned aerial vehicle and the real-time position of the inertial navigation system of the unmanned boat, the relative position relationship is calculated, and the path planning accuracy is improved.

[0052] DQN algorithm application: introduce deep reinforcement learning algorithm (DQN), generate optimal path strategy through training, consider speed, heading and obstacle avoidance, which is more adaptable than traditional A or Dijkstra algorithm.

[0053] Dynamic feedback adjustment: the propulsion system adjusts the speed and heading command in real time according to the path planning result, inertial navigation and camera feedback data, to ensure the unmanned ship to avoid obstacles smoothly and reach the target position accurately.

[0054] Other features and advantages of the embodiments of the present application will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 is a structure diagram of the unmanned ship cooperative operation system according to the present application;

[0056] Figure 2 is a work flow diagram of the unmanned ship cooperative operation system according to the present application. DETAILED DESCRIPTION

[0057] In order to make the personnel in the art better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application, so that the purpose, characteristics and advantages of the present application can be better understood. It should be understood that the embodiments shown in the drawings are not a limitation on the scope of the present application, but are only to illustrate the essential spirit of the technical solutions of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts should belong to the scope of protection of the present application.

[0058] Unless the context clearly requires otherwise, throughout the description and the claims, the words "comprise", "comprising", and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense; that is to say, in the sense of "including, but not limited to".

[0059] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of the phrase "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0060] As used in this specification and the appended claims, the singular forms "a", "an", and "the" include plural referents unless the content clearly dictates otherwise. It should be noted that the term "comprising" as used in this specification and the appended claims is to be construed as meaning "including, but not limited to".

[0061] In the following description, in order to clearly show the structure and working mode of the present application, many directional words will be described, but "front", "back", "left", "right", "outer", "inner", "outward", "inward", "up", "down" and the like should be understood as convenient language, and should not be understood as limiting language.

[0062] The implementation details of the embodiments of the present application will be specifically described below with reference to the accompanying drawings. The following implementation details are provided for the convenience of understanding, and are not essential for implementing the present application.

[0063] The present application proposes a water surface sea ice identification system relying on a UAV, and proposes a UAV-boat cooperative control system and a corresponding control method, so that the UAV and the boat can realize cooperative operation of the sea area through mutual cooperation.

[0064] The present application combines Figure 1 The UAV-boat cooperative operation system is described. The core components of the UAV-boat cooperative operation system include three parts, which are: a UAV, a UAV boat and a ground station. The system is provided with a UAV, a UAV boat double system platform, and a ground station monitoring and control platform.

[0065] Among them, the UAV includes: A1 is the propeller of the UAV, which is used to provide power for the UAV and is fixed on the UAV arm through a screw; A2 is a UAV on-board wireless communication module for wireless information transmission; A3 is a depth vision camera for collecting obstacle information in the scene. Among them, the depth vision camera (depth camera, also known as 3D camera) is a camera device that can capture the distance information between the object and the camera and generate a three-dimensional scene containing depth data.

[0066] The UAV boat includes: B1 is a wireless communication module of the UAV boat, which is used for wireless information transmission; B2 is a UAV boat propeller, which is used to provide power for the UAV boat.

[0067] The ground station includes: C1 is a wireless communication module of the ground station, which is used for wireless information transmission; C2 is a display screen module, which is used to display the digital map transmitted by the UAV; C3 is a ground station keyboard and mouse module, which is used to set the motion parameter information.

[0068] Specifically, during operation, the UAV collects information such as sea ice and obstacles on the sea surface through the depth vision camera. When the UAV collects data above the sea level, the UAV boat is located at the initial position. After the UAV obtains the digital map and the sea ice position, the map information is sent to the ground station, and the UAV waits after the sending is completed. Set the end point of the UAV boat motion on the ground station and plan the path. The ground station sends the mobile control instruction to the UAV boat through wireless communication technology, and the UAV boat waits after starting to move to the specified position.

[0069] In some embodiments, the present application provides a three-platform cooperative operation mode for the unmanned ship cooperative operation system. Specifically, the unmanned aerial vehicle is provided with an A platform, the unmanned ship is provided with a B platform, and the ground station is provided with a C platform. During operation, the A platform constructs a digital map and identifies the sea ice area through a depth vision camera. When the A platform obtains data information in the target area, the B platform waits at the shore end. After the A platform completes the digital map construction, the data is transmitted to the C platform by the unmanned aerial vehicle wireless communication technology, and the A platform waits after completion. The C platform starts to wait for the manual setting of the target position of the B platform. After obtaining the target position of the B platform, the C platform starts to plan the path and sends control instructions to the B platform through wireless communication technology. At the same time, after receiving the control instructions, the B platform starts to move to the specified target position area based on the planned path, and waits after the movement is completed. This cycle is repeated to realize the cooperative operation of the unmanned aerial vehicle and the unmanned ship.

[0070] Further, the best position calculation and path planning of the unmanned ship are as follows:

[0071] The ground station combines the target position collected by the unmanned aerial vehicle depth vision camera (three-dimensional coordinates provided by the depth vision camera) with the current position of the unmanned ship (position information provided by the inertial navigation system of the unmanned ship) to calculate the relative position of the target relative to the unmanned ship. A computer vision and path planning algorithm (DQN algorithm) is used to generate the optimal path of the unmanned ship from the current position to the target position. Obstacles, speed, heading, and other factors are considered to generate a smooth and obstacle-avoiding path. The processor calculates the heading adjustment instructions that the unmanned ship should take based on the target position and heading information. Based on the target position, path planning results, inertial navigation system, and camera feedback, the processor generates control instructions for the unmanned ship propulsion system. The control instructions include: speed instructions, which generate the speed that the unmanned ship should reach based on the target position and the speed of the path planning; heading instructions, which calculate the angular velocity or turning angle required to adjust the heading based on the target direction; execute the propulsion control instructions, and the propulsion system adjusts the thrust and direction of the propeller according to the control instructions to ensure that the unmanned ship moves towards the target position. The heading and speed of the unmanned ship are adjusted based on the real-time feedback data from the inertial navigation system and the camera to ensure accurate task execution.

[0072] In some embodiments, for the unmanned aerial vehicle and ship cooperative operation system of the present application, the following workflow diagram is combined Figure 2 to describe the setting of full function operation, and the steps are as follows:

[0073] Step 1, system initialization, complete parameter setting and function initialization of the unmanned aerial vehicle and the unmanned ship, and enter step 2 after completion;

[0074] Step 2, if there is initial position information set, read the position information, including the set flight height, the latitude and longitude of the specified position, etc., and enter step 3 after completion;

[0075] Step 3, the unmanned aerial vehicle reads the target point position set by the ground station, and enters step 4 after completion;

[0076] Step 4, the unmanned aerial vehicle reaches the set position through the motion control algorithm, and enters step 5 after completion;

[0077] Step 5, judgment, if the unmanned aerial vehicle reaches the set position, enter step 6, if not, enter step 3;

[0078] Step 6, if the unmanned aerial vehicle reaches the set position, the unmanned aerial vehicle hovers at a constant height, and enters step 7 after completion;

[0079] Step 7, the unmanned aerial vehicle carries a depth vision camera and rotates at a constant speed, and the camera collects the objects in the current area, and enters step 8 after completion;

[0080] Step 8, through the on-board depth vision camera, a digital map is constructed, and step 9 is entered after completion;

[0081] Step 9, load the YOLO target detection model and its weight parameters trained on the sea ice data set in advance;

[0082] Obtain the image or video frame to be detected by the acquisition device to obtain the sea ice region image to be recognized, and input the original sea ice image into the recognition system;

[0083] Through size adjustment, pixel normalization and data format conversion and other preprocessing operations on the input sea ice image, the image meets the input requirements of the YOLO target detection model, and the accuracy and efficiency of subsequent identification are improved;

[0084] By inputting the preprocessed sea ice image into the YOLO target detection model, performing forward inference operation, obtaining the original detection result containing sea ice target candidate box, class probability and confidence score;

[0085] Through post-processing of the target candidate box output by the YOLO target detection model, including screening valid sea ice targets according to the confidence threshold, using the non-maximum suppression (NMS) algorithm to remove overlapping redundant boxes, and finally obtaining accurate sea ice detection results; wherein, the non-maximum suppression (Non-Maximum Suppression, NMS) is a core post-processing algorithm in computer vision tasks such as target detection and image segmentation, which is used to eliminate redundant detection boxes or segmentation regions and retain the most representative results;

[0086] The visualization of the sea ice detection result is realized by drawing the detected sea ice target bounding box, class label and confidence score on the original image, and the result image is displayed on the user interface or stored in the specified location according to the requirement, which is convenient for subsequent analysis and application;

[0087] By completing the above steps, the automatic recognition process of the input sea ice image is realized, and the system enters the standby or end state, waiting for a new recognition task, and then enters step 10;

[0088] Step 10, mark the navigable area on the constructed digital map, and then enter step 11;

[0089] Step 11, transmit the map information from the unmanned aerial vehicle to the ground station through wireless communication technology, and then enter step 12;

[0090] Step 12, judge whether the data transmission of the unmanned aerial vehicle is completed, if yes, enter step 13, if not, enter step 11;

[0091] Step 13, the unmanned aerial vehicle is on standby, and then enter step 14;

[0092] Step 14, set the unmanned boat moving end position through the ground station, and then enter step 15;

[0093] Step 15, through the path planning algorithm, the unmanned boat path planning and obstacle avoidance are carried out, and the unmanned boat motion control information is obtained, and then enter step 16;

[0094] Step 16, send the ground station control instruction to the unmanned boat through wireless communication technology, and then enter step 17;

[0095] Step 17, the unmanned boat executes the control instruction and moves to the specified position, and then enters step 18;

[0096] Step 18, judge whether the unmanned boat reaches the specified position, if yes, enter step 19, if not, enter step 16;

[0097] Step 19, the system enters the standby state, waiting for a new task instruction, and then the process ends.

[0098] In some embodiments, the YOLO target detection model in step 9 is trained based on the sea ice data set, wherein YOLO is a real-time target detection algorithm, which adopts an advanced neural network architecture and has the characteristics of fast detection speed and high accuracy, and its core idea is to regard the target detection task as a regression problem, so as to realize end-to-end detection, and the training process is as follows:

[0099] Step1. Collect raw image data containing sea ice distribution through camera devices set in the target area, such as drones or shore-based cameras, ensuring that the collection process covers a variety of scenarios with different climates, lighting, and sea ice morphologies.

[0100] Step2. Clean the collected raw sea ice images by removing blurred, excessively noisy, repetitive, or invalid image data, and select a high-quality, sufficient, and evenly distributed dataset to improve the effectiveness and generalization ability of subsequent model training.

[0101] Step3. Perform data augmentation on the selected dataset, including rotation, scaling, cropping, color transformation, and other operations to further enrich sample diversity and improve the model's ability to recognize sea ice targets in different scenarios.

[0102] Step4. Accurately label the sea ice targets in the sea ice images using professional labeling software to generate label files containing target categories and bounding box coordinates; then divide the labeled dataset into training, testing, and validation sets according to a certain proportion to meet the needs of model training and evaluation.

[0103] Step5. Build a YOLO-based target detection algorithm framework, train the model using the training set, continuously optimize network parameters, and obtain the optimal training model suitable for sea ice recognition.

[0104] Step6. Evaluate the performance of the trained model using the validation and test sets, including detection accuracy, recall rate, inference speed, and other indicators to ensure the effectiveness and robustness of the model in actual applications.

[0105] Step7. Deploy the trained and evaluated model to the actual application system to achieve automatic recognition and detection of newly collected sea ice images, providing technical support for sea ice monitoring and management.

[0106] In some embodiments, the path planning algorithm of step 15 uses the Deep Q-Network (DQN) algorithm, with the following steps:

[0107] S1, Initialization: Initialize the UAV system, unmanned boat system, main network, target network parameters, experience replay buffer; set hyperparameters and reward-related parameters; reset the environment and obtain the initial state (including sea ice distribution, navigable area, etc.);

[0108] S2, Training process:

[0109] Loop each training episode: environment reset, initialize the position of UAV and USV; set the round reward sum and step number to zero; get the initial sea ice distribution and navigable area information;

[0110] Loop each step: select the moving action of USV according to the ε-greedy strategy; UAV performs sea ice identification and map construction tasks; execute the moving action of USV, after the USV performs an action, the environment will change, the system will get two key information, the next state and whether it is over; calculate the reward (combine factors such as approaching the target, reaching the target, collision, path smoothness, etc.); store the experience in the experience replay buffer; wherein, the ε-greedy strategy (Epsilon-Greedy Strategy) is a classic action selection strategy in reinforcement learning (Reinforcement Learning, RL), which is used to balance the contradiction between exploration and utilization. Its core idea is: select a random action with probability ε (exploration), and select the current optimal action with probability 1-ε (utilization);

[0111] If the experience replay buffer sample is sufficient: extract the sample; calculate the target action value function Q value and the current action value function Q value; calculate the loss and update the main network;

[0112] The update frequency determines the occurrence of the action of updating the target network parameters, if the update frequency is met: update the target network parameters; here, meeting the update frequency means that the number of training steps reaches the specified number;

[0113] Update the state of the UAV and the USV, the step number plus one, and the cumulative reward;

[0114] If it is over, record the information and jump out of the loop.

[0115] In the entire reinforcement learning training loop, each step of the USV will produce an instant reward (Instant Reward). The calculation rule of this instant reward is determined by the reward function composed of factors such as "approaching the target, reaching the target, collision, path smoothness". All these instant rewards are accumulated in each training episode to form a cumulative reward (Cumulative Reward). The goal of the algorithm is to maximize this cumulative reward by constantly adjusting the strategy, so that the USV learns how to efficiently and safely complete the task.

[0116] Reward function for approaching the target The design is as follows, which makes it more smooth and can effectively guide the USV to gradually approach the target:

[0117]

[0118] wherein, wherein d is the distance between the current unmanned surface vehicle and the target, wherein d is the distance between the current unmanned surface vehicle and the target,

[0119] wherein d is the distance between the current unmanned surface vehicle and the target,

[0120]

[0121] wherein d is the distance between the current unmanned surface vehicle and the target, wherein d is the distance between the current unmanned surface vehicle and the target, wherein d is the distance between the current unmanned surface vehicle and the target, wherein d is the distance between the current unmanned surface vehicle and the target,

[0122] The unmanned surface vehicle cooperative operation system and method provided by the present application realize efficient cooperative operation in polar sea areas through the three-platform cooperative mode of unmanned aerial vehicles, unmanned surface vehicles and ground stations, combined with deep learning and intelligent control technology: the unmanned aerial vehicle carries a deep vision camera to collect sea ice distribution and obstacle information, constructs a digital map of the polar region and identifies a navigable area, and then transmits the data to the ground station; the ground station calculates the relative position based on the three-dimensional target coordinates provided by the unmanned aerial vehicle and the inertial navigation information of the unmanned surface vehicle, and generates an optimal path that takes into account obstacle avoidance, speed and heading using the DQN algorithm, while sending propulsion control parameters containing speed and heading instructions to the unmanned surface vehicle through wireless communication; the unmanned surface vehicle adjusts the thrust and direction according to the instructions, and continuously corrects the heading and speed in combination with inertial navigation and camera real-time feedback, and finally accurately reaches the target position and waits, significantly improving the automation level and task execution accuracy of polar monitoring.

[0123] Although the present application has been described in detail with reference to the preferred embodiments, the present application is not limited to the preferred embodiments. Those skilled in the art can make various equivalent modifications or replacements to the embodiments of the present application without departing from the spirit and essence of the present application, and these modifications or replacements should be within the scope of the present application or any skilled person in the art can easily think of changes or replacements within the technical scope disclosed by the present application, and all should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An unmanned ship cooperative operation system, characterized by, The system comprises a UAV, an unmanned boat and a ground station; The UAV comprises blades for providing power to the UAV, fixed to the UAV arm by screws, a UAV wireless communication module, and a depth vision camera for collecting obstacle information in a scene; The unmanned boat comprises an unmanned boat wireless communication module and an unmanned boat propeller for providing power to the unmanned boat; The ground station comprises a ground station wireless communication module and a display screen module for displaying a digital map transmitted by the UAV; The UAV is provided with an A platform, the unmanned boat is provided with a B platform, and the ground station is provided with a C platform; the A platform constructs a digital map and identifies a sea ice area when operating, and after completion, transmits data to the C platform and waits; the C platform waits after manually setting a target position of the B platform, and then performs path planning and sends a control instruction; the B platform receives the control instruction, moves to the target position according to the planned path, and then waits; wherein the path planning adopts a deep Q network (DQN) algorithm, comprising: First, initialize the UAV system, the unmanned boat system, the main network, the target network parameters, and the experience replay buffer; set the hyperparameters and reward-related parameters; reset the environment and obtain the initial state; Then, training is performed, specifically including: Loop each training round: reset the environment, initialize the UAV and unmanned boat positions, set the round reward sum and step count to zero, and obtain the initial sea ice distribution and navigable area information; Loop each step: select the unmanned boat movement action according to the ε-greedy strategy; the UAV performs sea ice identification and map construction tasks; execute the unmanned boat movement action to obtain the next state and whether to end; calculate the reward; store the experience in the experience replay buffer; If the experience replay buffer samples are sufficient: extract samples; calculate the target action value function Q value and the current action value function Q value; calculate the loss and update the main network; If the update frequency is met, update the target network parameters, wherein the update frequency refers to the number of training steps reaching a specified number; Update the UAV and unmanned boat states, increment the step count, and accumulate the reward; If it is finished, record the information and exit the loop; During the entire reinforcement learning training loop, each step of the unmanned boat produces an immediate reward, which is determined by a reward function; the immediate reward is accumulated in each training round to form a cumulative reward, and the goal of the algorithm is to maximize the cumulative reward by continuously adjusting the strategy, so that the unmanned boat learns to complete the task efficiently and safely; Reward function close to target The design is as follows: wherein is the distance between the current unmanned surface vehicle and the target, is a scaling factor; The method also gives a reward according to the change in the heading of the unmanned boat, and if the heading change is small, a positive reward is given; otherwise, a negative reward is given, and the expression of the reward is as follows: wherein is an angle of change in heading of the unmanned vehicle, is a threshold of change in heading, is a maximum angle of change in heading.

2. A method for the unmanned ship-boat collaborative working system as claimed in claim 1, characterized in that, The method comprises: Step 1: system initialization, complete parameter setting and function initialization of the UAV and the unmanned boat; Step 2: read the initial position information according to the set initial position information; Step 3: the UAV reads the target position information set by the ground station; Step 4: the UAV reaches the target position through a motion control algorithm; Step 5: determine whether the target position is the set target position, if the UAV reaches the set target position, go to step 6, if not, go to step 3; Step 6, the unmanned aerial vehicle reaches the set target position, and the unmanned aerial vehicle hovers at a set height; Step 7, the unmanned aerial vehicle carries a depth vision camera and rotates at a set speed, and the camera collects images of objects in the current area; Step 8, a digital map is constructed through the depth vision camera; Step 9, sea ice in the images is identified through a deep learning algorithm; Step 10, the navigable area is marked on the constructed digital map, and after completion, step 11 is entered; Step 11, the map information is transmitted from the unmanned aerial vehicle end to the ground station through wireless communication technology, and after completion, step 18 is entered; Step 12, it is judged whether the unmanned aerial vehicle data is transmitted, if completed, step 13 is entered, if not completed, step 11 is entered; Step 13, the unmanned aerial vehicle is on standby; Step 14, the unmanned boat moving end position is set through the ground station; Step 15, the unmanned boat moving path is set through a path planning algorithm; Step 16, the ground station control instruction is sent to the unmanned boat through wireless communication technology; Step 17, the unmanned boat executes the control instruction and moves to the specified position; Step 18, it is judged whether the unmanned boat reaches the specified position, if reached, step 19 is entered, if not up to standard, step 16 is entered; Step 19, the system enters a standby state, waits for a new task instruction, and after completion, the process ends.

3. The method of claim 2, wherein, The initial position information in the above step 2 includes the set flight height and the longitude and latitude of the specified position.

4. The method of claim 2, wherein, The above step 6 includes: The unmanned aerial vehicle reads the target point position set by the ground station; the set position is reached through a motion control algorithm, and a hovering at a set height is performed.

5. The method of claim 2, wherein, The above step 9 includes: A YOLO target detection model and its weight parameters trained on a sea ice data set in advance are loaded; An original sea ice image to be recognized is obtained by acquiring an image or a video frame to be detected by an acquisition device, and the original sea ice image is input into a recognition system; Through size adjustment, pixel normalization and data format conversion preprocessing operations on the input sea ice image, the image meets the input requirements of the YOLO target detection model, and the accuracy and efficiency of subsequent recognition are improved; By inputting the preprocessed sea ice image into the YOLO target detection model, a forward inference operation is performed to obtain original detection results containing sea ice target candidate boxes, class probabilities and confidence scores; Through post-processing of the target candidate boxes output by the YOLO target detection model, including filtering valid sea ice targets according to a confidence threshold and removing overlapping redundant boxes using a non-maximum suppression algorithm, accurate sea ice detection results are finally obtained; By drawing the detected sea ice target bounding box, class label and confidence score on the original sea ice image, the visualization of the sea ice detection result is realized, and the result image is displayed on the user interface or stored in a specified position according to the demand, realizing the automatic recognition of the input sea ice image.

6. The method of claim 5, wherein, The YOLO target detection model is trained based on a sea ice data set, and the training process is as follows: Step 1. The original image data containing sea ice distribution is collected by the camera device set in the target area, ensuring that the collection process covers various scenes with different climates, lighting and sea ice morphological changes; Step 2. The collected original image data is cleaned; Step3. Data augmentation processing is performed on the screened data set; Step4. Accurate labeling of sea ice targets in the data set is performed using professional labeling software to generate a labeled file containing target categories and bounding box coordinates; then the labeled data set is divided into a training set, a test set, and a validation set according to a certain proportion; Step5. A YOLO-based target detection algorithm framework is built, the training set is used to train the model, the network parameters are continuously optimized, and the optimal training model suitable for sea ice identification is obtained; Step6. The performance of the trained model is evaluated using the validation set and the test set, and the evaluation parameters include detection accuracy, recall rate, and inference speed; Step7. The trained and evaluated YOLO target detection model is deployed to an actual application system for automatic identification and detection of newly collected sea ice images.

7. The method of claim 6, wherein, In Step2, the data cleaning of the collected original image data includes removing blurred, excessively noisy, repeated, or invalid image data.

8. The method of claim 6, wherein, In Step3, the data augmentation processing of the screened data set includes rotation, scaling, cropping, or color transformation.

Images