A mine countermeasures vehicle

Abstract

The present application belongs to the field of aircraft, and particularly relates to a mine countermeasure aircraft. The mine countermeasure aircraft provided by the present application comprises an aircraft body, and is characterized in that the mine countermeasure aircraft further comprises a mine target identification system, a mine countermeasure system, and a path planning system. The mine countermeasure aircraft provided by the present application can autonomously identify and stably track anchor mines and sunken mine targets, cut anchor chains or throw mine-clearing bombs according to the identification, set the detonation time of a communication sonar while clearing multiple sunken mines, and implement aircraft cluster control and task planning based on a multi-target point path planning algorithm based on estimated path cost.

Description

Technical Field

[0001] This invention belongs to the field of aircraft, specifically relating to a mine countermeasures aircraft. Background Technology

[0002] An unmanned underwater vehicle (UUV) is an autonomous underwater platform that relies on its own power source to navigate and can perform a variety of tasks depending on the payload it carries. UUVs are widely used in military and civilian fields and can perform a variety of tasks such as ocean exploration, underwater facility inspection, and mine countermeasures. The development of UUVs has become a hot topic in marine technology research in various countries.

[0003] Mines, low-cost, stealthy, and highly lethal weapons, have always played a vital role in naval warfare. With the development of modern naval warfare, the types and deployment methods of mines have become increasingly complex, undoubtedly increasing the difficulty and challenge of mine clearance. Against this backdrop, the development of underwater unmanned vehicles (UUVs) with specialized mine countermeasures capabilities has become particularly important. These UUVs can autonomously search for, identify, and clear mines underwater, significantly reducing the difficulty and danger of mine clearance and ensuring the unimpeded and safe passage of sea lanes. Therefore, the development of mine countermeasures UUVs is not only key to solving a major challenge in modern naval warfare, but also makes a significant contribution to protecting naval forces, ensuring the safety of sea lanes, and even to the overall protection of the marine environment.

[0004] The research on mine-clearing UUVs is of great significance. These UUVs can significantly improve the efficiency of mine clearance operations and ensure the safety of personnel and equipment at sea. Simultaneously, they can operate in complex and ever-changing marine environments, greatly expanding the range of underwater operations. Furthermore, mine-clearing UUVs provide new tactical options for modern naval operations, contributing to changes in the maritime strategic landscape. The development and application of mine-clearing UUVs reflect modern technological advancements, not only driving progress in military technology but also having a profound impact on other fields such as marine exploration and seabed resource development. Summary of the Invention

[0005] The present invention provides a mine countermeasures vehicle, comprising a vehicle body, characterized in that it further comprises a mine target identification system for identifying mines; a mine countermeasures system for countermeasures; and a path planning system for planning a navigation path.

[0006] As a further optimization of the present invention, the mine target identification system includes a forward-looking module, which is mounted on the vehicle body. The mine target identification system identifies mine targets by the following steps:

[0007] A1: Underwater image enhancement based on GAN network;

[0008] A2: The mines in image A1 were detected using YOLO recognition.

[0009] As a further optimization of the present invention, the mine clearance system includes a mechanical claw module mounted on the vehicle body. When the anchor chain identification frame based on YOLO recognition overlaps with the cutting operation area of ​​the mechanical claw module, and the fitted anchor mine identification frame and the image ratio satisfy the formula [formula missing], the system is optimized accordingly.

[0010]

[0011] At that time, the mechanical claw module performs the shearing operation on the anchor chain, where It is the distance from the forward-looking module to the anchor mine, where It is the width of the anchor mine identification frame.

[0012] As a further optimization of the present invention, the mine target identification system includes a downward-looking module; the mine countermeasures system includes a delivery module; when the effective throwing area of ​​the mine-extinguishing bomb detected by the downward-looking module coincides with the identification area of ​​the bottom mine, the delivery module delivers the mine-extinguishing bomb.

[0013] As a further optimization of the present invention, the vehicle body is also provided with a sonar communication module; the sonar communication module includes a sonar and an underwater acoustic transducer, the underwater acoustic transducer receives the echo of the sonar and converts the acoustic signal into an electrical signal, and sets the explosion time through pulse width modulation.

[0014] As a further optimization of the present invention, when the pulse width modulation duty cycle exceeds 60%, it enters the detonation start state; when the pulse width modulation duty cycle is 70%, it detonates after 8 seconds; when the pulse width modulation duty cycle is 80%, it detonates after 20 seconds; when the pulse width modulation duty cycle is 90%, it detonates after 30 seconds; and when the pulse width modulation duty cycle is 100%, it detonates after 90 seconds.

[0015] As a further optimization of the present invention, the path planning system plans a path by including the following steps:

[0016] C1: Obtain sea area data from a known sea area database and set the starting point and target point;

[0017] C2: A three-dimensional environment model is created using the grid method;

[0018] C3: Perform global path planning;

[0019] C4: Detect unknown path information and combine deep reinforcement learning for local path planning;

[0020] C5: Global path planning with comprehensive path cost estimation and local path planning with deep reinforcement learning yield 3D path planning results for a specified sea area.

[0021] As a further optimization of the present invention, global path planning for estimating path cost includes the following steps:

[0022] D1: Calculate the Euclidean distance between any two points as the path cost and then construct the Euclidean distance cost matrix;

[0023] D2: After training a neural network to estimate the path cost between any two points, construct the estimated path cost matrix;

[0024] D3: Construct the actual path cost matrix after actually planning the path between the two points using the Informed-RRT algorithm;

[0025] D4: Using a genetic algorithm, the matrices constructed by D1, D2, and D3 are substituted into the TSP solution to obtain the traversal order, and finally the global path planning is obtained.

[0026] As a further optimization of the present invention, it also includes a cluster control system; the cluster control system distributes mine countermeasures tasks and plans paths for several mine countermeasures vehicles.

[0027] As a further optimization of the present invention, it also includes a buoy unit and a terminal. The buoy unit is connected to the vehicle via an electric power line, and the buoy unit and the vehicle communicate with each other via electric power line carrier waves. The buoy unit and the terminal communicate wirelessly.

[0028] This invention provides a mine countermeasures vehicle capable of autonomously identifying and stably tracking anchored mines and bottom mines; performing operations such as cutting anchor chains or dropping mine-clearing bombs based on identification; simultaneously clearing multiple bottom mines by setting detonation times using communication sonar; employing a multi-target point path planning algorithm based on estimated path costs; and realizing vehicle cluster control and mission planning. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the structure of a mine countermeasures vehicle;

[0030] Figure 2 This is a map showing the effective shearing area for the aircraft to handle moored mines;

[0031] Figure 3 It is a graph showing the correspondence between image scale and distance;

[0032] Figure 4 This is a map showing the effective throwing area for mine-clearing bombs to destroy bottom mines;

[0033] Figure 5 It is a feature map environment model and global path planning result diagram;

[0034] Figure 6 It is the algorithm network structure diagram;

[0035] Figure 7 This is a diagram of the framework for setting up a gym environment. Detailed Implementation

[0036] like Figure 1 As shown, this embodiment provides a mine countermeasures vehicle, including a vehicle body with a propulsion module that propels the vehicle body to navigate in the sea. This embodiment also includes a mine target identification system for identifying mines, a mine countermeasures system, and a path planning system for planning the navigation path.

[0037] The mine target identification system includes a forward-looking module located at the front of the vehicle. This module comprises a starlight-level camera equipped with a high-definition USB lens and a wide-angle lens, enabling automatic exposure, automatic white balance, automatic gain control, and high sensitivity, making it ideal for use in low-light underwater environments. Specifically, the camera is equipped with a single-degree-of-freedom gimbal mount to adapt to complex underwater environments.

[0038] The complexity of the underwater environment presents numerous challenges to the quality of visual data, including light refraction and absorption, suspended particles in the water, and color distortion. To overcome these issues, this embodiment employs a Generative Adversarial Network (GAN) to generate underwater images. The GAN improves image generation quality through adversarial learning between the generator and discriminator. In the GAN, the generator network G attempts to generate realistic underwater images to deceive the discriminator network D, which is trained to distinguish between real underwater images and those generated by the generator. By minimizing the minimax problem between the generator and discriminator, the generator-generated image is made as close as possible to a real underwater image.

[0039]

[0040] The discriminator uses the sigmoid cross-entropy loss function as the classifier, but in practice, gradient vanishing and mode collapse problems occur. These problems cause the generator's training gradient to vanish, making it difficult to train effectively. To avoid this problem, this embodiment adopts the following network loss function:

[0041]

[0042] This embodiment collects underwater data and constructs datasets for bottom mines and moored mines. The datasets are manually divided into two categories: a set of distorted underwater images (X) and a set of undistorted underwater images (Y). A network is trained to learn mappings so that set X, after processing by the network model, resembles set Y. By employing a generative adversarial network, image noise is effectively reduced, image quality is improved, and the accuracy of bottom mine and moored mine identification is further enhanced.

[0043] This embodiment uses YOLO to identify sea mines after image enhancement via a GAN network. Specifically, YOLOv8 is used.

[0044] YOLOv8's backbone network uses a C2f structure with richer gradient flow instead of the C3 structure in YOLOv5. To reduce computational cost, the number of blocks in the largest stage of the backbone network has been reduced. Furthermore, models with different scaling factors (N / S / M / L / X) no longer share a single set of model parameters. For models with large scaling factors (M / L / X), the number of output channels in the last stage has been reduced, further decreasing the number of parameters and computational cost. Similarly, the Neck module replaces the C3 module with a C2f module, and to further reduce computational cost, the two upsampling convolutions in the top-down stage of the Neck structure have been removed. The Head part has been replaced with the currently mainstream Decoupled-Head structure, separating the classification and localization branches and alleviating the inherent conflict between classification and regression tasks. The label assignment and loss functions have been changed from Anchor-Based to Anchor-Free, employing TAL dynamic matching and introducing a loss function combining DFL and CIoU Loss for the regression branch, resulting in high consistency between classification and regression tasks.

[0045] Specifically, this embodiment combines YOLOv8 with the Deep SORT algorithm to achieve real-time tracking and identification of underwater targets. Based on YOLOv8, target detection is performed on consecutive image frames to obtain mine and bounding box information. Then, the Deep SORT algorithm is used to track the target, and Kalman filtering is used to predict and extrapolate the target's state. This enables accurate tracking of mine targets and the acquisition of their trajectory and future state prediction information.

[0046] Stable tracking and accurate identification of underwater targets are achieved by combining YOLOv8 and Deep SORT. This method improves the hit rate of mine strike missions and has high real-time performance and accuracy. Furthermore, leveraging YOLOv8's efficient target detection capabilities, simultaneous tracking and identification of multiple mine targets can be realized.

[0047] The mine clearance system in this embodiment includes a mechanical claw module, which is mounted on the vehicle body. The mechanical claw module specifically includes a waterproof servo motor and a mechanical gripper. The waterproof servo motor drives the mechanical gripper to perform a shearing operation.

[0048] Figure 2This indicates that the mechanical gripper performs the cutting operation in the gripper cutting area 2. The size of the gripper cutting area 2 is directly related to the recognition frame. When the ratio between the width of the recognition frame and the width of the image changes within a specific range, the size of the gripper cutting area 2 will also adjust accordingly. As the size of the recognition frame increases, the gripper cutting area 2 will also expand. The gripper cutting area 2 actually defines the spatial range in which the aircraft gripper performs the cutting operation.

[0049] The formula is derived by fitting the relationship between the bounding box and image ratio and the distance from the forward-looking camera to the anchor chain mine:

[0050]

[0051] in It is the distance from the forward-looking module to the anchor mine, where The width of the anchor mine identification frame is such that the anchor chain area 1 overlaps with the mechanical gripper shearing operation area 2, which satisfies the requirement. Figure 3 Under the conditions specified within the Chinese frame, the shearing operation can be performed. When these conditions are met, the mechanical grippers will cut the anchor chain, preventing the moored mine from remaining secured at the bottom. This operation eliminates the threat of the moored mine and improves water safety.

[0052] In this embodiment, the mine target identification system includes a downward-looking module. The downward-looking module is similar to the forward-looking module and will not be described in detail here. The mine clearance system also includes a delivery module, which also employs a waterproof servo motor and a rotating delivery device for accurately delivering specific mine-clearing bombs used to destroy bottom mines. The downward-looking module also uses a GAN network for underwater image enhancement and YOLO recognition to detect bottom mines. Figure 4 As shown, when the effective throwing area of ​​the mine-extinguishing bomb detected by the downward-looking module coincides with the identification area of ​​the bottom mine, the deployment module deploys the mine-extinguishing bomb.

[0053] In this embodiment, the vehicle body is also equipped with a sonar communication module, which includes a sonar and an underwater acoustic transducer. Radio waves propagate very poorly in the underwater environment, so underwater communication requires the use of sound waves. The underwater acoustic transducer can convert sound into electrical energy, thereby enabling underwater communication. Using underwater acoustic communication for setting and transmitting the detonation time enhances the equipment's underwater communication capabilities, broadens data transmission pathways, and improves operational efficiency.

[0054] This embodiment employs pulse width modulation (PWM) technology to precisely control the timing of the bomb's detonation, ensuring accurate control of the mine-suppressing bomb. The underwater acoustic transducer receives the sonar echo and converts the acoustic signal into an electrical signal, which is then used to set the detonation time via PWM. The detonation time is precisely set by changing the duty cycle of the PWM. Considering factors such as practical operation and demonstration time constraints, the detonation time and duty cycle for each stage are set as follows: In the initial stage, when the duty cycle exceeds 60%, the detonation is initiated. As the subsequent mine-suppressing process begins, the detonation time is adjusted accordingly based on the activation frequency of the mine-suppressing device. Specifically, detonation occurs 8 seconds after the duty cycle is adjusted to 70%; 20 seconds after 80%; 30 seconds after 90%; and 90 seconds after 100%.

[0055] This embodiment employs a combination of global and local path planning to plan the navigation path of the vehicle. Global path planning is applicable to known marine environments, while local path planning is used for unknown marine environments. Although known data such as depth and tides can be obtained from various databases, changing factors such as ocean currents and obstacles cannot be obtained in advance. Using only partially known marine environmental data for global path planning will result in a path that is difficult to operate; while relying entirely on local path planning increases computational complexity and cannot meet the requirements of real-time path planning. Therefore, it is necessary to make full use of known environmental information and reasonably combine global and local path planning to ensure that feasible paths can be generated quickly. Specifically, the following steps are included:

[0056] C1: Obtain sea area data from a known sea area database and set the starting point and target point;

[0057] C2: A three-dimensional environment model is created using the grid method;

[0058] C3: Perform global path planning;

[0059] C4: Detect unknown path information and combine deep reinforcement learning for local path planning;

[0060] C5: Global path planning with comprehensive path cost estimation and local path planning with deep reinforcement learning yield 3D path planning results for a specified sea area.

[0061] This embodiment utilizes the path planning characteristics of the Informed-RRT algorithm, namely, the path approaches optimality and is relatively stable, and the relevant parameter characteristics are easy to summarize. By training a neural network for path cost estimation, the path cost calculated by the Informed-RRT algorithm in obstacle-filled environments is estimated. This algorithm can achieve the following objectives:

[0062] It avoids obstacles while quickly planning the optimal closed path, traversing preset target points and finally returning to the starting point.

[0063] While ensuring the speed and efficiency of the path, we must also ensure that the total cost of the planned path is minimized.

[0064] Specifically, it includes the following steps:

[0065] (1) Constructing the estimated path cost matrix: Calculate the representation parameters of the network input. Taking the case where a line segment connecting two points intersects with a single obstacle as an example, the parameters include the line segment length l, the height h11 and h12 of the arc formed by the intersection of the circle and the line segment, and the normalized distance d' from the perpendicular point of the arc to the midpoint of the line segment. Select the corresponding network according to the representation parameters and calculate the estimated path cost. Calculate the cost estimates for all paths and construct the estimated path cost matrix Des as an n+1 order square matrix, where dij (i, j=0,1,...,n) represents the estimated path cost between the i-th point and the j-th point.

[0066]

[0067] (2) Construct the actual path cost matrix. Call the Informed-RRT* algorithm to plan the path between any two points, and construct the actual path cost matrix Dre based on the planning results.

[0068] (3) Construct the Euclidean distance cost matrix to calculate the Euclidean distance between any two points and construct the Euclidean distance cost matrix Ded. Where (x1,y1) and (x2,y2) are the path costs, and (x1,y1) and (x2,y2) are the coordinates of the starting point and the ending point of a certain planning, respectively.

[0069] (4) Solve the traversal order. Input three cost matrices respectively, and use the genetic algorithm to solve the TSP problem to obtain the traversal order that minimizes the total path cost of each.

[0070] like Figure 5 As shown, this embodiment transforms the mine countermeasures unmanned aerial vehicle (UAV) investigation and processing task into a multi-target point path planning problem, utilizing reconnaissance-acquired information on the location of obstacles and suspected mines in the mission area. A two-dimensional feature map is created to represent the mission area environment, including the UAV's starting point, target points, and the position and size information of the obstacle circles obtained by expanding the circumcircle of the obstacles. The expansion radius Rp is equal to the width of the UAV. Three cost matrices are constructed: the actual cost matrix Dre, the estimated cost matrix Des, and the Euclidean distance matrix Ded. These are substituted into the TSP problem for solution. Substituting Dre and Des yields the same traversal order (order 1), while substituting Ded yields order 2. The Informed-RRT* algorithm is then used to plan the globally optimal path starting from the starting point, passing through all target points in sequence, and returning to the starting point.

[0071] The actual marine environment is complex, and factors affecting vehicle navigation are difficult to detect in advance, rendering global path planning results ineffective. Therefore, it is necessary to use the vehicle's own sensors to detect the environment in real time and perform local path planning based on the detected information. This embodiment employs the Proximal Policy Optimization (PPO) algorithm as a key algorithm in the Actor-Critic framework. It borrows the concept of policy-based estimation and optimizes the step size selection mechanism, making it an effective tool for handling local path planning problems involving high-dimensional state spaces and continuous action spaces.

[0072] Specifically, the PPO algorithm using the clip truncation method has the following network structure: Figure 6 As shown. The network architectures of Actor and Critic are the same, both consisting of three fully connected layers with 128, 128, and 64 hidden neurons respectively, and using tanh as the activation function. The objective function is:

[0073]

[0074] In the PPO algorithm, a state transition sequence needs to be generated under the Actor network, and a small batch of samples is extracted from it for training. At fixed time intervals of n_steps, the corresponding objective function is calculated, and the parameters are updated using policy gradient ascent and mean squared error gradient descent.

[0075] During deep reinforcement learning training, such as Figure 7 As shown, if the completion flag of the gym environment is true, the current round ends, and the round reward and single-step reward are added together to obtain the total reward. After recording the total reward, the reset function is called to reset the gym environment and start a new round of training.

[0076] Preferably, it also includes a cluster control system; the cluster control system distributes mine-clearing tasks and plans paths for several mine-clearing vehicles. In the execution of minefield clearance tasks by a cluster of multiple autonomous underwater mine-clearing vehicles, the focus is on achieving effective and accurate task allocation and path planning, enabling each vehicle to find a suitable navigation path in the marine environment, reach the mine target, and carry out safe and effective disposal. In this process, a peer-to-peer network macro-agent architecture and vehicle cluster obstacle avoidance split / reunification maneuvering technology are used. Dynamic changes in the agent network topology provide effective maneuvering strategies for vehicle movement in obstacle-filled environments. The combination of cluster control and task planning technologies not only enables rapid and effective handling of large-scale minefields but also maintains high maneuverability and stability in the face of complex and changing environments.

[0077] Preferably, the system also includes a buoy unit and a terminal. The buoy unit is connected to the vessel via a power line, and the buoy unit and the vessel communicate with each other via power line carrier waves. The buoy unit also communicates wirelessly with the terminal. Since the vessel needs to operate over long distances and cannot directly connect to a shore-based monitoring computer via a network cable, a power line carrier module is used in the underwater portion of the vessel. This module modulates the output network signal and then transmits it via power waves to the power line carrier module in the buoy unit. After demodulation, the signal is re-output as a network signal, achieving lossless and transparent transmission from underwater to the surface. Simultaneously, the power line carrier line also powers the various modules within the buoy unit.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A mine countermeasures vehicle, comprising a vehicle body, characterized in that, It also includes a mine target identification system to identify mines; a mine countermeasures system to counter mines; and a path planning system to plan navigation paths. The mine target identification system includes a forward-looking module, which is mounted on the vehicle body. The mine target identification system identifies mine targets through the following steps: A1: Underwater image enhancement based on GAN network; A2: Mines were detected in image A1 based on YOLO recognition; The mine clearance system includes a mechanical claw module mounted on the vehicle body. When the anchor chain identification frame based on YOLO overlaps with the cutting operation area of ​​the mechanical claw module, and the fitted anchor mine identification frame and the image ratio satisfy the formula [formula missing], the system will be able to clear mines. At that time, the mechanical claw module performs the shearing operation on the anchor chain, where It is the distance from the forward-looking module to the anchor mine, where It is the width of the anchor mine identification frame.

2. The mine countermeasures vehicle according to claim 1, characterized in that, The mine target identification system includes a downward-looking module; the mine countermeasures system includes a delivery module; when the effective throwing area of ​​the mine-suppressing bomb detected by the downward-looking module coincides with the identification area of ​​the bottom mine, the delivery module delivers the mine-suppressing bomb.

3. The mine countermeasures vehicle according to claim 2, characterized in that, The vehicle body is also equipped with a sonar communication module; the sonar communication module includes a sonar and an underwater acoustic transducer. The underwater acoustic transducer receives the echo of the sonar and converts the acoustic signal into an electrical signal, and sets the explosion time through pulse width modulation.

4. The mine countermeasures vehicle according to claim 3, characterized in that, When the pulse width modulation duty cycle exceeds 60%, it enters the detonation start state; when the pulse width modulation duty cycle is 70%, it detonates after 8 seconds; when the pulse width modulation duty cycle is 80%, it detonates after 20 seconds; when the pulse width modulation duty cycle is 90%, it detonates after 30 seconds; when the pulse width modulation duty cycle is 100%, it detonates after 90 seconds.

5. A mine countermeasures vehicle according to claim 1, characterized in that, The path planning system plans a path by including the following steps: C1: Obtain sea area data from a known sea area database and set the starting point and target point; C2: A three-dimensional environment model is created using the grid method; C3: Perform global path planning; C4: Detect unknown path information and combine deep reinforcement learning for local path planning; C5: Global path planning with comprehensive path cost estimation and local path planning with deep reinforcement learning yield 3D path planning results for a specified sea area.

6. A mine countermeasures vehicle according to claim 5, characterized in that, Global path planning for estimating path costs includes the following steps: D1: Calculate the Euclidean distance between any two points as the path cost and then construct the Euclidean distance cost matrix; D2: After training a neural network to estimate the path cost between any two points, construct the estimated path cost matrix; D3: Construct the actual path cost matrix after actually planning the path between the two points using the Informed-RRT algorithm; D4: Using a genetic algorithm, the matrices constructed by D1, D2, and D3 are substituted into the TSP to obtain the traversal order, and finally the global path planning is obtained.

7. A mine countermeasures vehicle according to claim 1, characterized in that, It also includes a cluster control system; the cluster control system distributes mine countermeasures tasks and plans paths for several mine countermeasures vehicles.

8. A mine countermeasures vehicle according to claim 1, characterized in that, It also includes a buoy unit and a terminal. The buoy unit is connected to the aircraft via power lines, and the buoy unit and the aircraft communicate with each other via power line carrier waves. The buoy unit and the terminal communicate wirelessly.

Images