Multi-uav target coverage path planning method based on cluster-pooling graph neural network

Abstract

This application discloses a multi-UAV target coverage path planning method based on clustered pooling graph neural networks, belonging to the field of path planning. The method includes: obtaining the explored sub-maps of each UAV at the current moment based on their current position information and explored area information; if the total coverage area of ​​the explored sub-maps corresponding to all UAVs is less than the target area, inputting the explored sub-maps corresponding to at least one UAV into a graph neural network model to obtain the target motion path for at least one UAV; controlling at least one UAV to move based on the target motion path and obtaining its position information and explored area information after movement to update the explored sub-maps of at least one UAV; and controlling each UAV to stop moving if the total coverage area of ​​the updated explored sub-maps corresponding to all UAVs overlaps with the target area. This application achieves high efficiency in completing the coverage task.

Description

Technical Field

[0001] This application belongs to the field of path planning technology, and in particular relates to a multi-UAV target coverage path planning method based on clustered pooling graph neural networks. Background Technology

[0002] In industrial parks, VOCs (volatile organic compounds) inspections require checking for gas leaks at pipeline pollution sources and collecting gas samples from suspected leaks for chemical analysis in the laboratory. With the continuous improvement of drone technology, using multiple drones for VOCs inspection has become an emerging technological approach. While some methods rely on manually controlled drones for inspection, these methods are time-consuming and labor-intensive, and prone to missed or duplicate checks, resulting in low inspection efficiency. Summary of the Invention

[0003] This application aims to address at least one of the technical problems existing in the related art. To this end, this application proposes a multi-UAV target coverage path planning method based on clustered pooling graph neural networks. This method can automatically collect environmental information for path planning, without consuming a large amount of manpower, and promotes coordination among multiple UAVs, reducing missed or duplicate inspections and improving the efficiency of coverage task completion.

[0004] Firstly, this application provides a multi-UAV target coverage path planning method based on clustered pooling graph neural networks, the method comprising:

[0005] Based on the current location information of each UAV and the information of the explored areas of each UAV in the target area, obtain the explored sub-map of each UAV at the current time;

[0006] If the total coverage area of ​​the explored sub-maps corresponding to all the UAVs is less than the target area, at least one of the explored sub-maps corresponding to the UAVs is input into the graph neural network model to obtain at least one target motion path corresponding to the UAVs output by the graph neural network model.

[0007] Control at least one of the UAVs to move based on the target movement path, and acquire the position information and explored area information of at least one of the UAVs after movement, so as to update the explored sub-map of at least one of the UAVs;

[0008] If the total coverage area of ​​the updated explored sub-map corresponding to all the drones overlaps with the target area, control each drone to stop moving.

[0009] The multi-UAV target coverage path planning method based on clustered pooling graph neural network provided in this application performs local path planning for UAVs by using environmental information such as the explored area information and the location information of the UAVs. It also generates new local path plans based on the environmental information after the UAVs move, iterating until the inspection task is finally completed. The method effectively utilizes environmental information during the inspection process, can automatically collect environmental information for path planning, does not require a lot of manpower, promotes coordination between multiple UAVs, reduces missed or repeated inspections, and improves the efficiency of coverage task completion.

[0010] One embodiment of this application provides a multi-UAV target coverage path planning method based on a clustered pooling graph neural network, wherein the graph neural network model is trained based on the following steps:

[0011] Obtain the explored sub-map of multiple samples corresponding to a single data collection cycle using a sample drone;

[0012] Based on map information from the complete raster map, determine the sample movement path corresponding to the explored sub-map for each sample; at least some of the sample movement paths constitute a coherent movement path.

[0013] The graph neural network model is trained using the explored subgraphs of the multiple samples as samples and the sample motion paths corresponding to the explored subgraphs of each sample as sample labels.

[0014] One embodiment of this application provides a multi-UAV target coverage path planning method based on clustered pooling graph neural networks. The method includes a single acquisition cycle comprising multiple sample first moments. The step of acquiring the explored sub-graphs corresponding to the sample UAVs within a single acquisition cycle includes:

[0015] Based on each grid point in the complete grid map, obtain a set of incomplete obstacle grids from the complete grid map;

[0016] Based on the sample location information of the sample UAV in the complete grid map at the first moment of the sample and the distance between each of the incomplete obstacle grid points in the incomplete obstacle grid set, the incomplete obstacle grid set is processed to obtain the sample explored sub-map corresponding to the sample UAV at the first moment of the sample.

[0017] One embodiment of this application provides a multi-UAV target coverage path planning method based on clustered pooling graph neural networks. The method determines the sample movement path corresponding to each explored sub-map based on map information from a complete grid map, including:

[0018] Based on the area ratio of the region of interest corresponding to each grid point in the complete grid map, the set of interest grids is obtained.

[0019] Based on the interest grid set, the incomplete obstacle grid set, and the perception range parameters of the sample UAV, a first grid set is obtained;

[0020] Based on the viewpoint grid set and the interest grid set corresponding to the first grid set, a global keypoint grid set is constructed.

[0021] Based on the global key point grid set corresponding to the node set in the explored subgraph of the sample, path planning is performed on the sample UAV to obtain the sample motion path corresponding to the explored subgraph of the sample.

[0022] An embodiment of this application provides a multi-UAV target coverage path planning method based on a clustered pooling graph neural network, wherein training the graph neural network model using the plurality of explored sub-maps as samples and the motion paths corresponding to each explored sub-map as sample labels includes:

[0023] The multiple explored sub-graphs of the samples are input into the graph neural network model to obtain the predicted motion path corresponding to each explored sub-graph of the samples output by the graph neural network model;

[0024] Based on the predicted motion path and the motion path of the sample corresponding to the explored sub-graph of the sample, a target loss function is constructed.

[0025] The graph neural network model is trained based on the target loss function.

[0026] One embodiment of this application provides a multi-UAV target coverage path planning method based on a clustered pooling graph neural network. The graph neural network model includes a dead zone clustered pooling module, a graph aggregation module, and a policy module connected in sequence. The step of inputting the explored subgraph corresponding to at least one of the UAVs into the graph neural network model to obtain the target motion path corresponding to at least one UAV output by the graph neural network model includes:

[0027] Based on the dead zone cluster pooling module, cluster pooling operation is performed on the explored subgraph to obtain the first subgraph corresponding to the explored subgraph output by the dead zone cluster pooling module;

[0028] Based on the graph aggregation module, the first subgraph is processed to obtain multiple second subgraphs output by the graph aggregation module;

[0029] The multiple second sub-graphs are cascaded based on the strategy module to obtain the target motion path output by the strategy module.

[0030] One embodiment of this application provides a multi-UAV target coverage path planning method based on a clustered pooling graph neural network. The method involves performing clustering operations on the explored subgraph using the dead-zone clustering module to obtain a first subgraph corresponding to the explored subgraph output by the dead-zone clustering module. The method includes:

[0031] If, within the receptive field corresponding to the target drone, there are no other drones, regions of interest, or boundary regions among the drones mentioned above, the target cells are obtained based on the number of adjacent targets in the target region; the boundary region is a grid of unexplored regions adjacent to explored regions.

[0032] Based on the interconnected nodes within the target unit, obtain at least one cluster node corresponding to the target unit;

[0033] Process the connected cluster nodes in adjacent target units to obtain the edge features between the connected cluster nodes;

[0034] The first subgraph is obtained based on the at least one cluster node and each edge feature.

[0035] One embodiment of this application provides a multi-UAV target coverage path planning method based on clustered pooling graph neural networks. The graph aggregation module includes a graph encoding module, multiple sequentially connected graph aggregation operation modules, and multiple graph decoding modules. The output of the graph encoding module is connected to both the graph aggregation operation module and the graph decoding module, and the output of each graph aggregation module is connected to the graph decoding module. The step of processing the first sub-graph based on the graph aggregation module to obtain multiple second sub-graphs output by the graph aggregation module includes:

[0036] Based on the graph encoding module, a nonlinear transformation is performed on the cluster nodes and edge features corresponding to the first subgraph to obtain the third subgraph;

[0037] Multiple fourth subgraphs are obtained by performing multiple aggregation operations on the first subgraph based on multiple graph aggregation operation modules;

[0038] The third subgraph and the fourth subgraph are decoded using the multiple graph decoding modules to obtain the multiple second subgraphs.

[0039] An embodiment of this application provides a multi-UAV target coverage path planning method based on a clustered pooling graph neural network. The strategy module includes an activation function layer. The step of cascading multiple second subgraphs based on the strategy module to obtain the target motion path output by the strategy module includes:

[0040] The multiple second subgraphs are concatenated to obtain the fifth subgraph;

[0041] Based on the activation function layer, the probability values ​​corresponding to each candidate motion path in the fifth subgraph are obtained, and the candidate motion path corresponding to the largest probability value among multiple probability values ​​is determined as the target motion path.

[0042] Secondly, this application provides a multi-UAV target coverage path planning device based on clustered pooling graph neural networks, comprising:

[0043] The first processing module is used to obtain the explored sub-map of each UAV at the current time based on the location information of each UAV at the current time and the explored area information of each UAV in the target area.

[0044] The second processing module is used to input at least one of the explored sub-maps of all the UAVs into the graph neural network model when the total coverage area of ​​the explored sub-maps of all the UAVs is less than the target area, and to obtain at least one target motion path of the UAVs output by the graph neural network model.

[0045] The third processing module is used to control at least one of the UAVs to move based on the target movement path, and to obtain the position information and explored area information of at least one of the UAVs after movement, so as to update the explored sub-map of at least one of the UAVs;

[0046] The fourth processing module is used to control each of the drones to stop moving when the total coverage area of ​​the updated explored sub-map corresponding to all the drones overlaps with the target area.

[0047] According to the multi-UAV target coverage path planning device based on clustered pooling graph neural network provided in this application embodiment, the device performs local path planning for the UAV by using environmental information such as the explored area information and the location information of the UAV, and generates a new local path plan based on the environmental information after the UAV moves. The process is iterated until the inspection task is finally completed. The device effectively utilizes environmental information during the inspection process, can automatically collect environmental information for path planning, does not require a lot of manpower, promotes coordination between multiple UAVs, reduces missed or repeated inspections, and improves the efficiency of coverage task completion.

[0048] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multi-UAV target coverage path planning method based on clustered pooling graph neural networks as described in the first aspect above.

[0049] Fourthly, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-UAV target coverage path planning method based on clustered pooling graph neural networks as described in the first aspect above.

[0050] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the multi-UAV target coverage path planning method based on clustered pooling graph neural networks as described in the first aspect above.

[0051] The above-described one or more technical solutions in the embodiments of this application have at least one of the following technical effects:

[0052] By using environmental information such as the explored area information and the location information of the drone, local path planning is performed on the drone, and new local path planning is generated based on the environmental information after the drone moves. The process is iterated until the inspection task is finally completed. Environmental information is effectively utilized during the inspection process. Environmental information can be automatically collected for path planning without consuming a lot of manpower. It also promotes coordination between multiple drones, reduces missed inspections or duplicate inspections, and improves the efficiency of completing the coverage task.

[0053] Furthermore, by processing the explored subgraph based on the cluster pooling method, environmental information can be utilized more effectively when generating planned paths, promoting coordination among multiple UAVs and thus improving the efficiency of inspection tasks. By constructing a graph neural network model, UAVs can autonomously plan paths in complex environments, reducing manpower costs and improving the accuracy and precision of the generated motion paths.

[0054] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0055] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0056] Figure 1 This is one of the flowcharts illustrating the multi-UAV target coverage path planning method based on clustered pooling graph neural networks provided in this application embodiment;

[0057] Figure 2 This is a schematic diagram of the coverage path planning of a UAV based on the multi-UAV target coverage path planning method provided in the embodiments of this application;

[0058] Figure 3This is a schematic diagram of the graph neural network model of the multi-UAV target coverage path planning method based on clustered pooling graph neural network provided in the embodiments of this application;

[0059] Figure 4 This is a schematic diagram illustrating the principle of cluster pooling operation in the multi-UAV target coverage path planning method based on cluster pooling graph neural network provided in the embodiments of this application;

[0060] Figure 5 This is a schematic diagram illustrating the generation principle of the target motion path in the multi-UAV target coverage path planning method based on clustered pooling graph neural networks provided in this application embodiment;

[0061] Figure 6 This is the second flowchart illustrating the multi-UAV target coverage path planning method based on clustered pooling graph neural networks provided in this application embodiment;

[0062] Figure 7 This is a schematic diagram of the structure of the multi-UAV target coverage path planning device based on clustered pooling graph neural network provided in the embodiments of this application;

[0063] Figure 8 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0064] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0065] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0066] The following description, in conjunction with the accompanying drawings, details the multi-UAV target coverage path planning method, device, electronic device, and readable storage medium based on clustered pooling graph neural networks provided in this application, through specific embodiments and application scenarios.

[0067] Among them, the multi-UAV target coverage path planning method based on clustered pooling graph neural networks can be applied to the terminal, specifically executed by the hardware or software in the terminal.

[0068] The terminal includes, but is not limited to, portable communication devices such as mobile phones or tablets with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads). It should also be understood that, in some embodiments, the terminal may not be a portable communication device, but rather a desktop computer with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads).

[0069] The following embodiments describe a terminal including a display and a touch-sensitive surface. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, mouse, and joystick.

[0070] The multi-UAV target coverage path planning method based on clustered pooling graph neural networks provided in this application embodiment can be executed by an electronic device or a functional module or entity in an electronic device that can implement the multi-UAV target coverage path planning method based on clustered pooling graph neural networks. The electronic devices mentioned in this application embodiment include, but are not limited to, mobile phones, tablets, computers, cameras, and wearable devices. The following uses an electronic device as the execution subject to illustrate the multi-UAV target coverage path planning method based on clustered pooling graph neural networks provided in this application embodiment.

[0071] like Figure 1 As shown, the multi-UAV target coverage path planning method based on clustered pooling graph neural network includes steps 110, 120, 130 and 140.

[0072] Step 110: Based on the current location information of each UAV and the information of the explored areas of each UAV in the target area, obtain the explored sub-map of each UAV at the current time;

[0073] In this step, the current time is the current data collection time, and the location information of each drone at the current time can be collected.

[0074] Location information can include coordinates, etc.

[0075] The target area is the area where multiple drones need to be inspected.

[0076] like Figure 2 As shown, a grid map corresponding to the target area can be obtained. The grid map includes multiple grids. If the drone has visited a grid, that grid can be identified as the area explored by the drone.

[0077] The explored area information can include explored grid points, which can include non-obstacle grids in the explored area that are adjacent to the unexplored area.

[0078] The explored area is the area detected by the drone's perception camera, and the unexplored area is the area not detected by the drone's perception camera.

[0079] The set of grids explored by the drone is progressive over time, and the set of grids encompassed by the drone's camera perception area is constantly expanding.

[0080] It can acquire the region of interest information of the drone, where the region of interest is an area containing potential pollution sources. The region of interest information can include the ratio of the area of ​​the pollution source in the interest grid to the area of ​​the interest grid.

[0081] Based on the location information of each drone, the ratio of the number of drones in each grid to the area of ​​that grid can be obtained.

[0082] Based on the information of the explored areas, the location information of each UAV, and the information of the region of interest, the explored sub-map corresponding to each UAV can be obtained.

[0083] The explored subgraph can include graph structure data such as explored grid points and edges, with edges used to represent the connection relationships between grid points.

[0084] The explored subgraphs can also include features related to grid points and edges, such as gas concentration and obstacle location information.

[0085] Step 120: If the total coverage area of ​​the explored sub-maps corresponding to all UAVs is less than the target area, input the explored sub-maps corresponding to at least one UAV among all UAVs into the graph neural network model to obtain the target motion path corresponding to at least one UAV output by the graph neural network model.

[0086] In this step, the drones can communicate with each other and obtain the location information of other drones as well as the explored sub-maps.

[0087] The coverage areas of the explored sub-maps of each drone may overlap. We can perform deduplication on the explored sub-maps of each drone to obtain the total coverage area of ​​the explored sub-maps of all drones.

[0088] If the total coverage area is smaller than the target area, that is, if there are still areas in the target area that the drone has not explored, path planning can continue for at least one drone to control the drone to continue moving.

[0089] The input to the graph neural network model is the explored subgraph of the UAV, and the output is the target motion path corresponding to the UAV.

[0090] A graph neural network model can be trained using the explored subgraphs of a sample as samples and the motion paths of the samples corresponding to the explored subgraphs as sample labels.

[0091] The target motion path can include the step length of the UAV in four directions. For example, the candidate step lengths of the UAV in each direction can be calculated, and each candidate step length has a corresponding probability value. The candidate step length corresponding to the largest probability value among multiple probability values ​​can be determined as the target motion path of the UAV.

[0092] For example, candidate movement steps can include: moving forward 1 grid (with a probability of 10%), moving backward 1 grid (with a probability of 50%), moving left 1 grid (with a probability of 30%), and moving right 1 grid (with a probability of 10%). Then the target movement path can be determined as moving backward 1 grid.

[0093] Step 130: Control at least one UAV to move based on the target movement path, and obtain the position information and explored area information of at least one UAV after movement, so as to update the explored sub-map of at least one UAV;

[0094] In this step, after the target motion path is generated, the drone can be controlled to move based on its corresponding target motion path.

[0095] After the drone moves along the target's path, its current location and the area it has explored can be obtained.

[0096] Then, based on the updated location information and the information of the explored areas, the explored sub-map of the drone can be updated.

[0097] Step 140: When the total coverage area of ​​the updated explored sub-map corresponding to all drones coincides with the target area, control each drone to stop moving.

[0098] In this step, if the total coverage area of ​​the explored sub-maps corresponding to all drones coincides with the target area, that is, if there are no unexplored areas in the target area, the drones can be controlled to stop moving.

[0099] like Figure 2 As shown, Figure 2An example of an interest grid is provided. During its movement, the drone can inspect pipeline pollution sources for gas leaks and extract polluting gases, such as volatile organic compounds (VOCs), from the interest grid.

[0100] In this application, as Figure 5 As shown, for a drone, the explored sub-map of the drone at the current acquisition time can be obtained. If it is determined that there are unexplored areas in the target area, a target motion path with a step length of t can be generated based on the drone's explored sub-map. Then, the drone is controlled to move based on the target motion path, and the explored sub-map of the drone after the movement is updated. If there are still unexplored areas in the target area, a target motion path with a step length of t+1 can be generated based on the updated explored sub-map of the drone. Then, the drone is controlled to move based on the target motion path until the drone has visited all regions of interest and the drone's perception camera has detected all non-obstacle grids. It can be determined that the drone has completed the inspection task, and the drone is controlled to stop moving.

[0101] The multi-UAV target coverage path planning method based on clustered pooling graph neural network provided in this application performs local path planning for UAVs by using environmental information such as the explored area information and the location information of the UAVs. It also generates new local path plans based on the environmental information after the UAVs move, iterating until the inspection task is finally completed. The method effectively utilizes environmental information during the inspection process, can automatically collect environmental information for path planning, does not require a lot of manpower, promotes coordination between multiple UAVs, reduces missed or repeated inspections, and improves the efficiency of coverage task completion.

[0102] In some embodiments, the graph neural network model can be trained based on the following steps:

[0103] Obtain the explored sub-map of multiple samples corresponding to a single data collection cycle using a sample drone;

[0104] Based on the map information of the complete raster map, determine the sample movement path corresponding to the explored sub-map of each sample;

[0105] A graph neural network model is trained using multiple explored subgraphs as samples and the motion paths of the samples corresponding to each explored subgraph as sample labels.

[0106] In this embodiment, for a sample drone, multiple sample explored sub-maps corresponding to the sample drone can be collected in one collection cycle.

[0107] For example, a data acquisition cycle can include multiple data acquisition moments. At each data acquisition moment, the sample exploration sub-map corresponding to the sample drone can be acquired.

[0108] Depending on the location of the sample drone, the explored sub-map corresponding to the sample drone will be different.

[0109] Based on the map information of the complete raster map, multiple sample movement paths can be constructed. The time steps corresponding to the multiple sample movement paths are different. The sample explored submap at the target time t corresponds to the sample movement path at the time step t to t+1 corresponding to the target time.

[0110] For example, if the sampled drone collected at time t0 has an explored sub-map of x1, the corresponding sampled motion path y1 with a step length of t0 to t1 can be determined from multiple sampled motion paths; if the sampled drone collected at time t1 has an explored sub-map of x2, the corresponding sampled motion path y2 with a step length of t1 to t2 can be determined from multiple sampled motion paths.

[0111] At least some of the sample motion paths can form a coherent motion path. For example, the end point of sample motion path y1 is the starting point of sample motion path y2.

[0112] like Figure 6 As shown, in some embodiments, obtaining multiple sample-explored sub-maps corresponding to a sample drone in one collection cycle may include:

[0113] Based on each grid point in the complete grid map, obtain a set of incomplete obstacle grids from the complete grid map;

[0114] Based on the sample location information of the sample UAV in the complete grid map at the first moment and the distance between each incomplete obstacle grid point in the incomplete obstacle grid set, the incomplete obstacle grid set is processed to obtain the sample explored sub-map corresponding to the sample UAV at the first moment.

[0115] In this embodiment, a simulation environment can be built to create a complete grid map on the inspection flight altitude plane, such as... Figure 2 As shown.

[0116] An incomplete obstacle grid set can include multiple incomplete obstacle grid points.

[0117] Incomplete obstacles can include partially obstructed obstacles or obstacles that drones can bypass.

[0118] A collection period is the duration from the start of collection to the end of collection. A collection period can include multiple collection times, that is, a collection period includes multiple sample first times. For example, a collection period can be set from collection time t to collection time t+n (where n is a natural number greater than 0). A collection period can include n+1 sample first times.

[0119] The distance between each incomplete obstacle grid point can be obtained. Then, based on the sample position information of the sample UAV at the first moment of the sample and the distance between each incomplete obstacle grid point, the sample explored subgraph can be constructed. That is, the node features and edge features corresponding to the sample explored subgraph can be constructed to build the sample explored subgraph.

[0120] In actual execution, a simulation environment can be built to determine the obstacle space and flyable space in the map, and then a complete grid map can be created on the inspection flight altitude plane, such as... Figure 2 As shown, all incomplete obstacle grid points are marked as an incomplete obstacle grid set C0, and nodes are constructed from the grid points in C0 respectively. Where i = 1, 2, ..., m, Number of drones within a grid / grid area This represents the percentage of unexplored areas within the grid. This represents the area percentage of the region of interest (if a drone has already visited the grid, the area of ​​the region of interest degenerates to 0). The percentage of boundary grid cells (the boundary refers to the non-obstacle grid cells in the explored area that are adjacent to the unexplored area at time t).

[0121] In C0, if there is a passage for drones between adjacent grid points i and j, edge features corresponding to grid points i and j can be constructed. Edge features Let be the path distance between the center positions of grids i and j.

[0122] Obtain the grid where the sample drone is located (i.e., the sample location information), and then construct the graph structure data of the sample covering the mission environment at the first time step t:

[0123]

[0124] Then we can consider G at time t. t Node S, explored by drones t The sample has been explored and the subgraph is represented as G. t [S t ].

[0125] In some embodiments, determining the sample movement path corresponding to the explored sub-map for each sample based on map information from the complete raster map may include:

[0126] Based on the area ratio of the region of interest corresponding to each grid point in the complete grid map, obtain the set of interest grids;

[0127] The first grid set is obtained based on the interest grid set, the incomplete obstacle grid set, and the perception range parameters of the sample UAV;

[0128] Based on the viewpoint grid set and interest grid set corresponding to the first grid set, a global keypoint grid set is constructed.

[0129] Based on the set of nodes in the explored subgraph of the sample corresponding to the global key point grid set, path planning is performed on the sample UAV to obtain the sample motion path corresponding to the explored subgraph.

[0130] In this embodiment, the area ratio of the region of interest is used to characterize the access status of the grid point.

[0131] The sensing range parameters of the sample drone can be determined based on the configuration of the sample drone. For example, the sensing range parameters can be 9 grids, 16 grids, or 25 grids, etc., which are not limited in this application.

[0132] like Figure 2 As shown, the sensing range of the sample drone's camera can cover grid 1-grid 25. If the drone's sensing camera has detected grid 1-grid 25, it can be determined that grid 1-grid 25 has been explored by the drone.

[0133] During the detection process, the drone can reach the central grid (e.g., grid 13) among multiple grids (grid 1-grid 25), and grid 13 can be identified as a grid that has been visited by the drone.

[0134] For grid points that have been visited by the sample drone, the area percentage of the region of interest corresponding to that grid point is 0.

[0135] The set of interest grids includes at least one grid point, wherein the grid point is a grid containing a potential source of pollution that has not been visited by drones.

[0136] The coverage status of a grid point includes whether the sample drone has visited the grid point or not. Based on the sensing range parameters of the sample drone and the set of interest grids, the set of grids that the sensing range of the sample drone can completely cover can be obtained. Then, based on this set of grids, a first set of grids is obtained from the set of incomplete obstacle grids.

[0137] Based on the perception range parameter, a corresponding viewpoint grid set can be generated at equal intervals within the region corresponding to the first grid set. Then, based on the sum of the viewpoint grid set and the interest grid set, a global keypoint grid set can be constructed.

[0138] Based on the global keypoint grid set corresponding to the node set in the explored sub-graph of the sample, path planning can be performed on the sample UAV to obtain the motion path corresponding to the explored sub-graph of the sample.

[0139] In actual implementation, such as Figure 6 As shown, grid points with an area ratio greater than 0 in the region of interest can be marked as the set of interest grids C1. Assuming that there are |C1| drones located at the center of each grid point in C1, the grids that can be completely covered by the sensing range are set C2. The first grid set C3 = C0 - C2 can be obtained.

[0140] Based on the perception range parameter, a viewpoint grid set C can be generated at equal intervals in the C3 region. A This makes the distribution in C A The perception range of multiple sample drones at the center of each grid point can completely cover the C3 area.

[0141] Construct a global keypoint grid set C = C1 + C4. Set C corresponds to the node set V in the explored subgraph G0 of the sample. Path planning algorithms (such as A) can be used in G0. * The algorithm calculates the shortest path between any two elements in V and uses the path length as the edge length connecting the nodes in V to construct the key point graph. Then you can see the diagram. The problem of grouping n sample drones through |V| key points is modeled as an operations research (VRP) problem. For example, the OR-tools tool can be used to calculate the paths of the n drones and map the paths to graph G0 to obtain expert solution sample labels. The samples in the expert solution dataset are at each time step t. (See Figure G below.) t Subgraph G has been explored t [S t ], where S t This represents the set of nodes explored at time t, with the sample labels being the path selection actions (i.e., sample motion paths) of each UAV in four directions. in, like Figure 5 middle As shown.

[0142] According to the multi-UAV target coverage path planning method based on clustered pooling graph neural network provided in the embodiments of this application, the path planning of sample UAVs is performed based on global grid map information, and sample labels are generated for training graph neural network models. This enables the graph neural network model trained based on the explored subgraphs and motion paths of the samples to generate relatively accurate motion paths based on local explored subgraph information in practical applications, with high accuracy and precision in path planning.

[0143] like Figure 6 As shown, in some embodiments, using multiple explored sub-graphs as samples and the sample motion paths corresponding to each explored sub-graph as sample labels, training a graph neural network model may include:

[0144] Multiple sample explored subgraphs are input into a graph neural network model to obtain the predicted motion path corresponding to each sample explored subgraph output by the graph neural network model;

[0145] Based on the predicted motion path and the motion path of the sample corresponding to the explored sub-graph, a target loss function is constructed.

[0146] A graph neural network model is trained based on a target loss function.

[0147] In this embodiment, the target loss function is used to measure the difference between the model's predictions and the actual results.

[0148] The target loss function can include cross-entropy loss function, sum of squares loss function, or log loss function, etc.

[0149] The weights of a graph neural network model can be updated using the backpropagation algorithm to minimize the objective loss function until the graph neural network model's performance on the validation set meets the requirements.

[0150] In actual implementation, such as Figure 6 As shown, an expert dataset B = {(G} can be constructed on a small-scale map (|C| < 100, number of drones < 10)}. t [S t A t )} t＝1:T Among them, G t [S t [G at time t] t Node S, explored by drones t The sample has been explored in subgraph A. t This refers to the set of drone actions (sample motion paths) in the expert solution.

[0151] Model training can be based on imitation learning methods:

[0152]

[0153] Among them, G t [S t [G at time t] t Node S, explored by drones t The sample has been explored in subgraph A. t π represents the set of UAV actions (sample motion paths) in the expert solution, and π represents the controller of the graph neural network model. The cross-entropy loss is calculated between the predicted motion path obtained by the controller based on the explored subgraph and the sample motion path.

[0154] In some embodiments, step 120 may include:

[0155] Based on the dead zone cluster pooling module, cluster pooling operation is performed on the explored subgraph to obtain the first subgraph corresponding to the explored subgraph output by the dead zone cluster pooling module;

[0156] The graph aggregation module processes the first subgraph to obtain multiple second subgraphs output by the graph aggregation module.

[0157] The strategy module performs cascaded processing on multiple second subgraphs to obtain the target motion path output by the strategy module.

[0158] In this embodiment, such as Figure 3 As shown, the graph neural network model can include a sequentially connected dead zone cluster pooling module (DTCModule), graph aggregation module (K-layer AGG Module), and policy module.

[0159] The dead zone cluster pooling module can perform cluster pooling operations on the explored subgraph and output the first subgraph corresponding to the explored subgraph.

[0160] The size of the first subgraph is smaller than that of the explored subgraphs.

[0161] The graph aggregation module can encode, aggregate, and decode the first subgraph, and can output multiple second subgraphs at different aggregation stages.

[0162] The strategy module can perform cascading processing on multiple second subgraphs to output the target motion path.

[0163] In some embodiments, performing cluster pooling operations on the explored subgraph based on the dead-zone cluster pooling module to obtain the first subgraph corresponding to the explored subgraph output by the dead-zone cluster pooling module may include:

[0164] In the case that there are no other drones, regions of interest and boundary regions in the receptive field of at least one drone besides the target drone, the target cell is obtained based on the number of adjacent targets in the target region.

[0165] Based on the interconnected nodes within the target unit, obtain at least one cluster node corresponding to the target unit;

[0166] Process the connected cluster nodes in adjacent target units to obtain the edge features between the connected cluster nodes;

[0167] The first subgraph is obtained based on at least one cluster node and the features of each edge.

[0168] In this embodiment, the boundary region is a grid that is adjacent to the unexplored region and the explored region.

[0169] If the K-order receptive field of UAV node i does not contain other UAVs, regions of interest, or boundary regions, this receptive field region can be called the K-order dead region of UAV node i, and then cluster pooling operations can be performed on the explored subgraph.

[0170] The target unit includes one or more cluster nodes.

[0171] The position of a cluster node is the average position of all nodes within the cluster, and the edge feature between cluster nodes can be one-third of the distance between cluster nodes.

[0172] In actual implementation, such as Figure 4 As shown, when there are no other drones, regions of interest, or boundary region nodes within the K-order neighbors of the target drone, performing Clupool clustering on the explored subgraph can aggregate adjacent 3x3 grid regions into a single target cell, determine the connected set of nodes as a cluster, and construct a first-order cluster node. When multiple clusters exist within the target unit, multiple cluster nodes can be constructed. The formula for calculating the cluster node characteristics is as follows:

[0173]

[0174] Among them, c vi Let c be the set of grid points contained in node i. vp,1 Let p be the set of raster points contained in the cluster node obtained by first-order cluster pooling.

[0175] An edge between two adjacent cells can be constructed based on the cluster nodes that are connected in the adjacent cells.

[0176] The cluster node position is the average of the positions of all nodes within the cluster. The edge feature between the constructed cluster nodes is the distance between cluster nodes divided by 3. This applies to the explored subgraph G. t After processing by CluPool, G t It can be compressed into a smaller first subgraph consisting of cluster nodes.

[0177] Furthermore, if there are still drone nodes in the K-order dead zone in the first subgraph after pooling, CluPool calculation can be performed again.

[0178] In some embodiments, processing the first subgraph based on the graph aggregation module to obtain multiple second subgraphs output by the graph aggregation module includes:

[0179] The third subgraph is obtained by performing a nonlinear transformation on the cluster nodes and edge features corresponding to the first subgraph based on the graph encoding module.

[0180] Multiple graph aggregation operation modules are used to perform multiple aggregation operations on the first subgraph to obtain multiple fourth subgraphs;

[0181] The third subgraph and multiple fourth subgraphs are decoded using multiple graph decoding modules to obtain multiple second subgraphs.

[0182] In this embodiment, such as Figure 3 As shown, the graph aggregation module may include a graph encoding module (encoder), multiple sequentially connected graph aggregation operation modules (GN-block), and multiple graph decoding modules (decoder).

[0183] The output of the graph encoding module is connected to the graph aggregation operation module and the graph decoding module, respectively, and the output of each graph aggregation module is connected to the graph decoding module.

[0184] The graph encoding module can embed the cluster nodes and edge features corresponding to the first subgraph into a representation. Two multilayer perceptrons (MLPs) can be used to perform nonlinear transformations on the cluster nodes and edge features respectively, and output to a specified dimension to obtain the third subgraph.

[0185] The graph decoding module can decode both encoded graphs and graphs after aggregation operations of various orders.

[0186] In actual execution, the graph encoding module (encoder) can embed and represent the graph node features and edge features in the first subgraph. It can use two multilayer perceptrons (MLPs) to perform nonlinear transformations of node features and edge features respectively, and output them to the specified dimension.

[0187] The graph aggregation module GN-block is the backbone for performing aggregation operations on the graph. Each GN operation allows the current node to aggregate information about its neighboring nodes. The input graph data is: G = {{v i},{e k After an update, the graph data is G′={{v}. i ′},{e k The GN-block operation feature update formula is shown below:

[0188]

[0189] Among them, MLP v and MLP e All are 3-layer MLP networks with 16 nodes, e k Original edge features and Let e′ be the feature of the two endpoints of the edge. k For the updated edge features, v i For the original node features, v′ i For the updated node features, Let E′ be the average value of the features of the neighboring edges of node i. i Let i be the updated set of edges connected to node i.

[0190] The graph decoding module (decoder) can perform K stacked aggregation operations. It can decode both the encoded graph and the graphs after each level of aggregation operations. The decoder can be a 3-layer, 16-node MLP network, outputting to a specified dimension to obtain...

[0191] In some embodiments, cascading multiple second subgraphs based on the strategy module to obtain the target motion path output by the strategy module may include:

[0192] Cascade multiple second subgraphs to obtain a fifth subgraph;

[0193] The probability values ​​corresponding to each candidate motion path in the fifth subgraph are obtained based on the activation function layer, and the candidate motion path corresponding to the largest probability value among multiple probability values ​​is determined as the target motion path.

[0194] In this embodiment, the strategy module includes an activation function (softmax) layer.

[0195] In actual implementation, it is possible to... The concat cascade is performed, and then the feature values ​​of the edges connected to the node where the UAV is located are sampled based on the nonlinear MLP module. Then, the probability values ​​of each candidate motion path are obtained based on the softmax layer, and the one with the largest probability is taken as the target motion path.

[0196] The multi-UAV target coverage path planning method based on cluster pooling graph neural network provided in this application's embodiments processes the explored subgraph using a cluster pooling method, which can more effectively utilize environmental information when generating planned paths, promotes coordination among multiple UAVs, and thus improves the efficiency of inspection tasks. By constructing a graph neural network model, UAVs can autonomously plan paths even in complex environments, reducing manpower costs and improving the accuracy and precision of the generated motion paths.

[0197] The following describes the multi-UAV target coverage path planning device based on clustered pooling graph neural network provided in this application. The multi-UAV target coverage path planning device based on clustered pooling graph neural network described below can be referred to in correspondence with the multi-UAV target coverage path planning method based on clustered pooling graph neural network described above.

[0198] The multi-UAV target coverage path planning method based on clustered pooling graph neural networks provided in this application can be executed by a multi-UAV target coverage path planning device based on clustered pooling graph neural networks. This application example illustrates the multi-UAV target coverage path planning device based on clustered pooling graph neural networks provided in this application by using the multi-UAV target coverage path planning device based on clustered pooling graph neural networks to execute the multi-UAV target coverage path planning method based on clustered pooling graph neural networks.

[0199] This application also provides a multi-UAV target coverage path planning device based on clustered pooling graph neural networks.

[0200] like Figure 7 As shown, the multi-UAV target coverage path planning device based on clustered pooling graph neural network includes: a first processing module 710, a second processing module 720, a third processing module 730 and a fourth processing module 740.

[0201] The first processing module 710 is used to obtain the explored sub-map of each UAV at the current moment based on the position information of each UAV at the current moment and the explored area information of each UAV in the target area.

[0202] The second processing module 720 is used to input the explored sub-maps of at least one UAV among all UAVs into the graph neural network model when the total coverage area of ​​the explored sub-maps of all UAVs is less than the target area, and to obtain the target motion path of at least one UAV output by the graph neural network model.

[0203] The third processing module 730 is used to control at least one UAV to move based on the target movement path, and to acquire the position information and explored area information of at least one UAV after movement, so as to update the explored sub-map of at least one UAV.

[0204] The fourth processing module 740 is used to control each UAV to stop moving when the total coverage area of ​​the updated explored sub-map corresponding to all UAVs overlaps with the target area.

[0205] According to the multi-UAV target coverage path planning device based on clustered pooling graph neural network provided in this application embodiment, the device performs local path planning for the UAV by using environmental information such as the explored area information and the location information of the UAV, and generates a new local path plan based on the environmental information after the UAV moves. The process is iterated until the inspection task is finally completed. The device effectively utilizes environmental information during the inspection process, can automatically collect environmental information for path planning, does not require a lot of manpower, promotes coordination between multiple UAVs, reduces missed or repeated inspections, and improves the efficiency of coverage task completion.

[0206] In some embodiments, the multi-UAV target coverage path planning device based on clustered pooling graph neural networks may further include a fifth processing module for training a graph neural network model based on the following steps:

[0207] Obtain the explored sub-map of multiple samples corresponding to a single data collection cycle using a sample drone;

[0208] Based on the map information of the complete raster map, determine the sample movement path corresponding to the explored sub-map of each sample; at least some sample movement paths constitute a coherent movement path.

[0209] A graph neural network model is trained using multiple explored subgraphs as samples and the motion paths of the samples corresponding to each explored subgraph as sample labels.

[0210] In some embodiments, a single acquisition cycle includes multiple samples at a first moment, and the fifth processing module can also be used for:

[0211] Based on each grid point in the complete grid map, obtain a set of incomplete obstacle grids from the complete grid map;

[0212] Based on the sample location information of the sample UAV in the complete grid map at the first moment and the distance between each incomplete obstacle grid point in the incomplete obstacle grid set, the incomplete obstacle grid set is processed to obtain the sample explored sub-map corresponding to the sample UAV at the first moment.

[0213] In some embodiments, the fifth processing module may also be used for:

[0214] Based on the area ratio of the region of interest corresponding to each grid point in the complete grid map, obtain the set of interest grids;

[0215] The first grid set is obtained based on the interest grid set, the incomplete obstacle grid set, and the perception range parameters of the sample UAV;

[0216] Based on the viewpoint grid set and interest grid set corresponding to the first grid set, a global keypoint grid set is constructed.

[0217] Based on the set of nodes in the explored subgraph of the sample corresponding to the global key point grid set, path planning is performed on the sample UAV to obtain the sample motion path corresponding to the explored subgraph.

[0218] In some embodiments, the fifth processing module may also be used for:

[0219] Multiple sample explored subgraphs are input into a graph neural network model to obtain the predicted motion path corresponding to each sample explored subgraph output by the graph neural network model;

[0220] Based on the predicted motion path and the motion path of the sample corresponding to the explored sub-graph, a target loss function is constructed.

[0221] A graph neural network model is trained based on a target loss function.

[0222] In some embodiments, the graph neural network model includes a sequentially connected dead zone cluster pooling module, a graph aggregation module, and a policy module, and the second processing module 720 can also be used for:

[0223] Based on the dead zone cluster pooling module, cluster pooling operation is performed on the explored subgraph to obtain the first subgraph corresponding to the explored subgraph output by the dead zone cluster pooling module;

[0224] The graph aggregation module processes the first subgraph to obtain multiple second subgraphs output by the graph aggregation module.

[0225] The strategy module performs cascaded processing on multiple second subgraphs to obtain the target motion path output by the strategy module.

[0226] In some embodiments, the second processing module 720 may also be used for:

[0227] If, within the receptive field corresponding to the target drone, there are no other drones, regions of interest, or boundary regions among the target drones, the target cells are obtained based on the number of adjacent targets in the target region; the boundary region is a grid of unexplored regions adjacent to explored regions.

[0228] Based on the interconnected nodes within the target unit, obtain at least one cluster node corresponding to the target unit;

[0229] Process the connected cluster nodes in adjacent target units to obtain the edge features between the connected cluster nodes;

[0230] The first subgraph is obtained based on at least one cluster node and the features of each edge.

[0231] In some embodiments, the graph aggregation module includes a graph encoding module, a plurality of sequentially connected graph aggregation operation modules, and a plurality of graph decoding modules. The output of the graph encoding module is connected to the graph aggregation operation module and the graph decoding module, respectively, and the output of each graph aggregation module is connected to the graph decoding module. The second processing module 720 can also be used for:

[0232] The third subgraph is obtained by performing a nonlinear transformation on the cluster nodes and edge features corresponding to the first subgraph based on the graph encoding module.

[0233] Multiple graph aggregation operation modules are used to perform multiple aggregation operations on the first subgraph to obtain multiple fourth subgraphs;

[0234] The third subgraph and multiple fourth subgraphs are decoded using multiple graph decoding modules to obtain multiple second subgraphs.

[0235] In some embodiments, the strategy module includes an activation function layer; the second processing module 720 can also be used for:

[0236] Cascade multiple second subgraphs to obtain a fifth subgraph;

[0237] The probability values ​​corresponding to each candidate motion path in the fifth subgraph are obtained based on the activation function layer, and the candidate motion path corresponding to the largest probability value among multiple probability values ​​is determined as the target motion path.

[0238] The multi-UAV target coverage path planning device based on clustered pooling graph neural networks in this application embodiment can be an electronic device or a component within an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc. This application embodiment does not specifically limit the specific implementation.

[0239] The multi-UAV target coverage path planning device based on clustered pooling graph neural networks in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0240] The multi-UAV target coverage path planning device based on clustered pooling graph neural network provided in this application embodiment can achieve... Figures 1 to 6 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0241] In some embodiments, such as Figure 8 As shown, this application embodiment also provides an electronic device 800, including a processor 801, a memory 802, and a computer program stored in the memory 802 and executable on the processor 801. When the program is executed by the processor 801, it implements the various processes of the above-described multi-UAV target coverage path planning method embodiment based on clustered pooling graph neural network and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0242] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0243] On the other hand, this application also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the various processes of the above-described embodiments of the multi-UAV target coverage path planning method based on clustered pooling graph neural networks and achieve the same technical effect. To avoid repetition, these will not be described again here.

[0244] In another aspect, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the various processes of the above-described embodiments of the multi-UAV target coverage path planning method based on clustered pooling graph neural networks, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0245] On another note, this application also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-described multi-UAV target coverage path planning method based on clustered pooling graph neural networks, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0246] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0247] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0248] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0249] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A multi-UAV target coverage path planning method based on clustered pooling graph neural networks, characterized in that, include: Based on the current location information of each UAV and the information of the explored areas of each UAV in the target area, obtain the explored sub-map of each UAV at the current time; If the total coverage area of ​​the explored sub-maps corresponding to all the UAVs is less than the target area, at least one of the explored sub-maps corresponding to the UAVs is input into the graph neural network model to obtain at least one target motion path corresponding to the UAVs output by the graph neural network model. Control at least one of the UAVs to move based on the target movement path, and acquire the position information and explored area information of at least one of the UAVs after movement, so as to update the explored sub-map of at least one of the UAVs; If the total coverage area of ​​the updated explored sub-map corresponding to all the drones coincides with the target area, control each drone to stop moving. The graph neural network model includes a dead zone cluster pooling module, a graph aggregation module, and a policy module connected in sequence. The step of inputting the explored subgraph corresponding to at least one of the UAVs into the graph neural network model to obtain the target motion path corresponding to at least one UAV output by the graph neural network model includes: Based on the dead zone cluster pooling module, cluster pooling operation is performed on the explored subgraph to obtain the first subgraph corresponding to the explored subgraph output by the dead zone cluster pooling module; Based on the graph aggregation module, the first subgraph is processed to obtain multiple second subgraphs output by the graph aggregation module; The multiple second sub-graphs are cascaded based on the strategy module to obtain the target motion path output by the strategy module.

2. The multi-UAV target coverage path planning method based on clustered pooling graph neural networks according to claim 1, characterized in that, The graph neural network model is trained based on the following steps: Obtain the explored sub-map of multiple samples corresponding to a single data collection cycle using a sample drone; Based on map information from the complete raster map, determine the sample movement path corresponding to the explored sub-map for each sample; at least some of the sample movement paths constitute a coherent movement path. The graph neural network model is trained using the explored subgraphs of the multiple samples as samples and the sample motion paths corresponding to the explored subgraphs of each sample as sample labels.

3. The multi-UAV target coverage path planning method based on clustered pooling graph neural networks according to claim 2, characterized in that, The single acquisition cycle includes multiple samples at a first moment, and the multiple sample-explored sub-maps corresponding to the sample acquisition UAV in a single acquisition cycle include: Based on each grid point in the complete grid map, obtain a set of incomplete obstacle grids from the complete grid map; Based on the sample location information of the sample UAV in the complete grid map at the first moment of the sample and the distance between each of the incomplete obstacle grid points in the incomplete obstacle grid set, the incomplete obstacle grid set is processed to obtain the sample explored sub-map corresponding to the sample UAV at the first moment of the sample.

4. The multi-UAV target coverage path planning method based on clustered pooling graph neural networks according to claim 3, characterized in that, The map information based on the complete raster map determines the sample movement path corresponding to the explored sub-map for each sample, including: Based on the area ratio of the region of interest corresponding to each grid point in the complete grid map, obtain the set of interest grids; Based on the interest grid set, the incomplete obstacle grid set, and the perception range parameters of the sample UAV, a first grid set is obtained; Based on the viewpoint grid set and the interest grid set corresponding to the first grid set, a global keypoint grid set is constructed. Based on the global key point grid set corresponding to the node set in the explored subgraph of the sample, path planning is performed on the sample UAV to obtain the sample motion path corresponding to the explored subgraph of the sample.

5. The multi-UAV target coverage path planning method based on clustered pooling graph neural networks according to claim 2, characterized in that, The step of training the graph neural network model using the multiple explored sub-graphs as samples and the sample motion paths corresponding to each explored sub-graph as sample labels includes: The multiple explored sub-graphs of the samples are input into the graph neural network model to obtain the predicted motion path corresponding to each explored sub-graph of the samples output by the graph neural network model; Based on the predicted motion path and the motion path of the sample corresponding to the explored sub-graph of the sample, a target loss function is constructed. The graph neural network model is trained based on the target loss function.

6. The multi-UAV target coverage path planning method based on clustered pooling graph neural networks according to claim 1, characterized in that, The step of performing cluster pooling operations on the explored subgraph based on the dead-zone cluster pooling module to obtain the first subgraph corresponding to the explored subgraph output by the dead-zone cluster pooling module includes: If, within the receptive field corresponding to the target drone, there are no other drones, regions of interest, or boundary regions among the drones mentioned above, the target cells are obtained based on the number of adjacent targets in the target region; the boundary region is a grid of unexplored regions adjacent to explored regions. Based on the interconnected nodes within the target unit, obtain at least one cluster node corresponding to the target unit; Process the connected cluster nodes in adjacent target units to obtain the edge features between the connected cluster nodes; The first subgraph is obtained based on the at least one cluster node and each edge feature.

7. The multi-UAV target coverage path planning method based on clustered pooling graph neural networks according to claim 1, characterized in that, The graph aggregation module includes a graph encoding module, multiple sequentially connected graph aggregation operation modules, and multiple graph decoding modules. The output of the graph encoding module is connected to both the graph aggregation operation module and the graph decoding module, and the output of each graph aggregation module is connected to the graph decoding module. The step of processing the first subgraph based on the graph aggregation module to obtain multiple second subgraphs output by the graph aggregation module includes: Based on the graph encoding module, a nonlinear transformation is performed on the cluster nodes and edge features corresponding to the first subgraph to obtain the third subgraph; Multiple fourth subgraphs are obtained by performing multiple aggregation operations on the first subgraph based on multiple graph aggregation operation modules; The third subgraph and the fourth subgraph are decoded using the multiple graph decoding modules to obtain the multiple second subgraphs.

8. The multi-UAV target coverage path planning method based on clustered pooling graph neural networks according to claim 1, characterized in that, The strategy module includes an activation function layer; the step of cascading the multiple second subgraphs based on the strategy module to obtain the target motion path output by the strategy module includes: The multiple second subgraphs are concatenated to obtain the fifth subgraph; Based on the activation function layer, the probability values ​​corresponding to each candidate motion path in the fifth subgraph are obtained, and the candidate motion path corresponding to the largest probability value among multiple probability values ​​is determined as the target motion path.

9. A multi-UAV target coverage path planning device based on clustered pooling graph neural networks, characterized in that, include: The first processing module is used to obtain the explored sub-map of each UAV at the current time based on the location information of each UAV at the current time and the explored area information of each UAV in the target area. The second processing module is used to input at least one of the explored sub-maps of all the UAVs into the graph neural network model when the total coverage area of ​​the explored sub-maps of all the UAVs is less than the target area, and to obtain at least one target motion path of the UAVs output by the graph neural network model. The third processing module is used to control at least one of the UAVs to move based on the target movement path, and to obtain the position information and explored area information of at least one of the UAVs after movement, so as to update the explored sub-map of at least one of the UAVs; The fourth processing module is used to control each of the drones to stop moving when the total coverage area of ​​the updated explored sub-map corresponding to all the drones overlaps with the target area. The graph neural network model includes a sequentially connected dead-zone cluster pooling module, a graph aggregation module, and a policy module. The second processing module is further used for: Based on the dead zone cluster pooling module, cluster pooling operation is performed on the explored subgraph to obtain the first subgraph corresponding to the explored subgraph output by the dead zone cluster pooling module; Based on the graph aggregation module, the first subgraph is processed to obtain multiple second subgraphs output by the graph aggregation module; The multiple second sub-graphs are cascaded based on the strategy module to obtain the target motion path output by the strategy module.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the multi-UAV target coverage path planning method based on clustered pooling graph neural networks as described in any one of claims 1-8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the multi-UAV target coverage path planning method based on clustered pooling graph neural networks as described in any one of claims 1-8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-UAV target coverage path planning method based on clustered pooling graph neural networks as described in any one of claims 1-8.

Images