Real-time map route construction method and system

Abstract

The invention provides a real-time map route construction method and system. The method comprises the following steps: installing various sensors on an unmanned aerial vehicle to obtain multi-modal data of an environment; converting the multi-modal features into a high-dimensional embedding space, and constructing a real-time three-dimensional environment model based on a multi-modal point cloud reconstruction algorithm; deploying edge computing nodes, and distributing computing tasks to the edge computing nodes; planning a cooperative path among the multiple unmanned aerial vehicles; constructing a multi-unmanned aerial vehicle cooperative global adaptive adjustment mechanism to dynamically adjust the flight path of the unmanned aerial vehicle; and performing priority ranking on the tasks of the flight path of the unmanned aerial vehicle, constructing a path adjustment optimization mechanism according to the real-time flight condition to minimize the task failure risk, and updating the task priority and the scheduling strategy according to the feedback after each task is executed. According to the invention, a plurality of key problems in the prior art are effectively solved, and a technical guarantee is provided for successful application of the unmanned aerial vehicle in more complex application scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a method and system for constructing a real-time map route. Background Art

[0002] In modern information society, with the popularization of Internet technology and intelligent devices, users' activities on the Internet have become more and more diverse, generating a large amount of multimodal data, including text, images, audio, video, and behavior data, etc. These multimodal data not only reflect users' behaviors and interests, but also contain rich psychological state information. Therefore, effectively fusing and analyzing these multimodal data to construct a user's psychological portrait has important application value. However, existing user portrait technologies mainly focus on the processing of single-modal data, such as text-based sentiment analysis, image-based facial expression recognition, etc., lacking the comprehensive utilization of multimodal data, resulting in an incomplete and inaccurate description of users' psychological states.

[0003] Currently, some studies have attempted to fuse multimodal data for obtaining user psychological portraits, but there are still many problems. First, the heterogeneity and diversity of multimodal data make it difficult to effectively fuse different modal data. Existing methods mostly use simple feature concatenation or weighted averaging in cross-modal data fusion, unable to fully utilize the complementary information of different modal data, resulting in poor fusion effects. Second, existing methods mainly rely on static models in sentiment analysis and psychological state prediction, unable to dynamically capture the changes in users' psychological states. And users' psychological states are time-varying and complex, and static models are difficult to adapt to such dynamic changes. Third, social relationships play an important role in users' psychological states, but existing methods rarely consider users' social network data, ignoring the impact of users' interactions with others on their psychological states. Finally, the data processing and model training processes lack an adaptive optimization mechanism, making it difficult to dynamically adjust according to users' real-time behaviors and feedback, resulting in unsatisfactory performance of the model in practical applications. Summary of the Invention

[0004] The object of the present invention is to propose a method and system for constructing a real-time map route, which constructs a real-time map construction and path planning system for unmanned aerial vehicles that can handle complex dynamic environments through the organic combination of multimodal data fusion, edge computing, swarm intelligence collaboration, and adaptive path adjustment. These innovative points not only effectively solve multiple key problems in the prior art, but also provide technical guarantees for the successful application of unmanned aerial vehicles in more complex application scenarios.

[0005] To achieve the above object, in the first aspect of the present invention, a method for constructing a real-time map route is provided, and the method includes: S1. Install multiple sensors on the drone to obtain multi-modal data of the environment, and perform pre-processing on the multi-modal data before data fusion to obtain processed multi-modal feature vectors; among them, the pre-processing before data fusion includes time synchronization processing, normalization processing, spatial alignment of multi-modal data, and multi-modal data fusion; S2. Convert the multi-modal features into a high-dimensional embedding space to obtain high-dimensional multi-modal features Z. Use a multi-modal weighted graph convolutional network to perform spatial modeling on the high-dimensional embedding space to capture the spatial relationships between multi-modal features, and use a feature selection strategy based on a dynamic sparse attention mechanism to remove redundant information from the high-dimensional embedding space to obtain processed high-quality features. Construct a three-dimensional environment model based on a real-time multi-modal point cloud reconstruction algorithm according to the high-quality features; S3. Deploy edge computing nodes, allocate computing tasks to the edge computing nodes. Each edge computing node is deployed within the task area and is responsible for processing the data of the drones in the corresponding area. When a drone connects to the corresponding edge computing node, the drone transmits the collected high-dimensional multi-modal features Z to the edge computing node through real-time communication. The edge computing node performs parallel processing on the high-dimensional multi-modal features Z, constructs a dynamic load balancing mechanism, and if the load is too high due to multiple drones connecting to the same edge node at the same time, re-allocate the edge computing nodes. After the edge computing node processes the data, it feeds back the processing results to the drone in real time; S4. Plan the collaborative paths between multiple drones through a swarm intelligence algorithm combined with a local conflict detection mechanism according to the processing structure of the edge computing node. If the value detected by the conflict detection mechanism exceeds the preset threshold, the path is regarded as a conflict path and path re-selection is performed until a conflict-free path is found; S5. According to the optimized planned path, combined with the real-time collected situation awareness data, construct a global adaptive adjustment mechanism for multi-drone collaboration to dynamically adjust the flight paths of the drones; S6. Prioritize the tasks of the drone flight paths according to the task urgency, task complexity, remaining resources of the drone, and the impact of environmental changes. At the same time, construct a path adjustment and optimization mechanism according to the real-time flight situation to minimize the task failure risk, ensure the optimization of the path, update the task priority and scheduling strategy according to the feedback after each task execution, and improve the execution effect of future tasks; Among them, S4 specifically includes: S401. First, assign a communication node to each drone to form a multi-drone communication network with a distance-based communication weight matrix; S402. Take each drone as an independent ant colony individual and independently and parallelly perform path search and optimization; among them, the path selection strategy of the ant colony individual adopts the following probability model : Among them, represents the pheromone value of the path, represents the heuristic information, represents the path set, and represents the parameter for adjusting the importance of pheromone and heuristic information; represents the th drone's path set at time , and each path is composed of a series of waypoints ; S403. Define the local conflict detection function , which represents the conflict probability between path and its neighbor drone path , and is expressed as follows: Among them, represents the set of neighbor drones of the th drone, is the adjustment parameter for conflict detection; S404. After each drone independently completes path planning, it propagates its path pheromone to the neighbor drones in its communication network and performs global pheromone update.

[0006] Furthermore, the multiple sensors include a high-definition sensor, a lidar, and an environmental sensor; The sensor is used to obtain two-dimensional image data of the environment; The lidar is used to obtain three-dimensional point cloud data; The environmental sensor is used to detect environmental parameters.

[0007] Furthermore, the multi-modal data spatial alignment is based on the spatial transformation matrix for aligning the point cloud data of the lidar with the camera image data , and includes: By calibrating the positions and angles of the sensors on the drone, calculate their respective spatial transformation matrices ; Then, project the point cloud data onto the camera coordinate system through the transformation matrix so that the point cloud data and the image data are represented in the same space as: Among them, represents the aligned point cloud data, Represents the transformation matrix from the lidar coordinate system to the camera coordinate system.

[0008] Further, the S2 specifically includes: S201. Define a multi-modal embedding mapping function to convert the multi-modal data feature vectors into a shared high-dimensional embedding space; S202. Construct a spatial modeling based on the multi-modal weighted graph convolutional network, including: Construct a multi-modal weighted graph , where the node set represents the embedded feature points, and the edge set represents the correlation relationship between the feature points; According to the inter-modal relationship of the nodes, calculate the correlation relationship between the nodes through the Euclidean distance and the modal correlation metric, and define the weight of the edge as: where, and represent the hyperparameters for adjusting the weight, represents the modal correlation, which is calculated by statistically calculating the feature correlation; Define the graph convolution operation to perform convolution on the multi-modal graph and update the node features, which is expressed as follows: where, represents the weighted adjacency matrix, represents the degree matrix, represents the th layer of feature representation, represents the weight matrix of the graph convolution, represents the Frobenius norm regularization term introduced to prevent overfitting, represents the activation function; Perform layer-by-layer aggregation and update on the multi-modal features, and finally obtain the enhanced high-level feature representation ; S203. Construct a feature selection strategy based on the dynamic sparse attention mechanism to select the most important features for modeling the current environment. Among them, define the attention score , which is expressed as follows: where, and represent the trainable parameters of the attention mechanism; perform sparse selection on the features, which is expressed as follows: Here, Denotes an element-wise multiplication operation, and the features after sparse selection Unimportant redundant information is removed; S204. Utilize the selected high-quality features to construct a real-time 3D environmental model through a multi-modal point cloud reconstruction algorithm as follows: Among them, denotes time the features after sparse selection at time denotes time the environmental model at time denotes the model at the previous time the decay term of denotes the time decay coefficient denotes the reconstruction function.

[0009] Furthermore, the allocation of the computing task to the edge computing node is based on the positional relationship between the current position and the edge node, and the connected edge node is selected according to the nearest distance principle, and the calculation is as follows: Among them, denotes the current position of the UAV denotes the position of the selected edge computing node denotes the connection selection function.

[0010] Furthermore, the dynamic load balancing mechanism includes: The load of the edge node is defined as the total amount of tasks it is currently processing, denoted by If the load of node exceeds the threshold , then part of the processing tasks need to be reassigned to other neighboring nodes The reallocated node is selected by the following method of minimizing the load increment: Among them, denotes the load increment of the reallocated task is the target node for reallocating tasks.

[0011] Furthermore, in S404, the formula for global pheromone update is expressed as follows: Among them, denotes the decay factor of pheromone update Indicates the set of neighboring UAVs.

[0012] Furthermore, step S5 specifically includes: S501. The UAV continuously collects environmental data through multi-modal sensors , and converts this data into a multi-dimensional feature vector . An adaptive weight update mechanism is designed to dynamically adjust the feature weights in path planning according to the current sensed data and the data at the previous moment . ; S502. Construct an adaptive path optimization objective function , and optimize the path of the current UAV in real time according to the adjusted feature weights in path planning , to obtain the optimized path , which is expressed as follows: where , , and respectively represent path evaluation functions related to images, point clouds, and environmental data, represents the path smoothness constraint term, represents its weight coefficient, which is used to control the smoothness of the path; S503. Fuse the optimized path with the current environmental model to update the environment and obtain the updated environmental model ; S504. After each UAV completes adaptive path adjustment, share its optimized path and environmental perception data with other UAVs. Among them, the global collaborative objective function is defined as: where represents the number of UAVs, represents the path conflict penalty factor, represents the th th UAV and the

[0013] Furthermore, the priority ranking is calculated as follows: where represents the adjustment coefficient, represents the task completion rate prediction function based on historical data, represents the task urgency, represents the task complexity, represents the remaining resources of the drone, represents the impact of environmental changes; simultaneously construct a path adjustment and optimization mechanism according to the real-time flight situation to minimize the task failure risk and ensure the optimization of the path. Among them, define the path adjustment objective function of task execution , which is expressed as follows: Among them, represents the adjustment coefficient, represents the path smoothness regularization term, which is used to limit the sharp change of the path and ensure flight safety; represents the task completion degree, represents the resource consumption, represents the path safety.

[0014] In the second aspect of the present invention, a real-time map route construction system is provided. The system includes: A flight data acquisition unit, which is used to install a variety of sensors on the drone to obtain multi-modal data of the environment, perform pre-processing on the multi-modal data before data fusion, and obtain the processed multi-modal feature vector; among them, the pre-processing before data fusion includes time synchronization processing, normalization processing, spatial alignment of multi-modal data, and multi-modal data fusion; A three-dimensional environment model construction unit, which is used to convert multi-modal features into a high-dimensional embedding space to obtain high-dimensional multi-modal features Z, use a multi-modal weighted graph convolutional network to perform spatial modeling on the high-dimensional embedding space to capture the spatial relationship between multi-modal features, and use a feature selection strategy based on a dynamic sparse attention mechanism to remove redundant information from the high-dimensional embedding space to obtain processed high-quality features, and construct a real-time three-dimensional environment model based on a multi-modal point cloud reconstruction algorithm according to the high-quality features; An edge computing node deployment unit, which is used to deploy edge computing nodes, allocate computing tasks to edge computing nodes, each edge computing node is deployed in the task area and is responsible for processing the data of the drone in the corresponding area. When the drone is connected to the corresponding edge computing node, the drone transmits the collected high-dimensional multi-modal features Z to the edge computing node through real-time communication. The edge computing node performs parallel processing on the high-dimensional multi-modal features Z, constructs a dynamic load balancing mechanism, and reallocates edge computing nodes if the load is too high due to multiple drones connecting to the same edge node at the same time. After the edge computing node processes the data, it feeds back the processing result to the drone in real time; A path planning unit, which is used to plan the collaborative path among multiple UAVs through a swarm intelligence algorithm combined with a local conflict detection mechanism according to the processing structure of edge computing nodes. If the value detected by the conflict detection mechanism exceeds a preset threshold, the path is regarded as a conflict path and path re-selection is performed until a conflict-free path is found; A path optimization unit, which is used to construct a global adaptive adjustment mechanism for multi-UAV collaboration to dynamically adjust the UAV flight path according to the optimized planned path and the situation awareness data collected in real time; A path task priority sorting unit, which is used to sort the priorities of the tasks of the UAV flight path according to the task urgency, task complexity, remaining resources of the UAV, and the impact of environmental changes. At the same time, a path adjustment and optimization mechanism is constructed according to the real-time flight situation to minimize the task failure risk, ensure the optimization of the path, update the task priority and scheduling strategy according to the feedback after each task execution, and improve the execution effect of future tasks; Among them, the steps executed by the path planning unit specifically include: S401. First, assign a communication node to each UAV to form a multi-UAV communication network based on a distance-based communication weight matrix; S402. Take each UAV as an independent ant colony individual and independently and parallelly execute path search and optimization; among them, the path selection strategy of the ant colony individual adopts the following probability model : Among them, represents the pheromone value of the path, represents the heuristic information, represents the path set, and represent the parameters for adjusting the importance of pheromone and heuristic information; represents the th UAV at time 's path set, and each path consists of a series of waypoints ; S403. Define a local conflict detection function , which represents the conflict probability between path and the path of its neighbor UAV , and is expressed as follows: Among them, represents the set of neighbor UAVs of the th UAV, and is the adjustment parameter for conflict detection; S404. After each UAV independently completes path planning, its path pheromone Propagate to the neighboring drones in its communication network and perform global pheromone update.

[0015] The beneficial technical effects of the present invention are at least as follows: First, the present invention adopts a multi-modal data fusion technology. By using a variety of sensors carried by the drone (such as cameras, lidars, environmental sensors, etc.), it can perceive the environment in real time, and through a deep learning network for data fusion, a high-precision real-time three-dimensional map is generated. This method can effectively solve the problem of the traditional path planning's dependence on static maps, and improve the navigation ability of the drone in complex environments through the dynamically updated map.

[0016] Second, the present invention introduces an edge computing architecture, distributing the computational tasks of map construction and path planning to the edge nodes close to the task execution location. Compared with the traditional centralized computing mode, this architecture can significantly reduce the latency of data processing, improve the real-time performance of the system, and meet the strict requirements for real-time performance during the drone's task execution. In addition, the introduction of edge computing also reduces the computational burden on the drone itself, enabling it to operate efficiently under limited resources.

[0017] Third, to solve the path conflict and planning efficiency problems when multiple drones cooperate to execute tasks, the present invention proposes a cooperative path planning mechanism based on swarm intelligence. By using swarm intelligence algorithms (such as ant colony algorithm, particle swarm optimization algorithm), the system can achieve information sharing and real-time cooperation among multiple drones, effectively avoid path conflicts, and maximize the task execution efficiency. This cooperative mechanism is particularly suitable for complex task scenarios that require multiple drones to work together, such as large-area environmental monitoring or disaster relief.

[0018] Finally, the present invention also integrates an adaptive path adjustment function. Through real-time situation awareness technology, the drone can continuously monitor environmental changes during flight and dynamically adjust the flight path according to the latest perception data. This function effectively solves the problem that traditional path planning methods cannot cope with sudden environmental changes, and improves the safety and reliability of task execution. Brief Description of the Drawings

[0019] The present invention is further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the following drawings.

[0020] Figure 1 It is a flowchart of a real-time map route construction method of the present invention.

[0021] Figure 2 It is a framework diagram of a real-time map route construction system of the present invention. Detailed implementation manners

[0022] The embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation to the present invention.

[0023] As Figure 1 shown, a real-time map route construction method provided by an embodiment of the present invention includes the following steps: S1. Install a variety of sensors on the unmanned aerial vehicle (UAV) to obtain multi-modal data of the environment, and perform pre-processing on the multi-modal data before data fusion to obtain processed multi-modal feature vectors; wherein, the pre-processing before data fusion includes time synchronization processing, normalization processing, spatial alignment of multi-modal data, and multi-modal data fusion.

[0024] Specifically, install a variety of sensors on the UAV to obtain multi-modal data of the environment. Specific sensors include: High-definition camera: used to obtain two-dimensional image data of the environment, marked as , and its data format is , where and respectively represent the height and width of the image, and represents the three color channels of RGB.

[0025] LiDAR: Obtain three-dimensional point cloud data, marked as , and its data format is , representing the point cloud coordinates in space.

[0026] Environmental sensor: mainly used to detect environmental parameters such as temperature, humidity, and wind speed, marked as , and its data format is , respectively representing the scalar values of temperature, humidity, and wind speed.

[0027] Furthermore, the acquisition of multi-modal data must be time-synchronized to ensure the accuracy of data fusion. During the flight of the UAV, the data acquisition frequencies of different sensors may be different. Therefore, it is necessary to align the data of different sensors in time. Let the timestamps of the data collected by the sensors be , , , and the principle of time synchronization is based on the maximum timestamp: For the data with a lower sampling frequency, align it in time to the timestamp by an interpolation method (such as linear interpolation).。The interpolation calculation formula is as follows: Among them is the aligned data, and are the data of adjacent sampling points.

[0028] Furthermore, before performing multi-modal data fusion, it is necessary to preprocess and normalize the data to ensure that different data sources have the same scale and distribution. This is to avoid affecting the subsequent fusion process due to different data scales. The normalization formula is: Among them, is the normalized data, is the mean value of the data, is the standard deviation of the data. This process processes the image data , point cloud data and environmental data separately to ensure that they are on the same scale.

[0029] Furthermore, the data obtained by different sensors has spatial inconsistencies, so spatial alignment is required. For this purpose, a spatial transformation matrix is designed to align the point cloud data of the lidar with the camera image data . This process includes the following steps: First, by calibrating the positions and angles of the sensors on the drone, calculate their respective spatial transformation matrices .

[0030] Then, project the point cloud data through the transformation matrix into the camera coordinate system so that the point cloud data and the image data are represented in the same space: Among them, is the aligned point cloud data, is the transformation matrix from the lidar coordinate system to the camera coordinate system.

[0031] Furthermore, finally, fuse the preprocessed multi-modal data preliminarily. A fusion method based on weight coefficients is designed. Given the image data , point cloud data and environmental data , define the fused feature vector as : Among them, , and are feature extraction functions for different modality data respectively, , and are corresponding data weight coefficients, is the finally fused multi-modal feature vector, which is used for subsequent environment modeling and path planning.

[0032] S2. Convert the multi-modal features to a high-dimensional embedding space to obtain high-dimensional multi-modal features Z. Use a multi-modal weighted graph convolutional network to perform spatial modeling on the high-dimensional embedding space to capture the spatial relationships between multi-modal features and use a feature selection strategy based on a dynamic sparse attention mechanism to remove redundant information from the high-dimensional embedding space to obtain processed high-quality features. Construct a real-time three-dimensional environment model based on the multi-modal point cloud reconstruction algorithm according to the high-quality features.

[0033] Specifically, the multi-modal data feature vector obtained in step 1, includes image features , point cloud features and environmental features . These features need to be further transformed into a shared high-dimensional embedding space to achieve preliminary fusion and subsequent graph convolutional processing. To achieve this goal, we define a multi-modal embedding mapping function where , , represent the embedding mapping functions of image, point cloud and environmental data respectively, represents the concatenation operation of features. Through this mapping operation, the generated embedding representation is obtained, where is the dimension of the high-dimensional embedding space. This step ensures that the multi-modal data is represented in the same space, thus providing a unified input format for subsequent graph convolutional operations.

[0034] Furthermore, to effectively fuse features of different modalities and capture the complex spatial relationships between them, a multi-modal weighted graph convolutional network (MWGCN) is proposed for spatial modeling. First, construct a multi-modal weighted graph , where the node set represents the embedded feature points, and the edge set represents the association relationships between feature points. The association relationships are measured by Euclidean distance and modality correlation, and the edge weight is defined as: Among them, and are hyperparameters for adjusting weights, represents modal correlation and is calculated by statistically calculating feature correlation.

[0035] Next, define the graph convolution operation to perform convolution on the multi-modal graph and update the node features: Among them, is the weighted adjacency matrix, is the degree matrix, represents the feature representation of the th layer, is the weight matrix of the graph convolution, is the Frobenius norm regularization term introduced to prevent overfitting, is the activation function. This graph convolution process aggregates and updates the multi-modal features layer by layer, and finally obtains an enhanced high-level feature representation for modeling the complex environmental structure.

[0036] Furthermore, in traditional convolution operations, all neighbor nodes have the same importance for updating the features of the central node. However, in the actual environment, some features may be more critical for path planning. Therefore, a feature selection strategy based on a dynamic sparse attention mechanism is proposed. This mechanism calculates the importance scores of features, selects the features that are most important for modeling the current environment, and at the same time avoids the influence of redundant features on the model. Define the attention score as calculated by: Among them, and are the trainable parameters of the attention mechanism. Then, perform sparse selection on the features: Here, represents the element-wise multiplication operation. The features after sparse selection remove unimportant redundant information and retain the features that are most important for modeling the environment. These features will more accurately reflect the core elements of the complex environment.

[0037] Furthermore, finally, use the selected high-quality features to construct a real-time three-dimensional environmental model . To adapt to the flight tasks of drones in dynamic environments, the model needs to be updated in real time and reflect the latest environmental changes. The generation of the 3D model is achieved through a multi-modal point cloud reconstruction algorithm: where, is the environmental model at time , is the model at the previous time attenuation term, is the time decay coefficient, ensuring that the model can respond to environmental changes over time. The reconstruction function maps the enhanced features into 3D space to generate the final environmental model.

[0038] Through this dynamic update process, the 3D environmental model can not only reflect the current environmental state, but also provide stable scene understanding based on historical data, providing a stable and accurate basis for path planning and environmental perception.

[0039] S3. Deploy edge computing nodes, allocate computing tasks to edge computing nodes. Each edge computing node is deployed within the task area and is responsible for processing data of drones in the corresponding area. When a drone connects to the corresponding edge computing node, the drone transmits the collected high-dimensional multi-modal features Z to the edge computing node through real-time communication. The edge computing node processes the high-dimensional multi-modal features Z in parallel, constructs a dynamic load balancing mechanism, and reallocates edge computing nodes if the load is too high caused by multiple drones connecting to the same edge node at the same time. After the edge computing node processes the data, it feeds back the processing results to the drone in real time.

[0040] To reduce the latency of real-time data processing of drones, computing tasks are allocated to edge computing nodes. Each edge computing node is deployed within the task area and is responsible for processing data of nearby drones. Assume that the task area is divided into sub-regions, and each sub-region ( ) corresponds to an edge computing node . During the flight of the drone, based on the position relationship between the current position and the edge node, the drone selects the connected edge node by the principle of the nearest distance. We define a connection selection function , which is used to calculate the Euclidean distance between the drone and all edge nodes and select the nearest node: where, is the current position of the drone, is the position of the th edge node. is the selected edge computing node.

[0041] Further, when the drone is connected to the edge computing node After that, the drone transmits the collected multimodal data to the node through real-time communication To improve the computing efficiency, the data Is divided into multiple data shards { , where each shard contains data at different time points or of different modalities. Each shard is processed in parallel in different processing units within the edge node. The processed results Will be aggregated in the edge node: Among them, Represents the processing function for the data shard , Is the result of the final processing. This shard processing and parallel computing architecture ensures the processing of a large amount of data within a short time.

[0042] Further, to address The problem of excessive load that may be caused by the drone connecting to the same edge node simultaneously A dynamic load balancing mechanism is proposed. The load of the edge node is defined as the total amount of its current processing tasks, denoted by . If the load of node Exceeds the threshold , then some processing tasks need to be redistributed to other neighboring nodes . The redistributed node Is selected by the following method of minimizing the load increment: Among them, Represents the load increment of the redistributed task, Is the target node for the redistributed task. This step ensures that the processing capacity of the edge computing node is fully utilized and avoids the overload of individual nodes.

[0043] Further, after the edge computing node finishes processing the data, the result Needs to be fed back to the drone in real time. To ensure the lowest latency, a prediction-based feedback optimization mechanism is proposed. Assuming that the processing time of the edge node is , and the data transmission time is , the system learns from historical data, predicts the processing delay of future tasks, and adjusts the transmission path in advance to reduce the waiting time of data transmission. The total delay Is: Among them, The amount of time reduced by adjusting the transmission path. The feedback result is returned to the UAV within the time, enabling it to adjust the path planning in a timely manner according to the latest data.

[0044] S4. Plan the cooperative path among multiple UAVs through the swarm intelligence algorithm combined with the local conflict detection mechanism according to the processing structure of the edge computing node. If the value detected by the conflict detection mechanism exceeds the preset threshold, the path is regarded as a conflict path and path re-selection is performed until a conflict-free path is found.

[0045] In this step, first assign a communication node to each UAV to form a multi-UAV communication network. The position of each UAV node is represented as , where represents the th UAV. The communication network is represented by the adjacency matrix , where represents that there is a direct communication link between the th and the th UAVs, otherwise . To enhance the stability and robustness of the communication network, a distance-based communication weight matrix is defined and calculated as follows: where is the adjustment parameter for controlling the communication range. This communication network structure ensures the efficient sharing of information among UAVs and provides the necessary support for subsequent cooperative path planning.

[0046] Furthermore, after the communication network is established, a distributed ant colony optimization algorithm (Distributed Ant Colony Optimization, DACO) is proposed for cooperative path planning. Each UAV acts as an independent "ant colony" individual and independently and parallelly performs path search and optimization. Let be the path set of the th UAV at time , and each path consists of a series of waypoints . The path selection strategy of the ant colony algorithm adopts the following probability model: where is the pheromone value of the path, is the heuristic information, is the set of possible paths, and It is a parameter for adjusting the importance of pheromone and heuristic information. Each drone continuously updates its path set based on its own path selection and the path feedback from neighboring drones, thus gradually approaching the globally optimal path.

[0047] Furthermore, in actual multi-drone path planning, path conflicts or congestion problems may occur. To avoid this situation, a local conflict detection and optimization mechanism is proposed. When each drone selects a path, considering the collision risk with neighboring drones, a local conflict detection function is defined to represent the path and the path of its neighboring drone The conflict probability. The definition of the local conflict detection function is as follows: where represents the set of neighboring drones of the th drone, and is the adjustment parameter for conflict detection. If exceeds the preset threshold , then this path is regarded as a conflict path and path re-selection is carried out until a conflict-free path is found. This conflict detection and optimization process is carried out at each stage of path selection to ensure that multi-drones do not collide during flight.

[0048] Furthermore, to achieve cooperation among multi-drones, a path fusion strategy based on global pheromone update is proposed. After each drone independently completes path planning, it will spread its path pheromone to neighboring drones in its communication network and perform global pheromone update. The formula for global pheromone update is: where is the attenuation factor for pheromone update, and represents the set of neighboring drones. Through the update of global pheromone, the path planning of multiple drones gradually converges, thus achieving cooperative flight. This mechanism ensures the path consistency of multiple drones in a complex environment and maximizes the overall task execution efficiency.

[0049] S5. According to the path of the optimized plan and combined with the scenario awareness data collected in real time, construct a global adaptive adjustment mechanism for multi-drone cooperation to dynamically adjust the flight path of the drones.

[0050] During the flight of the drones, real-time environmental perception is crucial. For this reason, the drones continuously collect environmental data through multi-modal sensors (cameras, lidar, environmental sensors, etc.) , and convert these data into multi-dimensional feature vectors . Considering the dynamics and complexity of the environment, the UAV needs to dynamically adjust the weights of various features in its path planning to adapt to real-time changes. For this purpose, an adaptive weight update mechanism is proposed, which is based on the current perception data and the data at the previous moment to dynamically adjust the feature weights in path planning . The formula for weight update is: where is the learning rate, represents the feature change gradient, is the regularization coefficient, is the regularization term, which is used to control the stability of weight update. This formula ensures that the UAV can flexibly adjust the priorities of various features in path planning, making the planning result more in line with the actual situation of the current environment.

[0051] Further, based on the updated feature weights , the UAV optimizes its path in real time during flight . The core of path optimization lies in combining real-time perception data and current path information to form a new path optimization goal. Assuming that the current path consists of multiple waypoints , we define an adaptive path optimization objective function , comprehensively considering factors such as path safety, energy efficiency, obstacle avoidance, and real-time performance: where , and respectively represent path evaluation functions related to image, point cloud, and environmental data, is the path smoothness constraint term, is its weight coefficient, which is used to control the smoothness of the path. By minimizing , the UAV can adaptively adjust the path to ensure that its flight path in a complex environment is both safe and efficient.

[0052] Further, in order to further improve the real-time performance and accuracy of path adjustment, a scenario awareness and path adjustment feedback closed-loop system is proposed. After path adjustment, the optimized path will be fed back to the environment perception module in real time and fused with the current environment model . Through this feedback closed-loop, the environment perception data will be continuously updated, and the weight update and path optimization processes will be triggered again. We define the formula for environment model update as: Among them, is the learning rate for environmental model update, represents the environmental change amount brought by the current path adjustment, is the contribution of the current perception data to the environmental model. Through this closed-loop process, the UAV can achieve continuous adaptive optimization and maintain efficient path planning and adjustment capabilities in complex dynamic environments.

[0053] Furthermore, according to the complexity of multi-UAV collaborative tasks, a global adaptive adjustment mechanism is proposed to ensure the synchronization of path adjustments among multi-UAVs and reduce the risk of conflicts. After each UAV completes the adaptive path adjustment, it shares its optimized path with the environmental perception data to other UAVs. The global collaborative objective function is defined as: Among them, is the number of UAVs, is the path conflict penalty factor, represents the th possibility of path conflict between the th UAV and the th UAV,

[0054] S6. Prioritize the tasks of the UAV flight path according to the task urgency, task complexity, remaining resources of the UAV, and the impact of environmental changes. At the same time, construct a path adjustment optimization mechanism based on the real-time flight situation to minimize the task failure risk, ensure the optimization of the path, update the task priority and scheduling strategy according to the feedback after each task execution, and improve the execution effect of future tasks.

[0055] In a complex UAV task environment, the importance and urgency of different tasks will dynamically change with environmental changes. To reasonably allocate the resources of the UAV, the system must calculate and update the task priority in real time. Let the task set be , and the priority of each task is based on the following factors: task urgency , task complexity , remaining resources of the UAV (such as battery power, computing resources, etc.), and the impact of environmental changes . The priority calculation formula is: Among them, is the adjustment coefficient, is the task completion rate prediction function based on historical data. is the weight adjustment coefficient to ensure that tasks with higher success rates are preferentially selected under similar conditions. This priority calculation formula enables the system to adapt to dynamic changes in complex task environments, ensuring that the most important tasks are processed first when resources are limited. The calculation is as follows: Task urgency : Task urgency is determined by the difference between the deadline of the task and the current time. The calculation method is as follows: Among them, is the deadline of the task, is the current time, is a small constant to prevent the denominator from being zero. The closer the task is to the deadline, the larger the value, indicating that the task is more urgent.

[0056] Task complexity : Task complexity can be simply represented by the amount of resources required for the task, such as computational load, path length, or the number of obstacles. The calculation formula is: Among them, represents the amount of computation required for the task, is the task path length, is the number of obstacles on the path. The more complex the task, the larger the value.

[0057] Remaining resources of the UAV : Remaining resources of the UAV can be measured by the percentage of remaining battery power. The simple calculation is as follows: Among them, is the percentage of the remaining battery power of the UAV. The more resources, the larger the value.

[0058] Impact of environmental changes : Impact of environmental changes is simply represented by the current environmental conditions (such as wind speed). The calculation method is: Among them, is the wind speed at the current moment. The more unstable the environment is, the larger the value, indicating an increase in the difficulty of task execution.

[0059] Task Completion Rate Prediction : Task Completion Rate Prediction can be simply estimated based on the historical completion of the task. Assume the historical success rate is , then: Among them, is the historical completion rate of the task type under similar conditions.

[0060] Furthermore, during the task execution process, the system needs to continuously monitor the progress of the task and adjust the task scheduling and path planning according to the real-time feedback. Let the current task have a feedback of , and the feedback includes the task completion degree , resource consumption , path safety , and the impact of environmental changes . To ensure the successful execution of the task and the safety of the path, a path adjustment and optimization mechanism is proposed, whose goal is to minimize the task failure risk while ensuring the optimization of the path. Define the path adjustment objective function of the task execution as: Among them, is the adjustment coefficient, is the path smoothness regularization term, which is used to limit the sharp changes in the path and ensure flight safety. This path adjustment mechanism enables the UAV to dynamically adjust the path according to the real-time feedback during the task execution process, ensuring the smooth completion of the task.

[0061] Finally, to ensure the continuous optimization and adaptive adjustment of the system, a feedback loop and continuous optimization mechanism is proposed. The feedback after each task execution will be used to update the task priority and scheduling strategy to improve the execution effect of future tasks. The feedback loop is defined as follows: Among them, is the learning rate, is the task management optimization gradient based on the global feedback. Through this feedback loop, the system can continuously learn and adapt to the new task environment, thereby improving the dynamic response ability and task execution efficiency of the entire system.

[0062] As shown in Figure 2 , an embodiment of the present invention further provides a real-time map route construction system, which includes: A flight data acquisition unit 101, configured to install a variety of sensors on the unmanned aerial vehicle to obtain multi-modal data of the environment, and perform pre-data fusion processing on the multi-modal data to obtain a processed multi-modal feature vector; wherein, the pre-data fusion processing includes time synchronization processing, normalization processing, spatial alignment of multi-modal data, and multi-modal data fusion; A three-dimensional environment model construction unit 102, configured to convert multi-modal features into a high-dimensional embedding space to obtain high-dimensional multi-modal features Z, use a multi-modal weighted graph convolutional network to perform spatial modeling on the high-dimensional embedding space to capture the spatial relationship between multi-modal features, and use a feature selection strategy based on a dynamic sparse attention mechanism to remove redundant information from the high-dimensional embedding space to obtain processed high-quality features, and construct a real-time three-dimensional environment model based on a multi-modal point cloud reconstruction algorithm according to the high-quality features; An edge computing node deployment unit 103, configured to deploy edge computing nodes, allocate computing tasks to the edge computing nodes, each edge computing node is deployed within the task area and is responsible for processing the data of the unmanned aerial vehicle in the corresponding area. When the unmanned aerial vehicle is connected to the corresponding edge computing node, the unmanned aerial vehicle transmits the collected high-dimensional multi-modal features Z to the edge computing node through real-time communication. The edge computing node performs parallel processing on the high-dimensional multi-modal features Z, constructs a dynamic load balancing mechanism, and if the load caused by multiple unmanned aerial vehicles connecting to the same edge node at the same time is too high, reallocate the edge computing nodes. After the edge computing node processes the data, it feeds back the processing result to the unmanned aerial vehicle in real time; A path planning unit 104, configured to plan the collaborative path between multiple unmanned aerial vehicles through a swarm intelligence algorithm combined with a local conflict detection mechanism according to the processing structure of the edge computing node. If the value detected by the conflict detection mechanism exceeds a preset threshold, the path is regarded as a conflict path and path re-selection is performed until a conflict-free path is found; A path optimization unit 105, configured to construct a global adaptive adjustment mechanism for multi-unmanned aerial vehicle collaboration according to the optimized planned path and in combination with the real-time collected situation awareness data to dynamically adjust the flight path of the unmanned aerial vehicle; A path task priority sorting unit 106, configured to sort the tasks of the unmanned aerial vehicle flight path according to the task urgency, task complexity, remaining resources of the unmanned aerial vehicle, and the impact of environmental changes. At the same time, construct a path adjustment and optimization mechanism according to the real-time flight situation to minimize the task failure risk, ensure the optimization of the path, update the task priority and scheduling strategy according to the feedback after each task execution, and improve the execution effect of future tasks; Among them, the steps executed by the path planning unit specifically include: S401. First, assign a communication node to each drone to form a multi-drone communication network with a communication weight matrix based on distance; S402. Treat each drone as an independent ant colony individual and perform path search and optimization independently and in parallel; among them, the path selection strategy of the ant colony individual adopts the following probability model : Among them, represents the pheromone value of the path, represents the heuristic information, represents the set of paths, and represent the parameters that adjust the importance of pheromone and heuristic information; represents the th drone at time 's set of paths, and each path consists of a series of waypoints ; S403. Define a local conflict detection function , which represents the conflict probability between path and the path of its neighbor drone , and is expressed as follows: Among them, represents the set of neighbor drones of the th drone, is the adjustment parameter for conflict detection; S404. After each drone independently completes path planning, it propagates its path pheromone to the neighbor drones in its communication network and performs global pheromone update.

[0063] Unless otherwise specifically stated, the relative steps, numerical expressions, and numerical values of the components and steps set forth in these embodiments do not limit the scope of the present application.

[0064] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0065] In the description of this application, it should be noted that the orientation or positional relationship indicated by terms such as "upper" and "lower" is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the inventive product is usually placed during use. It is only for the convenience of describing this application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to this application.

[0066] In the description of this application, it should also be noted that unless otherwise clearly specified and limited, the terms "set", "install", "connected", and "connected to" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific situations.

[0067] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of this application, rather than to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the various embodiments of this application.

Claims

1. A real-time map route construction method, characterized in that: The method comprises: S1. Install multiple sensors on the drone to obtain multimodal data of the environment, and process the multimodal data before data fusion to obtain a processed multimodal feature vector; wherein the processing before data fusion includes time synchronization processing, normalization processing, multimodal data spatial alignment and multimodal data fusion; S2. Convert the multimodal features into a high-dimensional embedding space to obtain high-dimensional multimodal features Z, use a multimodal weighted graph convolutional network to perform spatial modeling on the high-dimensional embedding space to capture the spatial relationship between the multimodal features, and use a feature selection strategy based on a dynamic sparse attention mechanism to remove redundant information from the high-dimensional embedding space to obtain processed high-quality features. Based on the high-quality features, a real-time three-dimensional environment model based on a multimodal point cloud reconstruction algorithm is constructed; S3. Deploy edge computing nodes and assign computing tasks to edge computing nodes. Each edge computing node is deployed in the task area and is responsible for processing the data of drones in the corresponding area. When the drone is connected to the corresponding edge computing node, the drone transmits the collected high-dimensional multimodal features Z to the edge computing node through real-time communication. The edge computing node processes the high-dimensional multimodal features Z in parallel and builds a dynamic load balancing mechanism. If the load is too high due to multiple drones connecting to the same edge node at the same time, the edge computing node is reallocated. After the edge computing node processes the data, the processing results are fed back to the drone in real time. S4. According to the processing structure of the edge computing node, the collaborative path between multiple drones is planned by combining the swarm intelligence algorithm with the local conflict detection mechanism. If the value detected by the conflict detection mechanism exceeds the preset threshold, the path is regarded as a conflict path and the path is reselected until a conflict-free path is found. S5. Based on the optimized planned path and the real-time collected situational awareness data, a global adaptive adjustment mechanism for multi-UAV collaboration is constructed to dynamically adjust the flight path of the UAVs. S6. Prioritize the tasks on the UAV flight path according to the urgency of the task, the complexity of the task, the remaining resources of the UAV, and the impact of environmental changes. At the same time, build a path adjustment optimization mechanism based on the real-time flight situation to minimize the risk of task failure and ensure the optimization of the path. Update the task priority and scheduling strategy based on the feedback after each task execution; Wherein, the S4 specifically includes: S401, firstly assigning a communication node to each drone to form a multi-drone communication network based on a distance-based communication weight matrix; S402: Treat each drone as an independent ant colony individual and perform path search and optimization independently and in parallel; wherein the path selection strategy of the ant colony individual adopts the following probability model : in, represents the pheromone value of the path, represents heuristic information, Represents a set of paths. and Parameters representing the importance of regulating pheromone and heuristic information; Indicates Drone at the moment A collection of paths, each path consists of a series of waypoints composition; S403, define local conflict detection function , indicating the path Drone paths with its neighbors The conflict probability is expressed as follows: in, Indicates A collection of neighbor drones. is the adjustment parameter for conflict detection; S404. After each UAV has completed its path planning independently, it will send its path pheromone Propagate to neighboring drones in its communication network and perform global pheromone updates.

2. A real-time map route construction method according to claim 1, characterized in that: The multiple sensors include high-definition sensors, lidars, and environmental sensors; The sensor is used to obtain two-dimensional image data of the environment; The laser radar is used to obtain three-dimensional point cloud data; The environmental sensor is used to detect environmental parameters.

3. A real-time map route construction method according to claim 1, characterized in that: The multimodal data spatial alignment is based on the spatial transformation matrix Used to convert laser radar point cloud data With camera image data Alignment, including: By calibrating the position and angle of each sensor on the drone, calculate the respective spatial transformation matrix ; Then, the point cloud data By transforming the matrix Projected to the camera coordinate system, the point cloud data and image data are expressed in the same space as: in, Represents the aligned point cloud data, Represents the transformation matrix from the lidar coordinate system to the camera coordinate system.

4. A real-time map route construction method according to claim 1, characterized in that: The S2 specifically includes: S201. Define a multimodal embedding mapping function to convert a multimodal data feature vector into a shared high-dimensional embedding space; S202. Constructing spatial modeling based on multimodal weighted graph convolutional network, including: Constructing a multimodal weighted graph , where the node set Represents the embedded feature points and edge sets Indicates the association relationship between feature points; According to the modal relationship between nodes, the association relationship between nodes is calculated by Euclidean distance and modal correlation measurement, and the weight of the edge is defined for: in, and represents the hyperparameters for adjusting weights, represents the modal correlation, which is calculated by statistical feature correlation; Define graph convolution operations on multimodal graphs Perform convolution and update node features, which are expressed as follows: in, represents the weighted adjacency matrix, represents the degree matrix, Indicates The feature representation of the layer, represents the weight matrix of graph convolution, Represents the Frobenius norm regularization term introduced to prevent overfitting. represents the activation function; Aggregate and update multimodal features layer by layer to finally obtain enhanced high-level feature representation ; S203. Construct a feature selection strategy based on a dynamic sparse attention mechanism to select the most important features for modeling the current environment, where the attention score is defined , which is expressed as follows: in, and Represents the trainable parameters of the attention mechanism; sparse selection of features is performed as follows: here, Represents element-by-element multiplication, through the sparsely selected features Removed unimportant redundant information; S204. Using selected high-quality features , build a real-time 3D environment model through a multimodal point cloud reconstruction algorithm , which is expressed as follows: in, Indicates time The features after sparse selection at the moment, Indicates time The environment model at the moment, Represents the model of the previous moment The attenuation term, represents the time attenuation coefficient, Represents the reconstruction function.

5. A real-time map route construction method according to claim 1, characterized in that: The computing task is assigned to the edge computing node based on the positional relationship between the current position and the edge node, and the connected edge node is selected by the principle of the closest distance. The calculation is as follows: in, Indicates the current position of the drone. Indicates The location of the edge nodes, Indicates the selected edge computing node, Represents the connection selection function.

6. A real-time map route construction method according to claim 5, characterized in that: The dynamic load balancing mechanism includes: The load of an edge node is defined as the total amount of tasks it is currently processing. It means that if the node The load exceeds the threshold , then some processing tasks need to be reallocated to other neighboring nodes , the reallocated nodes Select by minimizing the load increment as follows: in, represents the load increment of the redistributed tasks, The target node to which the task is reassigned.

7. A real-time map route construction method according to claim 1, characterized in that: In S404, the formula for updating the global pheromone is expressed as follows: in, represents the decay factor of pheromone renewal, Represents the set of neighbor drones.

8. A real-time map route construction method according to claim 4, characterized in that: The S5 specifically includes: S501, UAV continuously collects environmental data through multimodal sensors , and convert these data into multidimensional feature vectors , design an adaptive weight update mechanism based on the current perception data With the previous moment data Dynamically adjust the feature weights in path planning ; S502: Constructing an adaptive path optimization objective function , according to the feature weights in the adjusted path planning Optimize the current drone’s path in real time , get the optimal path , which is expressed as follows: in, , and Respectively represent the path evaluation functions related to image, point cloud and environment data, represents the path smoothness constraint, Represents its weight coefficient, which is used to control the smoothness of the path; S503, optimize the path With the current environmental model Fusion is performed to update the environment and obtain an updated environment model ; S504: After each UAV completes the adaptive path adjustment, optimize its path Environmental perception data Shared with other drones, where the global collaborative objective function Defined as: in, Indicates the number of drones, represents the path conflict penalty factor, Indicates The first drone The possibility of conflicting drone paths. Represents the global environment consistency constraint.

9. A real-time map route construction method according to claim 1, characterized in that: The priority ranking is calculated as follows: in, represents the adjustment coefficient, represents the task completion rate prediction function based on historical data, Indicates the urgency of the task. Indicates the complexity of the task, Indicates the remaining resources of the drone. Indicates the impact of environmental changes; At the same time, the path adjustment optimization mechanism is constructed according to the real-time flight situation to minimize the risk of mission failure and ensure the optimization of the path, wherein the path adjustment objective function of the mission execution is defined , which is expressed as follows: in, represents the adjustment coefficient, represents the path smoothness regularization term, which is used to limit the sharp changes of the path and ensure flight safety; Indicates the degree of task completion. Indicates resource consumption, Indicates path security.

10. A real-time map route construction system, characterized in that: The system comprises: A flight data acquisition unit is used to install multiple sensors on the UAV to obtain multimodal data of the environment, and to process the multimodal data before data fusion to obtain a processed multimodal feature vector; wherein the processing before data fusion includes time synchronization processing, normalization processing, multimodal data spatial alignment and multimodal data fusion; A three-dimensional environment model construction unit is used to convert multimodal features into a high-dimensional embedding space to obtain a high-dimensional multimodal feature Z, use a multimodal weighted graph convolutional network to perform spatial modeling on the high-dimensional embedding space to capture the spatial relationship between multimodal features, and use a feature selection strategy based on a dynamic sparse attention mechanism to remove redundant information from the high-dimensional embedding space to obtain processed high-quality features, and construct a real-time three-dimensional environment model based on a multimodal point cloud reconstruction algorithm based on the high-quality features; The edge computing node deployment unit is used to deploy edge computing nodes and assign computing tasks to edge computing nodes. Each edge computing node is deployed in the task area and is responsible for processing the data of drones in the corresponding area. When the drone is connected to the corresponding edge computing node, the drone transmits the collected high-dimensional multimodal features Z to the edge computing node through real-time communication. The edge computing node processes the high-dimensional multimodal features Z in parallel and builds a dynamic load balancing mechanism. If the load caused by multiple drones connecting to the same edge node is too high at the same time, the edge computing node is reallocated. After the edge computing node processes the data, the processing results are fed back to the drone in real time. The path planning unit is used to plan the collaborative path between multiple drones based on the processing structure of the edge computing node through a swarm intelligence algorithm combined with a local conflict detection mechanism. If the value detected by the conflict detection mechanism exceeds a preset threshold, the path is regarded as a conflict path and the path is reselected until a conflict-free path is found; The path optimization unit is used to dynamically adjust the flight path of drones by building a global adaptive adjustment mechanism for multi-drone collaboration based on the optimized planned path and combining it with the real-time collected situational awareness data; The path task priority sorting unit is used to prioritize the tasks of the UAV flight path according to the urgency of the task, the complexity of the task, the remaining resources of the UAV and the impact of environmental changes. At the same time, a path adjustment optimization mechanism is built according to the real-time flight situation to minimize the risk of task failure and ensure the optimization of the path. The task priority and scheduling strategy are updated according to the feedback after each task execution to improve the execution effect of future tasks; The steps performed by the path planning unit specifically include: S401, firstly assigning a communication node to each drone to form a multi-drone communication network based on a distance-based communication weight matrix; S402: Treat each drone as an independent ant colony individual and perform path search and optimization independently and in parallel; wherein the path selection strategy of the ant colony individual adopts the following probability model : in, represents the pheromone value of the path, represents heuristic information, Represents a set of paths. and Parameters representing the importance of regulating pheromone and heuristic information; Indicates Drone at the moment A collection of paths, each path consists of a series of waypoints composition; S403, define local conflict detection function , indicating the path Drone paths with its neighbors The conflict probability is expressed as follows: in, Indicates A collection of neighbor drones. is the adjustment parameter for conflict detection; S404. After each UAV has completed its path planning independently, it will send its path pheromone Propagate to neighboring drones in its communication network and perform global pheromone updates.

Images