Unmanned aerial vehicle positioning method based on low-orbit satellite network

Abstract

The invention provides an unmanned aerial vehicle positioning method based on a low earth orbit satellite network, the method is applied to a first unmanned aerial vehicle, and the method comprises the following steps: issuing a first patrol parameter through a distributed information channel of the low earth orbit satellite network; acquiring three-dimensional point cloud data of the surrounding environment through a plurality of sensors; inputting the three-dimensional point cloud data into a preset feature extraction network to obtain a first environment feature; determining a target residual feature according to the first environment feature, generating a supplementary request according to the target residual feature and the first patrol parameter, and sending the supplementary request to a distributed information channel, so that a second unmanned replies a plurality of second environment features and corresponding second patrol parameters; inputting the first patrol parameter, the second patrol parameter, the first environment feature and the second environment feature into a preset visual fusion network to obtain a three-dimensional environment model of the surrounding environment; and determining a navigation landmark according to the three-dimensional environment modeling, and performing navigation positioning according to the navigation landmark.

Description

Technical Field

[0001] This application relates to the technical field of UAV positioning, and particularly to a UAV positioning method based on a low-earth orbit satellite network. Background Art

[0002] Currently, with the rapid development of UAV technology, its applications in fields such as surveying and mapping, monitoring, and search and rescue are becoming increasingly widespread. However, traditional UAV positioning methods mainly rely on satellite navigation systems such as GPS, and are susceptible to factors such as signal occlusion and multipath effects in complex environments, resulting in a decrease in positioning accuracy or even failure. This problem severely restricts the application scope and effect of UAVs in complex environments such as urban canyons and dense forest areas. For UAVs with visual positioning and navigation functions, due to the need to operate multiple sensors and visual analysis modules, there are problems of high power consumption and small payload. Summary of the Invention

[0003] This application provides a UAV positioning method based on a low-earth orbit satellite network, which is used to reduce the power consumption of UAV visual navigation and increase the payload of UAVs.

[0004] In a first aspect, an embodiment of this application provides a UAV positioning method based on a low-earth orbit satellite network, which is applied to a first UAV. The first UAV and a second UAV establish a distributed information channel through the low-earth orbit satellite network. The method includes:

[0005] Publish first inspection parameters through the distributed information channel;

[0006] Obtain three-dimensional point cloud data of the surrounding environment through multiple sensors;

[0007] Input the three-dimensional point cloud data into a preset feature extraction network to obtain a first environmental feature;

[0008] Determine a target residual feature according to the first environmental feature, generate a supplementary request according to the target residual feature and the first inspection parameters, and send the supplementary request to the distributed information channel so that the second UAV can reply with multiple second environmental features and corresponding second inspection parameters;

[0009] Input the first inspection parameters, the second inspection parameters, the first environmental feature, and the second environmental feature into a preset visual fusion network to obtain a three-dimensional environmental model of the surrounding environment;

[0010] Determine navigation landmarks according to the three-dimensional environmental model, and perform navigation positioning according to the navigation landmarks.

[0011] In a second aspect, an embodiment of this application provides a UAV positioning device based on a low-earth orbit satellite network. The device includes:

[0012] A communication module, configured to publish first inspection parameters through the distributed information channel;

[0013] A data acquisition module, configured to acquire three-dimensional point cloud data of the surrounding environment through a variety of sensors;

[0014] A feature extraction module, configured to input the three-dimensional point cloud data into a pre-set feature extraction network to obtain first environmental features;

[0015] A request generation module, configured to determine target residual features according to the first environmental features, generate a supplementary request according to the target residual features and the first inspection parameters, and send the supplementary request to the distributed information channel, so that the second unmanned aerial vehicle replies with a plurality of second environmental features and corresponding second inspection parameters;

[0016] A visual fusion module, configured to input the first inspection parameters, the second inspection parameters, the first environmental features and the second environmental features into a pre-set visual fusion network to obtain a three-dimensional environmental model of the surrounding environment;

[0017] A navigation and positioning module, configured to determine navigation landmarks according to the three-dimensional environmental model and perform navigation and positioning according to the navigation landmarks.

[0018] In a third aspect, an embodiment of the present application provides an electronic device, where the electronic device includes a memory and a processor;

[0019] The memory is used to store a computer program;

[0020] The processor is configured to execute the computer program and implement the method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network according to any one of the embodiments of the present application when executing the computer program.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to implement the method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network according to any one of the embodiments of the present application.

[0022] An embodiment of the present application provides a method for positioning an unmanned aerial vehicle (UAV) based on a low-Earth orbit (LEO) satellite network, which is applied to a first UAV. The first UAV and a second UAV establish a distributed information channel through the LEO satellite network. The method includes: publishing first inspection parameters through the distributed information channel; acquiring three-dimensional point cloud data of the surrounding environment through a variety of sensors; inputting the three-dimensional point cloud data into a preset feature extraction network to obtain first environmental features; determining target residual features based on the first environmental features, generating a supplement request according to the target residual features and the first inspection parameters, and sending the supplement request to the distributed information channel so that the second UAV can reply with multiple second environmental features and corresponding second inspection parameters; inputting the first inspection parameters, the second inspection parameters, the first environmental features, and the second environmental features into a preset visual fusion network to obtain a three-dimensional environmental model of the surrounding environment; determining navigation landmarks based on the three-dimensional environmental model and performing navigation positioning according to the navigation landmarks. Through the above method, a distributed information channel with low latency and high transmission rate is established using LEO satellites, realizing multi-UAV formation and information sharing. When the first UAV cannot achieve model convergence based on the first environmental features, that is, when accurate navigation landmarks cannot be obtained, by sending a supplement request to the UAV, according to the target residual features with missing visual information and the first inspection parameters, the second UAV provides more detailed and clearer second environmental parameters and second inspection parameters, expanding the machine vision of the first UAV. From multiple directions, the first UAV outputs more accurate three-dimensional modeling and navigation positioning. At the same time, during the above process, through joint navigation, the hardware requirements for the first UAV are reduced, and part of the computing tasks are concentrated on the second UAV for processing, enabling the first UAV to increase its load capacity and endurance. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0024] Figure 1 FIG. is a schematic diagram of a UAV formation based on a LEO satellite network provided by an embodiment of the present application;

[0025] Figure 2 FIG. is a schematic flowchart of a method for positioning a UAV based on a LEO satellite network provided by an embodiment of the present application;

[0026] Figure 3 FIG. is a schematic block diagram of a UAV positioning device based on a LEO satellite network provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] The flowchart shown in the accompanying drawings is only an example illustration, and does not necessarily include all contents and operations / steps, nor does it necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined or partially merged, so the actual execution order may be changed according to the actual situation.

[0029] It should also be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0030] It should be further understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0031] Explanation of the low-earth orbit satellite network: The low-earth orbit satellite network consists of a large number of small communication satellites with an orbital altitude of about 500 - 1200 kilometers. Compared with traditional communication satellites, it has advantages such as wide coverage, low transmission delay (20 - 40 milliseconds), and high communication rate (above 100 Mbps), and can provide high-speed and stable communication services for unmanned aerial vehicles globally. The low-earth orbit satellite unmanned aerial vehicle can communicate with the low-earth orbit satellite network (hereinafter referred to as the unmanned aerial vehicle).

[0032] Please refer to Figure 1 , Figure 1 which is a schematic diagram of a formation of unmanned aerial vehicles based on a low-earth orbit satellite network provided by an embodiment of this application. As Figure 1 shown, a plurality of first unmanned aerial vehicles 11 and a plurality of second unmanned aerial vehicles 12 are communicatively connected through the low-earth orbit satellite network of the low-earth orbit satellite 13. The low-earth orbit satellite 13 is a low-orbit satellite and can provide high-speed and low-latency communication services.

[0033] Please refer to Figure 2 , Figure 2 which is also a schematic flowchart of a method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network provided by an embodiment of this application. As Figure 2 shown, the method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network is applied to the one as Figure 1The first unmanned aerial vehicle (UAV) shown, the first UAV and the second UAV establish a distributed information channel through a low-earth orbit satellite network, and the specific steps of this method include: S101 - S106.

[0034] S101. Publish the first patrol parameters through the distributed information channel.

[0035] Exemplarily, in a UAV cooperation system, one second UAV serves multiple first UAVs, and there is a group relationship between them. The second UAV subscribes to the session channel of the first UAV on the distributed information channel to obtain the first patrol parameters. In some cases, the second UAV can also provide services to the first UAVs in different groups. The first patrol parameters include: longitude and latitude, altitude, current attitude information, future attitude information, yaw angle, camera parameters, currently executed tasks, task progress, and future direction information.

[0036] Publishing the patrol parameters of the first UAV through the distributed information channel can ensure the efficient transmission and processing of information. This approach is particularly important in multi-UAV cooperative operations, especially in cases where real-time data sharing and task coordination are required.

[0037] S102. Obtain three-dimensional point cloud data of the surrounding environment through a variety of sensors.

[0038] Exemplarily, the sensors at least include: lidar and cameras, and other devices such as optical flow sensors can also be added. The sensors of the first UAV are as few as possible, and instead, it relies on the information sent by the second UAV for three-dimensional modeling. In this way, part of the computing task is handed over to the second UAV, and the self-weight of the first UAV is also reduced, improving the load capacity and endurance of the first UAV.

[0039] S103. Input the three-dimensional point cloud data into a preset feature extraction network to obtain the first environmental features.

[0040] Exemplarily, the preset feature extraction networks include PointNet, PointNet++, DGCNN (Dynamic Graph CNN), etc. These networks can effectively extract meaningful features from unordered point cloud data. The extracted features can be used for various tasks, such as object recognition, scene classification, semantic segmentation, etc. These features provide a high-level understanding of the environment and help the UAV make more intelligent decisions. The parameters of the preset feature extraction network are set according to specific task requirements, and the hyperparameters of the network, such as learning rate, batch size, number of iterations, etc., are adjusted.

[0041] Through the above steps, the first UAV can extract the necessary first environmental features from the three-dimensional point cloud data, and these features can be used for various downstream tasks, such as environmental understanding, positioning and navigation, and cooperation with other UAVs.

[0042] S104. Determine the target residual features according to the first environmental features, generate a supplement request based on the target residual features and the first inspection parameters, and send the supplement request to the distributed information channel so that the second drone can reply with multiple second environmental features and corresponding second inspection parameters.

[0043] In the UAV cooperative inspection system, the first UAV performs 3D modeling based on the acquired first environmental features (such as perception data like images and radar ranging). The features that cause difficulty in model convergence are the target residual features. These residual features represent fuzzy objects that are difficult to accurately identify or classify by the first UAV. Due to the existence of these target residual features, the recognition model of the first UAV is difficult to converge to an ideal state. To solve this problem, a supplement request is generated based on the target residual features and the inspection parameters of the first UAV (such as attitude information, camera shooting parameters, and future direction information, etc.). This request is sent to other UAVs, especially the second UAV, through the distributed information channel. After receiving the request, the second UAV will send back multiple second environmental features collected from different angles or conditions, as well as the corresponding second inspection parameters. This collaborative method helps to improve the understanding and recognition ability of the entire system for complex environments.

[0044] S105. Input the first inspection parameters, the second inspection parameters, the first environmental features, and the second environmental features into a pre-set visual fusion network to obtain a 3D environmental model of the surrounding environment.

[0045] Exemplarily, the second inspection parameters are multiple sets of supplementary flight parameters intelligently generated by the second UAV according to the first inspection parameters. It analyzes the flight path and shooting angle of the first UAV, and then plans multiple flight paths and shooting angles that can supplement the perspective of the first UAV. The second UAV executes the flight tasks specified by the second inspection parameters, takes pictures from multiple supplementary angles, and performs feature extraction to obtain the second environmental features. The visual fusion network is a pre-trained deep learning model specifically used to process and fuse visual data from multiple sources. This is the ultimate goal of the entire process, that is, to construct a comprehensive and accurate 3D model of the surrounding environment.

[0046] This method can obtain more comprehensive, accurate, and detailed environmental information through the collaborative work of two UAVs and multi-angle and multi-scale data collection, combined with advanced data fusion technology, so as to construct a high-quality and high-precision 3D environmental model.

[0047] S106. Determine the navigation landmarks according to the 3D environmental model, and perform navigation and positioning according to the navigation landmarks.

[0048] Exemplarily, the first drone identifies and filters out significant features suitable as navigation landmarks in the three-dimensional environment model, then precisely calculates the position coordinates of these landmarks to generate a navigation map containing these landmarks. During the flight of the drone, the real-time position is determined by identifying the landmarks in the current field of view and matching them with the navigation map, and path planning is carried out accordingly. The position is continuously updated during flight and the path is dynamically adjusted. This method enables the drone to achieve accurate autonomous navigation and positioning in complex environments, improving the task execution efficiency and safety.

[0049] An embodiment of the present application provides a drone positioning method based on a low-earth orbit satellite network, which is applied to the first drone. The first drone and the second drone establish a distributed information channel through the low-earth orbit satellite network. The method includes: publishing the first inspection parameter through the distributed information channel; obtaining three-dimensional point cloud data of the surrounding environment through multiple sensors; inputting the three-dimensional point cloud data into a preset feature extraction network to obtain the first environmental feature; determining the target residual feature according to the first environmental feature, generating a supplement request according to the target residual feature and the first inspection parameter, and sending the supplement request to the distributed information channel so that the second drone replies with multiple second environmental features and corresponding second inspection parameters; inputting the first inspection parameter, the second inspection parameter, the first environmental feature and the second environmental feature into a preset visual fusion network to obtain a three-dimensional environmental modeling of the surrounding environment; determining navigation landmarks according to the three-dimensional environmental modeling and performing navigation positioning according to the navigation landmarks. Through the above method, a distributed information channel with low latency and high transmission rate is established by using low-earth orbit satellites, realizing multi-drone formation and information sharing. When the first drone cannot achieve model convergence according to the first environmental feature, that is, when accurate navigation landmarks cannot be obtained, by sending a supplement request to the drone, according to the target residual feature with visual information loss and the first inspection parameter, the second drone provides more detailed and clearer second environmental parameters and second inspection parameters, expanding the machine vision of the first drone, enabling the first drone to output more accurate three-dimensional modeling and navigation positioning from multiple directions. At the same time, in the above process, through joint navigation, the hardware requirements for the first drone are reduced, and part of the computing tasks are concentrated on the second drone for processing, enabling the first drone to increase the load capacity and endurance.

[0050] To more clearly introduce the technical solution of the present application, the technical solution of the present application will also be introduced through specific embodiments below. It should be noted that the specific embodiment is used to expand the description of the technical solution of the present application, rather than limiting the present application.

[0051] In some embodiments, the three-dimensional point cloud data is input into a pre-set feature extraction network to obtain the first environmental feature. The specific steps include: Based on the pre-set feature extraction network, perform downsampling on the three-dimensional point cloud data to obtain an initial point cloud set. Construct an octree structure according to the initial point cloud set, perform feature encoding on each leaf node to obtain multi-scale point cloud features. Perform global pooling operation on the multi-scale point cloud features to obtain a global feature vector. Perform feature enhancement on each point in the initial point cloud set according to the global feature vector to obtain enhanced point cloud features. Perform edge convolution operation on the enhanced point cloud features to extract local geometric information and obtain a local feature matrix. Concatenate the local feature matrix and the global feature vector, and perform feature fusion through a multi-layer perceptron to obtain the first environmental feature.

[0052] Exemplarily, the pre-set feature extraction network performs downsampling on the point cloud data to obtain an initial point cloud set. An octree structure is constructed and feature encoding is performed on the leaf nodes to obtain multi-scale point cloud features. Global pooling obtains a global feature vector, which is used to perform feature enhancement on each point in the initial point cloud set. Edge convolution operation extracts local geometric information to obtain a local feature matrix. The local feature matrix and the global feature vector are concatenated, and feature fusion is performed through a multi-layer perceptron to obtain the final environmental feature. This method combines the global information and local geometric features of the point cloud, and gradually extracts and fuses features through multiple processing steps, and can obtain a richer and more effective environmental representation. The technical effect is to make full use of the spatial structure information of the point cloud data, extract discriminative environmental features, and provide strong support for downstream applications such as scene understanding and object detection.

[0053] In some embodiments, determining the target residual feature according to the first environmental feature, the specific steps include: S201-S206.

[0054] S201. Perform a downsampling operation on the first environmental feature to obtain a feature map with a reduced size.

[0055] Exemplarily, methods such as max pooling or average pooling are used to reduce the size of the first environmental feature to 1 / 2 or 1 / 4 of the original. For example, a 2x2 max pooling kernel with a stride of 2 can be used to perform downsampling on the feature map. This can reduce the computational amount and memory occupancy while retaining the main features.

[0056] S202. Construct a multi-layer spatial grid according to the feature map with a reduced size, perform feature reprojection on each layer of the grid, and obtain multi-angle spatial features.

[0057] Exemplarily, 3-5 layers of spatial grids can be constructed, and the resolution of each layer of the grid gradually increases. For each layer of the grid, methods such as bilinear interpolation are used to reproject the features onto the grid. In this way, spatial feature representations at different scales and perspectives can be obtained.

[0058] S203. Voxelize the multi - angle spatial features, extract spatial information through a 3D convolutional network, and obtain a preliminary spatial estimation map.

[0059] A 3 - 5 - layer 3D convolutional network can be used, with a convolutional kernel size of 3x3x3 and a stride of 1. Finally, a fully - connected layer is used to integrate the features to obtain a preliminary spatial estimation result.

[0060] S204. Magnify and transform the original first - environment feature according to the preliminary spatial estimation map to obtain an extended spatial feature.

[0061] Exemplarily, use a transposed convolution or a magnification transformation method to magnify the preliminary spatial estimation map to the same size as the original feature. 2 - 3 layers of transposed convolution can be used, and each layer magnifies the feature map size by 2 times, finally obtaining the extended spatial feature.

[0062] S205. Perform a residual connection between the extended spatial feature and the first - environment feature, and perform feature integration through a multi - layer perceptron to obtain a comprehensive feature.

[0063] Exemplarily, use element - wise addition for the residual connection, and then use 3 - 4 layers of fully - connected layers for feature fusion. The number of neurons in each fully - connected layer can be reduced layer by layer, such as 1024 - 512 - 256 - 128, and finally obtain a comprehensive feature representation.

[0064] S206. Perform boundary recognition and block division on the comprehensive feature, extract the core - region feature, and obtain the target residual feature.

[0065] Exemplarily, use an edge - detection algorithm (such as Canny or Sobel) for boundary recognition, and then use a region - growing or watershed algorithm for block division. Aggregate the features of the divided blocks, extract the core feature of each block. Finally, combine these core features to obtain the final target residual feature.

[0066] In some embodiments, input the first inspection parameter, the second inspection parameter, the first - environment feature, and the second - environment feature into a pre - set visual fusion network to obtain a 3D environmental modeling of the surrounding environment. The specific steps include: S301 - S309.

[0067] S301. Perform a multi - scale pyramid transformation on the first - environment feature and the second - environment feature, and generate multi - level feature maps with different resolutions through continuous convolution and down - sampling operations.

[0068] S302. Calculate a spatial registration matrix according to the first inspection parameter and the second inspection parameter, and perform a spatial transformation and alignment on the multi - level feature maps according to the spatial registration matrix to obtain corrected multi - level feature maps.

[0069] Exemplarily, according to the first inspection parameter, the second inspection parameter, and a preset spatial transformation algorithm, such as the trilateration method, a spatial registration matrix T is generated. The multi-level feature map of the first perspective is spatially transformed through the spatial registration matrix T. Similarly, the multi-level feature map of the second perspective is subjected to the same spatial transformation to obtain the corrected multi-level feature map. In this way, the feature misalignment caused by different perspectives is eliminated.

[0070] S303. Based on the corrected multi-level feature map, use the adaptive feature extraction method and the preset perspective direction to extract texture information, and perform variance-based feature aggregation to obtain the fused enhanced point cloud.

[0071] Exemplarily, texture information is extracted in multiple preset perspective directions through the adaptive feature extraction method to obtain texture features. The multiple preset perspective directions include: the main perspective direction, the left perspective direction, and the right perspective direction. Feature aggregation based on variance is performed on the texture features. Feature aggregation is performed based on variance, thereby retaining the main information of the data in dimensionality reduction or feature extraction and improving the performance of the model.

[0072] S304. Combine the fused enhanced point cloud with the normalized coordinates to construct comprehensive feature data and form a high-dimensional feature point cloud model.

[0073] S305. Uniformly sample the high-dimensional feature point cloud model along the target perspective direction to generate multiple groups of spatial offset candidate points, where the target perspective direction is one of the preset perspective directions.

[0074] Exemplarily, when the target perspective direction is the main perspective direction, the formula for generating multiple groups of spatial candidate offset points includes:

[0075] V t =R (θ,φ) ·V0;

[0076] S i =P i +Δd i ;

[0077] Δd=α·V t +β·n i

[0078]

[0079] where V t is the vector of the target perspective direction. R (θ，φ) is the rotation matrix used to rotate the initial perspective vector V0 to the target perspective direction, where θ and φ are the azimuth angle and elevation angle respectively. V0 is the vector of the initial perspective direction. Si is the position of the i-th sampling point. P i为 is the position of the ii-th original point cloud point. Δdi is the spatial offset of the i-th point. α is a scaling factor used to adjust the offset in the target viewing direction. β is a scaling factor used to adjust the offset in the normal vector direction. n i is the normal vector of the i-th point. C is the set of candidate points. N: the total number of candidate points.

[0080] S306. Establish a dynamic adjacency graph using multiple sets of spatial offset candidate points, apply edge convolution operations to extract local geometric information, and obtain a multi-level local feature set.

[0081] Exemplarily, a dynamic adjacency graph is established using multiple sets of spatial offset candidate points generated in the previous step. A dynamic adjacency graph is a graph structure where each node represents a spatial offset candidate point and the edges represent the proximity relationships between candidate points. Then, edge convolution operations are applied to extract local geometric information. Edge convolution is a special convolution operation that can extract features from the local neighborhoods in point cloud data. In this way, the local geometric relationships between spatial offset candidate points can be captured, resulting in a multi-scale local feature set. The multi-scale local feature set not only contains the geometric information of each candidate point but also the geometric structure information within its neighborhood, providing rich input features for subsequent feature fusion and optimization.

[0082] S307. Input the multi-level local feature set into a deep neural network, calculate the weights of each candidate point through probability distribution, and perform weighted fusion to obtain the point cloud flow vector.

[0083] Exemplarily, the multi-scale local feature set obtained in the previous step is input into a deep neural network. The deep neural network will further process these features and calculate the weights of each candidate point through probability distribution. The weights are calculated based on the similarity and importance between features and can reflect the contribution of each candidate point in the overall feature fusion. Next, the candidate points are weighted and fused according to these weights, and finally, the point cloud flow vector is obtained. The point cloud flow vector represents the moving direction and distance of each point in three-dimensional space, and these vectors will be used for subsequent point cloud optimization and update. In this way, the spatial changes and dynamic characteristics of the point cloud data can be captured more accurately.

[0084] S308. Iteratively optimize the high-dimensional feature point cloud model according to the point cloud flow vector to obtain the optimized point cloud model.

[0085] Exemplarily, based on the point cloud flow vectors obtained in the previous step, the high-dimensional feature point cloud model is iteratively optimized. The iterative optimization process includes updating the three-dimensional spatial coordinates of each point multiple times, making it gradually tend to the real position. Guided by the point cloud flow vectors, the position of the point can be adjusted in each iteration, thereby reducing errors and improving the accuracy of the point cloud model. After each iteration, the point cloud flow vectors are recalculated and the optimization continues until convergence to a stable state. Finally, the obtained high-dimensional feature point cloud model will have higher spatial accuracy and consistency, providing reliable basic data for subsequent environmental modeling and navigation.

[0086] S309. Based on the first inspection parameter and the second inspection parameter, perform global consistency verification and adaptive correction on the optimized point cloud model to obtain a three-dimensional environmental modeling of the surrounding environment.

[0087] Exemplarily, based on the first inspection parameter and the second inspection parameter, perform global consistency verification and adaptive correction on the optimized point cloud model. Global consistency verification refers to checking whether the point cloud model conforms to the expected geometric and topological structures as a whole, ensuring the consistency of the model in the global range. Adaptive correction is to adjust and optimize the problem areas in the model according to the verification results to eliminate errors and inconsistencies. Through these operations, it can be ensured that the finally obtained three-dimensional environmental modeling has high accuracy and high consistency. The finally obtained three-dimensional environmental modeling can not only accurately reflect the geometric structure of the surrounding environment, but also provide reliable reference data for the navigation and positioning of the unmanned aerial vehicle.

[0088] This technical solution realizes high-precision three-dimensional positioning and environmental modeling of an unmanned aerial vehicle based on a low-earth orbit satellite network through multi-level image processing, multi-scale feature extraction and fusion, dynamic adjacency graph construction, and deep neural network optimization, improving the accuracy and reliability of positioning and navigation, and providing strong support for the autonomous navigation and environmental perception of the unmanned aerial vehicle.

[0089] In some embodiments, determine the target residual feature according to the first environmental feature, and generate a supplement request according to the target residual feature and the first inspection parameter, including: performing adaptive threshold segmentation on the target residual feature to obtain multiple feature regions, and calculating the regional importance weights according to the spatial distribution of the feature regions. Sort the feature regions according to the regional importance weights, select the top N important regions, and perform spatial clustering on these N important regions to obtain a key information cluster. Determine the future attitude information and future direction information according to the first inspection parameter. Generate a supplement request according to the key information cluster, future attitude information, and future direction information.

[0090] Determine the target residual features according to the first environmental feature to ensure timely capture of environmental change information and improve the perception ability. Process the target residual features through adaptive threshold segmentation to identify multiple feature regions. Adaptive threshold segmentation can dynamically adjust the segmentation threshold according to different environmental conditions, accurately identify the feature regions, and improve the accuracy and robustness of the identification.

[0091] After identifying the feature regions, calculate the spatial distribution of these regions, determine their importance weights, effectively evaluate the relative importance of each feature region in the current environment, and provide a basis for subsequent steps. Sort the feature regions and select the top N important regions to focus on the most critical regions, avoid resource waste, and improve the processing efficiency.

[0092] Perform spatial clustering on the selected top N important regions to generate key information clusters. By grouping similar feature regions together, simplify the information processing process, reduce the interference of redundant information, and improve the efficiency and accuracy of key information extraction. Determine the future attitude information and future direction information according to the first inspection parameter, make preparations in advance, and enhance the predictability and response speed to future environmental changes.

[0093] Combine the key information clusters, future attitude information, and future direction information to generate a supplementary request, integrate the key information extracted and processed in each step, and provide specific guidance for subsequent operations. Through this series of steps, achieve a rapid response and intelligent processing to environmental changes, and significantly improve the adaptability, accuracy, and processing efficiency.

[0094] In some embodiments, the second inspection parameter is determined by the second unmanned aerial vehicle according to the future attitude information and future direction information, and the second environmental feature is obtained by the second unmanned aerial vehicle through shooting and feature extraction according to the second inspection parameter.

[0095] In some embodiments, determine the navigation landmarks according to the three-dimensional environmental modeling, and perform navigation positioning according to the navigation landmarks, including: extract feature points from the three-dimensional environmental modeling to obtain a set of candidate feature points. Screen the set of candidate feature points according to the preset landmark screening rules to obtain a set of navigation landmarks. Use the low-earth orbit satellite network to obtain real-time position information, and match the real-time position information with the set of navigation landmarks. Calculate the relative position relationship between the current position and the navigation landmarks according to the matching result. Perform attitude estimation and path planning based on the relative position relationship to achieve navigation positioning.

[0096] Please refer to Figure 3 , Figure 3 which is a schematic block diagram of an unmanned aerial vehicle positioning device based on a low-earth orbit satellite network provided by an embodiment of the present application. The unmanned aerial vehicle positioning device 200 based on a low-earth orbit satellite network is used to execute the aforementioned unmanned aerial vehicle positioning method based on a low-earth orbit satellite network. Among them, the unmanned aerial vehicle positioning device 200 based on a low-earth orbit satellite network can be configured in a server.

[0097] Among them, the server can be an independent server, a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0098] Such as Figure 3 As shown, the UAV positioning device 300 based on the low-earth orbit satellite network includes: a communication module 301, a data acquisition module 302, a feature extraction module 303, a request generation module 304, a vision fusion module 305, and a navigation and positioning module 306.

[0099] The communication module 301 is used to publish the first inspection parameters through the distributed information channel.

[0100] The data acquisition module 302 is used to obtain three-dimensional point cloud data of the surrounding environment through a variety of sensors.

[0101] The feature extraction module 303 is used to input the three-dimensional point cloud data into a pre-set feature extraction network to obtain the first environmental features.

[0102] The request generation module 304 is used to determine the target residual features according to the first environmental features, generate a supplementary request according to the target residual features and the first inspection parameters, and send the supplementary request to the distributed information channel, so that the second UAV can reply with a plurality of second environmental features and corresponding second inspection parameters.

[0103] The vision fusion module 305 is used to input the first inspection parameters, the second inspection parameters, the first environmental features, and the second environmental features into a pre-set vision fusion network to obtain a three-dimensional environmental modeling of the surrounding environment.

[0104] The navigation and positioning module 306 is used to determine navigation landmarks according to the three-dimensional environmental modeling and perform navigation and positioning according to the navigation landmarks.

[0105] An embodiment of the present application provides an electronic device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the UAV positioning method based on the low-earth orbit satellite network as described in any one of the embodiments of the present application when executing the computer program.

[0106] An embodiment of the present application provides a computer-readable storage medium storing a computer program, which when executed by a processor causes the processor to implement the method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network as described in any one of the embodiments of the present application.

[0107] As described above, the foregoing is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network, characterized in that, Applied to the first unmanned aerial vehicle (UAV), the first UAV and the second UAV establish a distributed information channel through a low-earth orbit satellite network. The method includes: Publish the first inspection parameters through the distributed information channel; Obtain three-dimensional point cloud data of the surrounding environment through multiple sensors; Input the three-dimensional point cloud data into a preset feature extraction network to obtain the first environmental feature; Determine the target residual feature according to the first environmental feature, generate a supplement request according to the target residual feature and the first inspection parameters, and send the supplement request to the distributed information channel, so that the second UAV returns multiple second environmental features and corresponding second inspection parameters; Input the first inspection parameters, the second inspection parameters, the first environmental feature, and the second environmental feature into a preset visual fusion network to obtain a three-dimensional environmental model of the surrounding environment; Determine navigation landmarks according to the three-dimensional environmental model and perform navigation positioning according to the navigation landmarks.

2. The method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network according to claim 1, wherein The inputting the three-dimensional point cloud data into a preset feature extraction network to obtain the first environmental feature includes: Based on the preset feature extraction network, perform downsampling processing on the three-dimensional point cloud data to obtain an initial point cloud set; Construct an octree structure according to the initial point cloud set, perform feature encoding on each leaf node to obtain multi-scale point cloud features; Perform global pooling operation on the multi-scale point cloud features to obtain a global feature vector; Enhance the features of each point in the initial point cloud set according to the global feature vector to obtain enhanced point cloud features; Perform edge convolution operation on the enhanced point cloud features to extract local geometric information and obtain a local feature matrix; Concatenate the local feature matrix with the global feature vector, perform feature fusion through a multi-layer perceptron to obtain the first environmental feature.

3. The UAV positioning method based on the low-earth orbit satellite network according to claim 1, wherein The determining the target residual feature according to the first environmental feature includes: Perform downsampling operation on the first environmental feature to obtain a feature map with reduced size; Construct a multi-layer spatial grid according to the feature map with reduced size, perform feature reprojection on each layer of the grid to obtain multi-angle spatial features; Perform voxelization processing on the multi-angle spatial features, extract spatial information through a three-dimensional convolutional network to obtain a preliminary spatial estimation map; Perform magnification transformation on the original first environmental feature according to the preliminary spatial estimation map to obtain an extended spatial feature; Perform residual connection between the extended spatial feature and the first environmental feature, perform feature integration through a multi-layer perceptron to obtain a comprehensive feature; Perform boundary recognition and block division on the comprehensive feature, extract core region features to obtain the target residual feature.

4. The UAV positioning method based on the low-earth orbit satellite network according to claim 1, characterized in that, The inputting the first inspection parameters, the second inspection parameters, the first environmental feature, and the second environmental feature into a preset visual fusion network to obtain a three-dimensional environmental model of the surrounding environment includes: Perform multi-scale pyramid transformation on the first environmental feature and the second environmental feature, and generate multi-level feature maps with different resolutions through continuous convolution and downsampling operations; Calculate a spatial registration matrix based on the first inspection parameter and the second inspection parameter, and perform spatial transformation and alignment on the multi-level feature maps according to the spatial registration matrix to obtain corrected multi-level feature maps; Based on the corrected multi-level feature maps, use an adaptive feature extraction method and a preset viewing direction to extract texture information, and perform variance-based feature aggregation to obtain a fused enhanced point cloud; Combine the fused enhanced point cloud with the normalized coordinates to construct comprehensive feature data and form a high-dimensional feature point cloud model; Uniformly sample the high-dimensional feature point cloud model along the target viewing direction to generate multiple sets of spatial offset candidate points, where the target viewing direction is one of the preset viewing directions; Use the multiple sets of spatial offset candidate points to establish a dynamic adjacency graph, and apply edge convolution operations to extract local geometric information to obtain a multi-level local feature set; Input the multi-level local feature set into a deep neural network, calculate the weights of each candidate point through probability distribution, and perform weighted fusion to obtain a point cloud flow vector; Iteratively optimize the high-dimensional feature point cloud model according to the point cloud flow vector to obtain an optimized point cloud model; Based on the first inspection parameter and the second inspection parameter, perform global consistency verification and adaptive correction on the optimized point cloud model to obtain a three-dimensional environmental model of the surrounding environment.

5. The method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network according to claim 1, wherein, The determination of the target residual feature according to the first environmental feature and the generation of a supplementary request according to the target residual feature and the first inspection parameter include: Perform adaptive threshold segmentation on the target residual feature to obtain multiple feature regions, and calculate the regional importance weights according to the spatial distribution of the feature regions; Sort the feature regions according to the regional importance weights, select the top N important regions, and perform spatial clustering on these N important regions to obtain a key information cluster; Determine future pose information and future direction information according to the first inspection parameter; Generate the supplementary request according to the key information cluster, the future pose information, and the future direction information.

6. The method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network according to claim 5, wherein The second inspection parameter is determined by the second unmanned aerial vehicle according to the future pose information and the future direction information, and the second environmental feature is obtained by the second unmanned aerial vehicle through shooting and feature extraction according to the second inspection parameter.

7. The method for positioning an unmanned aerial vehicle based on a low-earth orbit satellite network according to claim 1, wherein The determination of the navigation landmark according to the three-dimensional environmental model and the navigation positioning according to the navigation landmark include: Extract feature points from the three-dimensional environmental model to obtain a set of candidate feature points; Screen the set of candidate feature points according to a preset landmark screening rule to obtain a set of navigation landmarks; Obtain real-time position information using a low-earth orbit satellite network and match the real-time position information with the set of navigation landmarks; Calculate the relative position relationship between the current position and the navigation landmark according to the matching result; Perform attitude estimation and path planning based on the relative position relationship to achieve navigation positioning.

Images