Wireless channel three-dimensional electromagnetic environment reconstruction method based on multi-source data

Abstract

The invention discloses a wireless channel three-dimensional electromagnetic environment reconstruction method based on multi-source data, and relates to the technical field of wireless channel research, and the method specifically comprises the following steps: 1) extracting building information by using an electronic map, and carrying out first-stage building reconstruction; 2) utilizing a satellite map to carry out ground object environment reconstruction of a second stage, introducing an image semantic segmentation technology to carry out environment perception on the satellite map, designing a segmentation mask optimization processing method, making up for the precision problem of semantic segmentation, extracting key ground object environment information, and carrying out environment three-dimensional surface reconstruction according to a segmentation result; and 3) performing electromagnetic parameter matching and ray tracing simulation in a third stage, and constructing an offline channel map of the target area. According to the method, three-dimensional electromagnetic environment reconstruction of a wireless communication scene is realized, electromagnetic environment information richer than that of a traditional building model is obtained, construction of an off-line channel map is completed, and the technical requirements of a 6G digital twinning on-line channel model can be efficiently met.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless channel research, and in particular to a method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data. Background Art

[0002] With the advancement of wireless communication technology, the sixth generation of wireless communication technology (6G) is expected to achieve "full coverage, full application, full digital, full spectrum, full perception and strong security". Digital Twin Online Channel Models (DTOCM) can dynamically and accurately reflect the channel characteristics of the real communication environment and improve the overall performance of the 6G network. At present, the construction of DTOCM for known scenarios includes four steps: identifying communication scenarios and initialization, reconstructing offline channel maps, sensing the physical environment online and dynamically updating the digital twin channel model, and the application of the digital twin channel model. Among them, the effective reconstruction of the three-dimensional (3D) twin electromagnetic environment is the key cornerstone of the DTOCM construction, which supports the accuracy of the entire digital twin model and lays a solid foundation for the subsequent real-time perception and efficient operation of DTOCM.

[0003] Currently, the environment construction of DTOCM relies on manual creation using software such as Blender, but it is time-consuming and laborious. In addition, open source databases such as OpenStreetMap (OSM) are used to create 3D building models. However, these databases may not be updated enough and lack information in many areas. Another method is to use drone mapping for scene reconstruction, however, this method is also costly and laborious.

[0004] In response to the above challenges, researchers have applied artificial intelligence technology to scene reconstruction. Many deep learning models have been used for 3D reconstruction. However, these studies mainly focus on the extraction and reconstruction of building information, without considering more environmental factors such as vegetation, roads and water bodies, which can significantly affect wireless channel propagation. At present, the construction process of DTOCM has the problem of complex and laborious construction of communication twin scenes. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data. The present invention is a fast and efficient three-dimensional electromagnetic environment reconstruction for DTOCM. The present invention realizes the initialization of the communication scene and the construction of the offline channel map, thereby improving the accuracy of the entire digital twin model.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] A method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data is proposed according to the present invention, comprising:

[0008] Obtain electronic maps and satellite maps of the target area for three-dimensional electromagnetic environment reconstruction;

[0009] Extracting coordinate information and height information of buildings from an electronic map of the target area;

[0010] According to the coordinate information and height information of the building, the three-dimensional reconstruction of the building is performed using the triangular mesh to obtain a three-dimensional building model;

[0011] Perform segmentation preprocessing on the satellite map of the target area to obtain a preprocessed satellite map;

[0012] The environment perception module is composed of a semantic segmentation network and a fully connected conditional random field. The environment perception module is used to perform environment perception classification on the preprocessed satellite map to obtain the classification results. The classification results are then spliced ​​to obtain the complete environment perception classification results of the target area.

[0013] Performing post-optimization processing on the environmental perception classification result to obtain the environmental perception classification result after optimization processing;

[0014] Extracting the coordinate information of the ground object from the optimized environmental perception classification result, converting the coordinate information of the ground object into the actual coordinate, and allocating the height information according to the optimized environmental perception classification result;

[0015] According to the converted actual coordinates and height information of the ground object environment, the ground object environment is three-dimensionally reconstructed using a triangular mesh to obtain a three-dimensional ground object environment model;

[0016] The obtained three-dimensional building model and the three-dimensional ground environment model are matched and integrated to obtain a three-dimensional twin environment model;

[0017] According to the results of the environmental perception classification after the final optimization processing, electromagnetic parameters are assigned to the three-dimensional twin environment model, and the channel characteristics are simulated in combination with ray tracing technology.

[0018] As a further optimization scheme of the method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data described in the present invention, the format, accuracy, and obtainable environmental information of the electronic map and satellite map used are first determined.

[0019] As a further optimization scheme of the method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data described in the present invention, the electronic map format is plantet, the satellite map is an image in tiff format, the accuracy of the electronic map is 5 meters, and the accuracy of the satellite map is 0.3 meters.

[0020] As a further optimization scheme of the method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data described in the present invention, a three-dimensional reconstruction of a building is performed using a triangular mesh according to the coordinate information and height information of the building to obtain a three-dimensional building model; the following steps are included:

[0021] Step 3.1, when generating the side of the building, the longitude and latitude information and height data of each building vertex are used to create rectangular sides; these rectangular sides are divided into two triangular meshes to form each side of the building, thereby obtaining the side model of the building;

[0022] Step 3.2, generating a roof model of a building: before generating the roof model of the building, determine the convexity of the roof polygon; if the roof polygon is a convex polygon, connect the vertices of the roof polygon of the building to the center point of the polygon, so as to obtain the roof model of the building model; if the roof polygon is concave, divide the roof polygon into multiple triangles or convex polygons, and then connect the vertices of each polygon to their respective center points, so as to obtain the roof model of the building model;

[0023] Step 3.3, combine the obtained side model and roof model of the building to create a complete three-dimensional building model.

[0024] As a further optimization scheme of the method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data described in the present invention, an environment perception module is formed by a semantic segmentation network and a fully connected conditional random field, and the environment perception module is used to perform environment perception classification on the preprocessed satellite map to obtain a classification result, and the classification result is spliced ​​to obtain a complete environment perception classification result of the target area; specifically, the following steps are included:

[0025] Step 5.1, determine the dataset of the environment perception module and select the LoveDa dataset as the semantic segmentation dataset;

[0026] Step 5.2: Perform environmental perception on the preprocessed satellite image to obtain preliminary classification results of environmental perception;

[0027] Step 5.3: Use the fully connected conditional random field to optimize the preliminary classification results of environmental perception, and concatenate the classification results to obtain the complete environmental perception classification results of the target area.

[0028] As a further optimization scheme of the method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data described in the present invention, the environmental perception classification result is optimized and processed to obtain the environmental perception classification result after optimization processing; including the following:

[0029] Step 6.1, optimizing the environmental perception classification results to obtain the classification results of each local environment;

[0030] Step 6.2: After performing mask optimization processing on the classification results of each object environment, obtain the environmental perception classification results that conform to the real environment.

[0031] As a further optimization scheme of the method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data described in the present invention, electromagnetic parameters are allocated to the three-dimensional twin environment model according to the result of the environmental perception classification after the final optimization processing, and the channel characteristics are simulated in combination with ray tracing technology; specifically, it includes the following:

[0032] Step 10.1, import the obtained 3D twin environment model into the ray tracing simulation software Wireless InSite;

[0033] Step 10.2, in Wireless InSite, according to the optimized environmental perception classification results, electromagnetic parameters are assigned to the environmental information of buildings, roads, trees, and water bodies in the 3D twin environment model;

[0034] Step 10.3: Set up the antenna and transceiver in Wireless InSite, and perform ray tracing simulation.

[0035] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0036] (1) The present invention comprehensively utilizes electronic maps and satellite map data, combined with deep learning and image processing, to reconstruct a three-dimensional twin electromagnetic environment for DTOCM, thereby accelerating the construction of offline channel maps;

[0037] (2) First, accurate building information is extracted from the electronic map to reconstruct the three-dimensional model. Secondly, the semantic segmentation network is applied to the satellite map to perform environmental perception. The fully connected conditional random field and mask optimization method are used to enhance the results of environmental perception. After extracting the ground object information, the three-dimensional environment is reconstructed, which provides more environmental information for the twin scene and can greatly improve the accuracy of channel characteristic simulation. Finally, based on the three-dimensional reconstruction results of the previous two stages, the model is multi-sourced, electromagnetic parameters are assigned to the three-dimensional twin environment model, and channel characteristic simulation is performed. In summary, the present invention is an effective three-dimensional electromagnetic environment reconstruction method, which provides a prerequisite for DTOCM to achieve efficient offline channel map construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A model block diagram of a method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data provided in Embodiment 1 of the present invention;

[0039] Figure 2 This is a schematic diagram of part of the electronic map in Example 1 of the present invention;

[0040] Figure 3 This is a schematic diagram of a portion of a satellite map in Example 1 of the present invention;

[0041] Figure 4 This is a schematic diagram of a three-dimensional model of a portion of a building in Example 1 of the present invention;

[0042] Figure 5 This is a schematic diagram of an environment perception module in Example 1 of the present invention;

[0043] Figure 6 This is a schematic diagram of the road mask edge connection result in Example 1 of the present invention;

[0044] Figure 7 This is a schematic diagram of the reconstruction result of the three-dimensional electromagnetic environment in Example 1 of the present invention;

[0045] Figure 8 Schematic diagram of path loss comparison in Example 1 of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Embodiment 1:

[0048] See also Figure 1 This embodiment provides a method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data, which mainly includes the following steps:

[0049] Step 1: Obtain an electronic map and a satellite map of the target area for three-dimensional electromagnetic environment reconstruction, and determine the format and accuracy of the electronic map and satellite map to be used, as well as the environmental information that can be obtained.

[0050] Specifically, in this embodiment, the electronic map format is determined to be "plantet", the satellite map is determined to be an image in "tiff" format, the electronic map accuracy is 5 meters, and the satellite map accuracy is 0.3 meters. The key layers of the electronic map are analyzed, including but not limited to the Buildings layer, the Height layer, and the Vector layer, to obtain accurate building information in the electronic map. The ground objects in the satellite map are buildings, roads, water bodies, wastelands, woods, and cultivated land. These ground objects have a significant impact on the wireless channel. The schematic diagram of the electronic map and satellite map of the target area is shown in the figure below. Figure 2 , Figure 3 shown.

[0051] Step 2: Extract the coordinate information and height information of the building from the electronic map of the target area and save it in a txt file.

[0052] Specifically, in this embodiment, the Buildings layer, Height layer and Vector layer in the electronic map provide the coordinate information and height information of the buildings. The coordinate information and height information of the buildings in the target area are extracted and saved in txt files respectively. The coordinate files are stored in sets for the next step of reconstructing the three-dimensional surface of the buildings.

[0053] Step 3: According to the txt file of the building, read the vertex coordinates and height data of the building, and use the triangular mesh to reconstruct the three-dimensional surface of the building.

[0054] Specifically, in this embodiment, the vertex coordinates and height data of the building in the txt file are read, and the side of the building is generated by looping the vertex coordinates of the building to form each vertex of the polygon, and each side rectangle is constructed with two triangle meshes. The roof of the building will generate a center point based on the polygon vertices, and connect the vertices and the center point to construct a corresponding triangle mesh. The generated three-dimensional model of the building includes the side and the roof, and is saved as an obj file.

[0055] Among them, in the generation of the roof, it is necessary to determine whether the polygon is a convex polygon. If it is a concave polygon, the center point generated by the vertex will appear outside the polygon, etc., resulting in errors in the generation of the roof. Therefore, it is necessary to determine the convexity of the polygon and split it into several convex triangles and polygons, including the following steps:

[0056] 1) Vertex check: Check the geometric characteristics of the polygon to ensure that its boundary is closed (the starting point and the end point are consistent) and there is no self-intersection, and sort the vertices;

[0057] 2) Ear cutting method initialization: Identify the polygonal ear shape area (convex vertices) to prepare for the ear cutting method segmentation process;

[0058] 3) Decomposition execution: cut off the lug areas one by one, generate triangles, remove the processed vertices, and update the vertex list of the remaining part of the polygon;

[0059] 4) Boundary update: After each subdivision, recheck the remaining polygon boundaries and repeat the ear cutting operation until the polygon is completely subdivided into triangles;

[0060] 5) Result output: After the segmentation is completed, the coordinates or vertex indices of all the triangles obtained by segmentation are obtained.

[0061] For subsequent geometric reconstruction processing.

[0062] Finally, height information is assigned to the segmented roof polygons, and a three-dimensional model of the building is further generated. Figure 4 The 3D models of some buildings generated in the example are shown in Figure .

[0063] Step 4: Segment and preprocess the satellite map of the target area.

[0064] The specific steps are as follows: the satellite map is segmented into images with a size of 1024×1024 pixels.

[0065] Step 5: The semantic segmentation network and the fully connected conditional random field form an environment perception module, which is used to perform environment perception classification on the preprocessed satellite map. The results of the environment perception module are spliced ​​to obtain the complete environment perception classification results of the target area. The schematic diagram of the environment perception module is shown in the figure. Figure 5 As shown. In the example, the semantic segmentation network used is the UNetFormer network, which is trained using the training set in the dataset. UNetFormer consists of ResNet18 as an encoder and a Transformer-based decoder. The encoder includes ResBlocks arranged in four stages, each of which downsamples the feature map. The decoder consists of three global local transformer blocks (GLTB), three weighted sum modules (WS), and a feature refinement head (FRH). GLTB introduces a global-local attention mechanism that uses a dual-branch structure to capture global and local context information. The local branch extracts local information using parallel convolutional layers with kernel sizes of 3 and 1, and applies batch normalization before the final summation operation. The global branch captures global context through a window-based multi-head self-attention mechanism. Each stage of the feature map generated by the encoder is connected to the corresponding feature map in the decoder through a 1×1 convolution. WS is responsible for feature aggregation, thereby learning more general multi-source features. The formula of WS can be expressed as:

[0066] BF=α·RBF+(1-α)·GLBF

[0067] In the formula, BF represents multi-source features, RBF represents features generated by Resblocks, GLBF represents features generated by GLTB, and α is a weight coefficient.

[0068] The fully connected conditional random field improves the overall segmentation quality by smoothing the segmentation results, reducing noise and misclassification. The goal of the fully connected conditional random field is to maximize the conditional probability P(y|x) of the label configuration y:

[0069]

[0070] In the formula, y={y1,y2,…,y N}, x={x1,x2,…,x N}, x represents the observed feature, y i represents the label of the i-th pixel, x represents the feature vector of the i-th pixel, i∈[1,N], N is the total number of pixels in the image, Z(x)=∑ y exp(-E(y|x)) is the partition function that normalizes the probability distribution. E(y|x) is the energy function, which consists of unary and pairwise potential functions and can be expressed as:

[0071]

[0072] In the formula, ψ u (y i |x) is a unary potential used to calculate the label y based on the observed feature x. i The cost assigned to pixel i, ψ p (y i ,y j |x) is a binary potential that measures the interaction between pixel pair i and j, making spatially similar pixels share the same label.

[0073] The initial prediction results of the neural network are processed through a fully connected conditional random field to make them more consistent with the actual physical boundaries.

[0074] In the environment perception module, the dataset is the LoveDa dataset. In the trained environment perception module, the segmentation accuracy of the environment perception module is evaluated by the pixel accuracy, where the pixel accuracy is:

[0075]

[0076] In the formula, TP represents the number of pixels correctly predicted as positive, and FP represents the number of pixels incorrectly predicted as positive. The higher the pixel accuracy, the higher the accuracy of model perception.

[0077] The specific steps are as follows:

[0078] Step 501, data set training. Prepare a training data set, which is divided into a training set, a validation set, and a test set, and train the UNetFormer model;

[0079] Step 502: Perform preliminary classification through the UNetFormer module. In the trained environment perception module, perform environmental perception on the segmented satellite images to obtain preliminary classification results of environmental perception;

[0080] Step 503: Fully connected conditional random field optimization. Calculate ψ according to the classification result graph u (y i |x) and ψ p (y i ,y j |x), minimize the energy function E(y|x) to maximize P(y|x), perform multiple inferences, iteratively optimize label assignment, find the most likely category for each pixel, and save the final processed image. The number of inferences can be adjusted appropriately according to the accuracy requirements.

[0081] Step 504, evaluation and tuning. Finally, the performance of the module needs to be evaluated, usually using cross-validation or other evaluation indicators (such as accuracy, precision, recall, etc.). This example uses pixel accuracy for evaluation. The simulation results are shown in Table 1. The segmentation accuracy of roads, buildings, water bodies, bare land, woodlands, and cultivated land are 68%, 84%, 84%, 66%, 61%, and 92%, respectively, and the overall accuracy is about 70%. Satisfactory results were achieved through the environmental perception module. After the introduction of the fully connected conditional random field, the boundary segmentation of various ground environments has been significantly improved.

[0082] Table 1 Simulation results

[0083] category the way building Water Bare Land forest arable land Overall accuracy Accuracy 68％ 84％ 84％ 66％ 61％ 92％ 70％

[0084] Step 6: Perform mask optimization post-processing on the results of environmental perception classification to obtain a perception classification result that is more in line with the actual environment. Since Google Earth satellite images from different regions are used, their quality is different from that of the images in the dataset, and the performance of the segmentation network is affected. In addition, the occlusion caused by vegetation and other geospatial features in the satellite images affects the accurate road segmentation. In this embodiment, a series of image processing steps are designed and implemented to enhance the continuity and integrity of the road boundaries in the perception segmentation output. The specific steps are as follows:

[0085] Step 601: Morphological operation: Apply morphological operation to smooth and expand the edge of the road mask and eliminate the breakage and noise caused by segmentation;

[0086] Step 602: Line segment detection: A line detection algorithm based on Hough transform is used to fit straight lines along the road boundary to ensure their straightness and consistency.

[0087] Step 603: Edge connection. The result of edge connection is shown in the figure below: Figure 6 As shown. When multiple lines are detected or some areas lack edge detection, an edge connection algorithm is used to connect these lines to ensure edge continuity. Specifically, if the current pixel is a road edge (pixel value is 255), the neighborhood pixels around each pixel are determined based on the center size of the neighborhood, and the surrounding neighborhood pixels are checked. If only one neighbor pixel is 255: mark the current pixel as the end point of the straight line and record the only neighbor pixel. Calculate the distance between the current end point and its neighbor pixel. If the distance is less than or equal to the threshold, the two points are connected by linear interpolation. The neighborhood center size and connection threshold can be determined by experiment to determine the optimal size.

[0088] Step 7: Extract coordinate information from the environmental perception results and mask optimization results, convert them into actual coordinates, and assign height information according to the environmental perception classification results. Specifically, the environmental perception results and mask optimization results are both images, each pixel represents a real resolution of 0.3 meters, and the image pixel coordinates are mapped to real coordinates. The precise coordinates of the building in the electronic map are used as anchor points to correct and match the coordinates of the object information. According to the environmental perception classification results, the height value is assigned to the object information, and the coordinate information and height information are saved as txt files.

[0089] Step 8: Based on the coordinate information and height information of the object environment, the object environment is reconstructed in three dimensions using a triangular mesh to obtain a three-dimensional object environment model. Specifically, in this embodiment, similar to the method for generating a three-dimensional model of a building, the vertex coordinates and height data of the object information in the txt file are read, and the corresponding triangular mesh is constructed to generate a three-dimensional model of the object environment and save it as an obj file.

[0090] Step 9: Match and fuse the obtained 3D building model and 3D land object environment model to obtain a 3D twin environment model. Specifically, match and fuse the building model and land object environment model generated in the first two stages to obtain a more comprehensive 3D twin environment model in the target area. In this example, the 3D electromagnetic environment reconstruction result is as follows: Figure 7 shown.

[0091] Step 10: Based on the results of the environmental perception classification, electromagnetic parameters are assigned to the three-dimensional twin environment model, and channel characteristics simulation is performed in combination with ray tracing technology.

[0092] Step 1001, importing the obtained three-dimensional twin environment model into the ray tracing simulation software Wireless InSite;

[0093] Step 1002: In Wireless InSite, electromagnetic parameters are assigned to the environmental information of buildings, roads, trees, and water bodies in the three-dimensional twin environment model according to the optimized environmental perception classification results;

[0094] Step 1003: Perform simulation settings of antennas, transceivers, etc. in Wireless InSite, and perform ray tracing simulation to obtain the required channel characteristics;

[0095] Step 1004: Evaluate the channel characteristics and compare them with the actual channel measurement data to ensure the accuracy of the three-dimensional twin electromagnetic environment and verify the improvement of the channel characteristic simulation by the model. In Example 1, the channel measurement environment is a suburban scene near a building. The transmitter is placed in a fixed position near the building, and the receiver changes with the point. The measurement frequency is 2.6 GHz, with two scenes. The channel characteristic comparison results in this example are shown in Figure 8 ,Compared with the simulation based on the building model alone, the simulation results obtained from the 3D twin electromagnetic environment show good consistency with the channel measurement data. This consistent relationship comes from the rich environmental perception information provided by the 3D twin electromagnetic environment model.

[0096] In summary, the method for reconstructing the three-dimensional electromagnetic environment of wireless channels based on multi-source data established by the present invention effectively extracts environmental information from electronic maps and satellite maps, and ultimately effectively realizes the reconstruction of the wireless communication environment, thereby promoting and accelerating the construction of offline channel maps and improving the overall accuracy of the digital twin channel model.

[0097] The above description is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a technician familiar with the technical field within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for reconstructing the three-dimensional electromagnetic environment of a wireless channel based on multi-source data, characterized in that: include: Obtain electronic maps and satellite maps of the target area for three-dimensional electromagnetic environment reconstruction; Extracting coordinate information and height information of buildings from an electronic map of the target area; According to the coordinate information and height information of the building, the three-dimensional reconstruction of the building is performed using the triangular mesh to obtain a three-dimensional building model; Perform segmentation preprocessing on the satellite map of the target area to obtain a preprocessed satellite map; The environment perception module is composed of a semantic segmentation network and a fully connected conditional random field. The environment perception module is used to perform environment perception classification on the preprocessed satellite map to obtain the classification results. The classification results are then spliced ​​to obtain the complete environment perception classification results of the target area. Performing post-optimization processing on the environmental perception classification result to obtain the environmental perception classification result after optimization processing; Extracting the coordinate information of the ground object from the optimized environmental perception classification result, converting the coordinate information of the ground object into the actual coordinate, and allocating the height information according to the optimized environmental perception classification result; According to the converted actual coordinates and height information of the ground object environment, the ground object environment is three-dimensionally reconstructed using a triangular mesh to obtain a three-dimensional ground object environment model; The obtained three-dimensional building model and the three-dimensional ground environment model are matched and integrated to obtain a three-dimensional twin environment model; According to the results of the environmental perception classification after the final optimization processing, electromagnetic parameters are assigned to the three-dimensional twin environment model, and the channel characteristics are simulated in combination with ray tracing technology.

2. The method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data according to claim 1, characterized in that: First determine the format, accuracy, and environmental information that can be obtained from the electronic maps and satellite maps used.

3. The method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data according to claim 2, characterized in that: The electronic map format is plantet, and the satellite map is a tiff format image. The accuracy of the electronic map is 5 meters, and the accuracy of the satellite map is 0.3 meters.

4. The method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data according to claim 1, characterized in that: According to the coordinate information and height information of the building, a three-dimensional reconstruction of the building is performed using a triangular mesh to obtain a three-dimensional building model; the steps include: Step 3.1, when generating the side of the building, the longitude and latitude information and height data of each building vertex are used to create rectangular sides; these rectangular sides are divided into two triangular meshes to form each side of the building, thereby obtaining the side model of the building; Step 3.2, generating a roof model of a building: before generating the roof model of the building, determine the convexity of the roof polygon; if the roof polygon is a convex polygon, connect the vertices of the roof polygon of the building to the center point of the polygon, so as to obtain the roof model of the building model; if the roof polygon is concave, divide the roof polygon into multiple triangles or convex polygons, and then connect the vertices of each polygon to their respective center points, so as to obtain the roof model of the building model; Step 3.3, combine the obtained side model and roof model of the building to create a complete three-dimensional building model.

5. The method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data according to claim 1, characterized in that: The environment perception module is composed of a semantic segmentation network and a fully connected conditional random field. The environment perception module is used to perform environment perception classification on the preprocessed satellite map to obtain the classification results. The classification results are then spliced ​​to obtain the complete environment perception classification results of the target area. The specific steps include: Step 5.1, determine the dataset of the environment perception module and select the LoveDa dataset as the semantic segmentation dataset; Step 5.2: Perform environmental perception on the preprocessed satellite image to obtain preliminary classification results of environmental perception; Step 5.3: Use the fully connected conditional random field to optimize the preliminary classification results of environmental perception, and concatenate the classification results to obtain the complete environmental perception classification results of the target area.

6. The method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data according to claim 1, characterized in that: Perform post-optimization processing on the environmental perception classification result to obtain the environmental perception classification result after optimization processing; including the following: Step 6.1, optimizing the environmental perception classification results to obtain the classification results of each local environment; Step 6.2: After performing mask optimization processing on the classification results of each object environment, obtain the environmental perception classification results that conform to the real environment.

7. The method for reconstructing a three-dimensional electromagnetic environment of a wireless channel based on multi-source data according to claim 1, characterized in that: According to the results of the final optimized environmental perception classification, electromagnetic parameters are assigned to the 3D twin environment model, and channel characteristics simulation is performed in combination with ray tracing technology; specifically, the following are included: Step 10.1, import the obtained 3D twin environment model into the ray tracing simulation software Wireless InSite; Step 10.2, in Wireless InSite, according to the optimized environmental perception classification results, electromagnetic parameters are assigned to the environmental information of buildings, roads, trees, and water bodies in the 3D twin environment model; Step 10.3: Set up the antenna and transceiver in Wireless InSite, and perform ray tracing simulation.

Images