Multi-band spectrum map construction method, device, electronic device and storage medium

Abstract

The present application proposes a method for constructing a multi-band spectrum map, which includes: obtaining map data of a target area, and rasterizing the map data to obtain a target grid map; randomly selecting multiple acquisition grids in the target grid map, and obtaining the radiation signal strength of each acquisition grid in different frequency bands; using the position information of each acquisition grid and the radiation signal strength of each acquisition grid in different frequency bands as heterogeneous graph node information, and modeling each acquisition grid in each frequency band as a heterogeneous graph node; constructing a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node; obtaining a pre-trained heterogeneous graph neural network, and inputting the spatial domain map, the frequency domain map and the heterogeneous graph node information of each heterogeneous graph node into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area. The embodiment of the present application can construct a high-quality multi-band spectrum map.

Description

Technical Field

[0001] The present application relates to the field of wireless communications, and in particular to a method, device, electronic device and storage medium for constructing a multi-band spectrum map. Background Art

[0002] In recent years, with the diversification of ultra-wideband services and the demand for high-precision sensing, wireless communication technology has developed rapidly and is expected to support the popularization of large-scale intelligent terminals. To this end, there are a series of design and control problems to be solved, such as network planning, resource allocation, dynamic spectrum access, spectrum sharing, etc., which require high-precision spectrum maps to accurately evaluate the radio environment. Spectrum maps can describe power spectrum density (PSD) information in different spaces, times, and frequencies.

[0003] In the related art, most traditional spectrum map reconstruction methods predict the power spectral density in a two-dimensional space of a given frequency band (single band), focusing on the spectrum distribution at different locations. However, many modern wireless applications have recently had an increasing demand for full-band information, so an efficient multi-band spectrum map construction method is an inevitable requirement for the development of modern wireless communications. Summary of the invention

[0004] The embodiments of the present application provide a method, device, electronic device and storage medium for constructing a multi-band spectrum map, which can construct a high-quality multi-band spectrum map.

[0005] To achieve the above object, a first aspect of an embodiment of the present application proposes a method for constructing a multi-band spectrum map, the method comprising:

[0006] Acquire map data of the target area, and perform rasterization processing on the map data to obtain a target raster map;

[0007] Randomly select multiple acquisition grids in the target grid map to obtain the radiation signal intensity of each acquisition grid in different frequency bands;

[0008] The location information of each acquisition grid and the radiation signal strength of each acquisition grid in different frequency bands are used as heterogeneous graph node information, and each acquisition grid in each frequency band is modeled as a heterogeneous graph node;

[0009] Constructing a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node;

[0010] A pre-trained heterogeneous graph neural network is obtained, and the spatial domain graph, the frequency domain graph, and the heterogeneous graph node information of each heterogeneous graph node are input into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area.

[0011] In some embodiments, the method of using the location information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid as heterogeneous graph node information and modeling each acquisition grid in each frequency band as a heterogeneous graph node includes:

[0012] According to the target grid map, each grid in each frequency band is modeled as an original node;

[0013] According to the position information of each acquisition grid, the occluder nodes in the original nodes are removed to obtain a heterogeneous graph node set;

[0014] Based on the heterogeneous graph node set, the position information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid are used as heterogeneous graph node information, and each acquisition grid in each frequency band is modeled as a heterogeneous graph node.

[0015] In some embodiments, constructing the spatial domain map and the frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node includes:

[0016] For heterogeneous graph nodes belonging to the same frequency band in the heterogeneous graph node set, the spatial domain graph is calculated based on a preset node distance threshold and a distance between every two heterogeneous graph nodes;

[0017] For the heterogeneous graph nodes belonging to the same acquisition grid in the heterogeneous graph node set, the frequency domain graph is calculated based on a preset frequency difference threshold and the frequency difference between every two heterogeneous graph nodes.

[0018] In some embodiments, for the heterogeneous graph nodes in the heterogeneous graph node set belonging to the same frequency band, the spatial domain graph is calculated based on a preset node distance threshold and the distance between every two heterogeneous graph nodes, including:

[0019] For heterogeneous graph nodes belonging to the same frequency band in the heterogeneous graph node set, calculating the inter-node distance between every two heterogeneous graph nodes;

[0020] When the distance between the nodes is less than the node distance threshold and there is no occluder node between the corresponding two heterogeneous graph nodes, a spatial domain heterogeneous edge is constructed between the two heterogeneous graph nodes to obtain the spatial domain graph.

[0021] In some embodiments, for the heterogeneous graph nodes in the heterogeneous graph node set belonging to the same acquisition grid, the frequency domain graph is calculated based on a preset frequency difference threshold and the frequency difference between every two heterogeneous graph nodes, including:

[0022] For the heterogeneous graph nodes belonging to the same acquisition grid in the heterogeneous graph node set, calculating the inter-node frequency difference between every two heterogeneous graph nodes;

[0023] When the frequency difference between the nodes is less than the frequency difference threshold, a frequency domain heterogeneous edge is constructed between two heterogeneous graph nodes to obtain the frequency domain graph.

[0024] In some embodiments, the heterogeneous graph neural network includes a spatial domain aggregate graph attention network and a frequency domain aggregate graph attention network, and the spatial domain graph, the frequency domain graph, and the heterogeneous graph node information of each heterogeneous graph node are input into the heterogeneous graph neural network so that the heterogeneous graph neural network outputs a multi-band spectrum map corresponding to the target area, including:

[0025] Inputting the spatial domain graph and the heterogeneous graph node information of each heterogeneous graph node into the spatial domain aggregate graph attention network of the heterogeneous graph neural network, so that the spatial domain aggregate graph attention network performs message propagation on the heterogeneous graph node information of each heterogeneous graph node according to the spatial domain graph to obtain intermediate spectrum map data;

[0026] The frequency domain graph and the intermediate spectrum map data are input into the frequency domain aggregate graph attention network of the heterogeneous graph neural network, so that the frequency domain aggregate graph attention network performs message transmission on the intermediate spectrum map data according to the frequency domain graph to obtain the multi-band spectrum map.

[0027] In some embodiments, after the heterogeneous graph neural network outputs the multi-band spectrum map corresponding to the target area, the method further includes:

[0028] Determining at least one heterogeneous graph node having a true label in the multi-band spectrum map output by the heterogeneous graph neural network;

[0029] Calculating a loss function according to the predicted value of the heterogeneous graph node in the multi-band spectrum map and the corresponding true label;

[0030] The heterogeneous graph neural network is back-propagated and trained according to the loss function, so that the heterogeneous graph neural network modifies the multi-band spectrum map according to the loss function.

[0031] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a multi-band spectrum map construction device, comprising:

[0032] A rasterization module is used to obtain map data of a target area and perform rasterization processing on the map data to obtain a target raster map;

[0033] An acquisition module is used to randomly select a plurality of acquisition grids in the target grid map and acquire the radiation signal strength in different frequency bands in each acquisition grid;

[0034] A modeling module, used to take the position information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid as heterogeneous graph node information, and model each acquisition grid in each frequency band as a heterogeneous graph node;

[0035] A construction module, used to construct a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node;

[0036] An input module is used to obtain a pre-trained heterogeneous graph neural network, and input the spatial domain graph, the frequency domain graph and the heterogeneous graph node information of each heterogeneous graph node into the heterogeneous graph neural network, so that the heterogeneous graph neural network outputs a multi-band spectrum map corresponding to the target area.

[0037] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the multi-band spectrum map construction method as described in the first aspect.

[0038] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a storage medium, which is a computer-readable storage medium, and the storage medium stores a computer program. When the computer program is executed by a processor, it implements the multi-band spectrum map construction method as described in the first aspect.

[0039] A method, device, electronic device and storage medium for constructing a multi-band spectrum map proposed in an embodiment of the present application first obtain map data of a target area, and rasterize the map data to obtain a target grid map; then, randomly select multiple acquisition grids in the target grid map to obtain the radiation signal strength of each acquisition grid in different frequency bands; next, use the position information of each acquisition grid and the radiation signal strength of each acquisition grid in different frequency bands as heterogeneous graph node information, and model each acquisition grid in each frequency band as a heterogeneous graph node; then, construct a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node; finally, obtain a pre-trained heterogeneous graph neural network, and input the spatial domain map, the frequency domain map and the heterogeneous graph node information of each heterogeneous graph node into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area.

[0040] With respect to the problems in the background technology, the multi-band spectrum map construction method provided by the embodiment of the present application first obtains the target grid map by rasterizing the map data of the target area. The target grid map provides a fine spatial framework for the construction of the spectrum map; then, by randomly selecting multiple acquisition grids in the target grid map, the radiation signal strength of each acquisition grid in different frequency bands is obtained. Next, the position information of each acquisition grid and the radiation signal strength of each acquisition grid in different frequency bands are used as heterogeneous graph node information, and each acquisition grid in each frequency band is modeled as a heterogeneous graph node. The spatial domain map and frequency domain map of the target grid map are constructed according to the heterogeneous graph node information of each heterogeneous graph node, and the spatial domain map, frequency domain map and heterogeneous graph node information of each heterogeneous graph node are input into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area. The spatial domain map reflects the spatial correlation between nodes, and the frequency domain map depicts the correlation between different frequencies. By constructing these two heterogeneous graphs, spectrum sensing data can be represented and modeled from different dimensions. Heterogeneous graph neural networks can effectively fuse and utilize heterogeneous information in spatial domain graphs and frequency domain graphs, and aggregate and update node features through the graph attention mechanism. This fusion method of heterogeneous information can capture high-order relationships in spectrum sensing data and improve the accuracy of spectrum map construction. Heterogeneous graph neural networks can complete and infer complete multi-band spectrum maps from sparse spectrum sensing data through iterative updates of node information. The network's information propagation capabilities enable it to utilize the correlation between nodes, fill in missing spectrum information, and generate high-quality multi-band spectrum maps.

[0041] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of a method for constructing a multi-band spectrum map provided in an embodiment of the present application;

[0043] Figure 2 An embodiment of the present application provides Figure 1 Flow chart of step 103;

[0044] Figure 3 An embodiment of the present application provides Figure 1 Flow chart of step 104;

[0045] Figure 4 An embodiment of the present application provides Figure 3 Another flow chart of step 301 in FIG.

[0046] Figure 5 An embodiment of the present application provides Figure 3 Another flow chart of step 302;

[0047] Figure 6 An embodiment of the present application provides Figure 1 Another flow chart of step 106;

[0048] Figure 7 A schematic diagram of the structure of a heterogeneous graph neural network provided in one embodiment of the present application;

[0049] Figure 8 An embodiment of the present application provides Figure 1 Another flow chart after step 106;

[0050] Fig. 9 A schematic diagram of a multi-band spectrum map construction device provided in an embodiment of the present application;

[0051] Fig.10 A schematic diagram of the hardware structure of an electronic device provided in yet another embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] It should be noted that although the functional modules are divided in the device schematic and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0055] In recent years, with the diversification of ultra-wideband services and the demand for high-precision sensing, wireless communication technology has developed rapidly and is expected to support the popularization of large-scale intelligent terminals. To this end, there are a series of design and control problems to be solved, such as network planning, resource allocation, dynamic spectrum access, spectrum sharing, etc., which require high-precision spectrum maps to accurately evaluate the radio environment. Spectrum maps can describe power spectrum density (PSD) information in different spaces, times, and frequencies, and are usually reconstructed from sparse observation data collected by deployed sensors and mobile devices. However, how to reconstruct high-precision full-band spectrum maps based on sparse samples from space, time, and space is a long-standing challenge. At present, most traditional spectrum map reconstruction methods predict the PSD in two-dimensional space of a given frequency band (single frequency band), focusing on the spectrum distribution at different locations. For example, radio propagation models can be applied to interpolate sparse spectrum maps, while data-driven methods are also applied to predict spectrum maps based on surrounding observations. However, many modern wireless applications have recently seen an increasing demand for full-band information, such as spectrum sharing, dynamic spectrum access, and ultra-wideband communications.

[0056] In the related technology, the existing multi-band spectrum map construction methods can be mainly divided into model-based and data-based methods. The model-based method assumes the signal propagation model and directly estimates the parameters in the assumed model; the data-based method considers the correlation between data and predicts unknown data through existing sparse data. However, the multi-band spectrum maps predicted by the above two methods cannot meet the accuracy requirements of existing high-precision spectrum maps.

[0057] In response to the above problems, the multi-band spectrum map construction method provided in the embodiment of the present application first obtains the target grid map by rasterizing the map data of the target area. The target grid map provides a fine spatial framework for the construction of the spectrum map; then, by randomly selecting multiple acquisition grids in the target grid map, the radiation signal strength of each acquisition grid in different frequency bands is obtained. Next, the position information of each acquisition grid and the radiation signal strength of each acquisition grid in different frequency bands are used as heterogeneous graph node information, and each acquisition grid in each frequency band is modeled as a heterogeneous graph node. The spatial domain map and frequency domain map of the target grid map are constructed according to the heterogeneous graph node information of each heterogeneous graph node, and the spatial domain map, frequency domain map and heterogeneous graph node information of each heterogeneous graph node are input into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area. The spatial domain map reflects the spatial correlation between nodes, and the frequency domain map depicts the correlation between different frequencies. By constructing these two heterogeneous graphs, spectrum sensing data can be represented and modeled from different dimensions. Heterogeneous graph neural networks can effectively fuse and utilize heterogeneous information in spatial domain graphs and frequency domain graphs, and aggregate and update node features through the graph attention mechanism. This fusion method of heterogeneous information can capture high-order relationships in spectrum sensing data and improve the accuracy of spectrum map construction. Heterogeneous graph neural networks can complete and infer complete multi-band spectrum maps from sparse spectrum sensing data through iterative updates of node information. The network's information propagation capabilities enable it to utilize the correlation between nodes, fill in missing spectrum information, and generate high-quality multi-band spectrum maps.

[0058] The following will further describe a method, device, electronic device and storage medium for constructing a multi-band spectrum map provided by an embodiment of the present application. First, a method for constructing a multi-band spectrum map in an embodiment of the present application will be specifically described.

[0059] Reference Figure 1 , which is an optional flowchart of the multi-band spectrum map construction method provided in an embodiment of the present application, Figure 1 The method may include but is not limited to steps 101 to 105. It can also be understood that this embodiment Figure 1 The order of step 101 to step 105 is not specifically limited, and the order of steps can be adjusted or some steps can be reduced or added according to actual needs.

[0060] Step 101: Obtain map data of a target area, and perform rasterization processing on the map data to obtain a target raster map.

[0061] The following is a detailed explanation of step 101.

[0062] Specifically, the target area is the area where a multi-band spectrum map needs to be constructed. First, the map data of the target area is obtained. The map data can be a map in the form of an image, such as a satellite map, a plane map, etc.; it can also be a map in the form of a three-dimensional model, such as a three-dimensional real-life map, etc. Then, the obtained map data is rasterized. Rasterization refers to dividing the map into multiple grids (also called grids) of the same size and arranged in sequence. Each grid is uniquely identified by its position coordinates in the grid array. The size of the grid can be set according to actual needs. For example, the map can be divided into a grid of 10 meters × 10 meters, or a grid of 50 meters × 50 meters, etc. Rasterization can be achieved by a variety of methods, such as dividing the map image using an image processing algorithm, dividing the three-dimensional map using three-dimensional modeling software, etc.

[0063] After the above-mentioned rasterization processing, a target grid map of the target area can be obtained. In the target grid map, the entire target area is divided into a regular grid array, and each grid has a unique position identifier. This rasterization representation is conducive to subsequent data processing and analysis.

[0064] For example, suppose you need to build a multi-band spectrum map for the central urban area of ​​a city, which is 1000 meters by 1000 meters. First, obtain a satellite map image of the central urban area. Then, divide the satellite map image into 100 grids of 100 meters by 100 meters. In this way, you get a target grid map consisting of 10×10 grids, where each grid has a unique row and column number.

[0065] It should be noted that the shape and size of the above grid can be flexibly set according to actual needs, and this embodiment does not limit this. For example, the grid can be square, rectangular, hexagonal or other shapes, and the size of the grid can also be flexibly determined according to factors such as the scope of the target area and the resolution of the required spectrum map.

[0066] In some embodiments, the grid map can be three-dimensional, and each grid has not only length and width, but also height, thereby forming a three-dimensional cubic grid. Using a three-dimensional grid map can more accurately represent the spatial information of the target area, especially when there are complex terrain or high-rise buildings in the target area.

[0067] Specifically, when generating a three-dimensional grid map, first obtain the three-dimensional map data of the target area, such as a three-dimensional real scene model, a three-dimensional topographic map, etc. Then, divide the three-dimensional map data into a three-dimensional cubic grid array. Each three-dimensional grid can be uniquely identified by its spatial position coordinates (such as x, y, z coordinates) in the three-dimensional grid array. The length, width and height of the three-dimensional grid can be flexibly set according to actual needs. For example, the three-dimensional map can be divided into a 10m×10m×10m cubic grid.

[0068] By introducing the height dimension, the 3D grid map can more realistically reflect the 3D spatial characteristics of the target area. For example, for mountainous areas with ups and downs, the 3D grid map can be used to distinguish the spatial information at different elevations; for high-rise building areas in cities, the 3D grid map can be used to distinguish the spatial information of different floors. This is crucial for the subsequent construction of multi-band spectrum maps, because the propagation characteristics of radiation signals at different heights may be significantly different and need to be distinguished during modeling.

[0069] For example, suppose you need to build a multi-band spectrum map for a central business district in a city. There are many high-rise buildings in the business district, and the highest building is more than 200 meters high. First, obtain the 3D real scene model data of the business district. Then, considering the height distribution of the buildings, divide the 3D real scene model into a 3D cube grid of 50 meters × 50 meters × 50 meters. In this way, you get a 3D target grid map, where each grid corresponds to a 3D spatial area of ​​the business district, and you can distinguish the spatial information of different floors of different buildings.

[0070] It should be noted that, although the use of three-dimensional grid maps can provide a richer and more accurate representation of spatial information, it also increases the complexity of data processing and calculation. Therefore, whether to use a three-dimensional grid map needs to be weighed according to factors such as the needs of the actual application scenario and the limitations of computing resources. In some scenarios where the requirements for spatial information are not too high, the use of a two-dimensional grid map may be sufficient. This embodiment does not limit whether to use a three-dimensional grid map, and it can be flexibly selected according to actual needs.

[0071] Step 102: randomly select multiple acquisition grids in the target grid map, and obtain the radiation signal intensity of each acquisition grid in different frequency bands.

[0072] The following is a detailed explanation of step 102.

[0073] After the target grid map is obtained in step 101, it is necessary to select some grids in the target grid map as acquisition grids, and obtain the radiation signal intensity data of these acquisition grids in different frequency bands for subsequent multi-band spectrum map construction.

[0074] Specifically, firstly, a plurality of grids are randomly selected in the target grid map as acquisition grids. The acquisition grids can be selected in a completely random manner, that is, a specified number of grids are randomly selected in the target grid map. Then, the radiation signal acquisition device is used to obtain the radiation signal strength of each acquisition grid in different frequency bands. Specifically, for each acquisition grid, its radiation signal strength in multiple preset frequency bands (such as the frequency band set F={f1, f2,..., f k ,...,f K Here, the radiation signal strength can be measured by indicators such as the received signal strength indication (RSSI). The radiation signal acquisition equipment can be various radio frequency sensors (such as IoT nodes, mobile communication base stations, etc.) randomly deployed in the target area.

[0075] It should be noted that since the acquisition devices at different locations may have different spectrum sensing capabilities, the acquired radiation signal strength data may be sparse in the frequency domain, that is, not every acquisition grid can obtain signal strength data of all frequency bands. At the same time, since the number of acquisition grids is limited and the distribution is not dense, the acquired radiation signal strength data is also sparse in the spatial domain.

[0076] For example, suppose that in the target grid map obtained in step 101, 100 grids are randomly selected as acquisition grids, and 5 preset frequency bands are set. The 100 acquisition grids are measured one by one using spectrum monitoring equipment, and the RSSI data of each acquisition grid in 5 frequency bands is obtained. However, since some acquisition grids are not equipped with monitoring equipment or the deployed monitoring equipment can only cover certain specific frequency bands, the collected RSSI data presents the characteristics of sparse distribution, which has both sparsity in the frequency domain and sparsity in the space domain. It is also possible to collect multiple RSSI data within a period of time and take the average value for subsequent calculation.

[0077] In some embodiments, there are M radiation sources randomly distributed on different grids in the target area, and the radiation source set is defined as M={1, 2, ..., m, ..., M}. Different radiation sources can transmit signals in different frequency bands. Assuming there are K frequency bands in total, the frequency band set is the set F={f1, f2, ..., f k ,...,f K}, and multiple radiation signal acquisition devices are sparsely deployed at different locations in the target area. The radiation signals are collected through these acquisition devices. Since the radiation signal acquisition devices deployed at different locations on the ground may have different perception capabilities, that is, different radiation signal acquisition devices may obtain measurements of different frequency bands, this leads to the sparseness of multi-band spectrum map measurements in the frequency domain. At the same time, in the same frequency band, the radiation signal acquisition device cannot cover each grid, which leads to the sparseness of multi-band spectrum maps in the spatial domain. In general, the multi-band spectrum map measurement of K frequency bands in the area of ​​interest has both spatial and frequency domain sparsity.

[0078] Step 103: taking the position information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid as heterogeneous graph node information, and modeling each acquisition grid in each frequency band as a heterogeneous graph node.

[0079] Step 103 is explained in detail below.

[0080] For each acquisition grid, its location information (such as the row and column number of the grid or the coordinates of the center point) and the radiation signal strength value in each frequency band are extracted to form the node information of a heterogeneous graph node. The node information of a heterogeneous graph node includes the location attribute and spectrum attribute of the node. For example, for the acquisition grid (i, j), assume that it is in the frequency band set F={f1,f2,...,f k}, the RSSI values ​​are {RSSI ij1 ,RSSI ij2 ,...,RSSI ijk}, then the corresponding heterogeneous graph node information can be expressed as [(i,j), (f1, RSSI ij1 ),(f2,RSSI ij2 ),...,(f k ,RSSI ijk )]. When constructing node information, if a certain acquisition grid does not collect RSSI data in a certain frequency band, the corresponding spectrum attribute value can be set to zero.

[0081] Through the above steps, the process of mapping the acquisition grid to the heterogeneous graph node is completed. The heterogeneous graph node contains not only the location information of the acquisition grid, but also the spectrum attribute information under each frequency band. The grid location information and frequency band of each node are known, and the radiation signal strength is sparse in both frequency and spatial domains, which provides rich feature representation for the subsequent spectrum map construction.

[0082] Reference Figure 2 In some embodiments, step 103 may also include but is not limited to steps 201 to 203.

[0083] Step 201 : Model each grid in each frequency band as an original node according to the target grid map.

[0084] The following is a detailed explanation of step 201.

[0085] First, for each frequency band in the frequency band set, all grids in the target grid map are traversed, and each grid is defined as an original node. These original nodes contain the complete spatial information of the target area, but the actual signal acquisition situation has not been considered. The attributes of the original node can include the location information of the grid (such as row and column numbers) and the radiation signal intensity value of the grid in the frequency band (if the data has not been collected, it can be set to the default value first). After this step, the original node set under each frequency band is obtained, and the nodes in each set correspond one by one to the grids in the target grid map.

[0086] Step 202 : According to the position information of each acquisition grid, the occluder nodes in the original nodes are removed to obtain a heterogeneous graph node set.

[0087] Step 202 is explained in detail below.

[0088] In the actual signal acquisition process, some grids may not be able to collect signals due to the presence of obstructions (such as buildings, terrain obstacles, etc.), and the corresponding nodes are called obstruction nodes. These nodes are invalid in the construction of the spectrum map and need to be removed from the original node set. Specifically, for each frequency band, based on the location information of the acquisition grid, determine whether each node in the original node set corresponds to an acquisition grid. If the grid corresponding to a node is not selected as an acquisition grid, it is marked as an obstruction node. After traversing all nodes, remove those nodes marked as obstruction nodes to obtain a heterogeneous graph node set that does not contain obstruction nodes.

[0089] For example, a single frequency band A single grid under Observations are defined as nodes , for the frequency band The sparse spectrum map of a single frequency band (including occluded nodes, unobserved nodes, and observed nodes) can be expressed as , where each element It can be expressed as

[0090]

[0091] in Indicates that the nth grid is The average signal receiving strength of . By removing the obstruction nodes in the element, the node set under this frequency band can be obtained as This process is repeated for each frequency band to complete the modeling of all nodes in the electromagnetic space of interest.

[0092] Step 203, based on the heterogeneous graph node set, the position information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid are used as heterogeneous graph node information, and each acquisition grid in each frequency band is modeled as a heterogeneous graph node.

[0093] Among them, after obtaining the heterogeneous graph node set, it is necessary to assign actual node information to each node. Similar to the aforementioned step 103, the node information here also includes position attributes and spectrum attributes. For each node in the heterogeneous graph node set, firstly, according to the position number of the corresponding acquisition grid, the position attribute is added to it; then, according to the radiation signal strength value collected by the acquisition grid in each frequency band, the spectrum attribute of the node is assigned. If the acquisition grid corresponding to a node does not collect data in certain frequency bands, the corresponding spectrum attribute value is set to zero.

[0094] Step 104: construct a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node.

[0095] Step 104 is explained in detail below.

[0096] After completing the construction of the heterogeneous graph nodes in step 103, it is necessary to further construct the connection relationship between the heterogeneous graph nodes to form a complete heterogeneous graph structure. The heterogeneous graph contains two types of subgraphs: a spatial domain graph and a frequency domain graph, which respectively characterize the mutual correlation of nodes in the spatial dimension and the frequency dimension. According to the signal propagation model, the propagation characteristics of signals with small distances and similar frequencies are similar, and occlusion also has an important impact on signal propagation. Among them, distance and occlusion can be used as spatial domain influencing factors, and frequency can be used as a frequency domain influencing factor, which respectively affects the connection of nodes in the space of interest in the spatial domain and the frequency domain. In the spatial domain graph, nodes with similar spatial positions and no occlusion are connected, reflecting the local correlation in the geographic space; in the frequency domain graph, nodes with similar frequencies are connected, reflecting the correlation of spectral features. The structural relationship of the nodes of the heterogeneous graph is characterized from the spatial dimension and the frequency dimension, respectively, providing important prior information for subsequent spectrum map learning.

[0097] See also Figure 3 In some embodiments, step 104 may include but is not limited to steps 301 to 302 .

[0098] Step 301 , for heterogeneous graph nodes belonging to the same frequency band in a heterogeneous graph node set, a spatial domain graph is calculated based on a preset node distance threshold and a distance between every two heterogeneous graph nodes.

[0099] Among them, the spatial domain graph reflects the adjacency relationship of heterogeneous graph nodes in the same frequency band in the geographic space. In this step, only nodes belonging to the same frequency band are considered, because the nodes in different frequency bands have different spatial correlations and need to be processed separately. In order to construct the spatial domain graph, a preset node distance threshold is introduced as a criterion for judging whether there is a connection between nodes. The size of the node distance threshold determines the sparsity of the spatial domain graph. The larger the threshold, the more edges there are in the graph and the denser the connections between nodes; conversely, the smaller the threshold, the fewer edges there are in the graph and the sparser the connections between nodes. The selection of the threshold needs to be determined according to the actual application scenario and data characteristics to balance the expression ability of the graph and the computational complexity. After determining the node distance threshold, it is also necessary to calculate the actual distance between each two heterogeneous graph nodes as a basis for judging whether they are connected.

[0100] See also Figure 4 In some embodiments, step 301 may include but is not limited to steps 401 to 402.

[0101] Step 401 : for heterogeneous graph nodes belonging to the same frequency band in the heterogeneous graph node set, calculate the inter-node distance between every two heterogeneous graph nodes.

[0102] The following is a detailed explanation of step 401.

[0103] In this step, the heterogeneous graph node set is traversed to find all node pairs belonging to the same frequency band and calculate the node distances between them. The node distances can be calculated based on the node's location attribute. For example, if the node's location attribute is a two-dimensional grid coordinate (x, y), the Euclidean distance formula can be used to calculate the distance between two nodes: if the node's location attribute is a three-dimensional grid coordinate (x, y, z), the square of the difference in z coordinates can be added to the Euclidean distance formula. Through the above calculations, the actual distance between each pair of nodes in the same frequency band in the heterogeneous graph node set can be obtained.

[0104] Step 402: When the distance between nodes is less than the node distance threshold and there is no occluder node between the two corresponding heterogeneous graph nodes, a spatial domain heterogeneous edge is constructed between the two heterogeneous graph nodes to obtain a spatial domain graph.

[0105] Step 402 is explained in detail below.

[0106] The distance between each pair of nodes obtained in step 401 is judged to determine whether there should be a spatial domain heterogeneous edge between them. There are two criteria for judgment: one is whether the distance between nodes is less than the preset node distance threshold, and the other is whether there is an occlusion node between the grids corresponding to the two nodes. Only when these two conditions are met at the same time can a spatial domain heterogeneous edge be constructed between the two nodes. Among them, the distance between nodes is less than the threshold, which means that the two nodes are close enough in space and have a certain correlation; the absence of an occlusion node means that the grids corresponding to the two nodes are connected in geographic space and there are no obstacles blocking them. In step 202, the occlusion node has been marked, so the judgment can be made directly in this step.

[0107] In the spatial domain, the node distance difference threshold is set to d th , and because if there is an obstruction between nodes, the received signal will most likely pass through different paths, resulting in a large difference in RSSI reception. and , the spatial domain edge connection criterion can be defined as

[0108]

[0109] in Indicates the same frequency band node and The edge connection indicator function, Representation Node and The distance between any nodes in each frequency band is calculated. , thus obtaining the spatial domain graph (spatial domain adjacency matrix).

[0110] By traversing all pairs of nodes in the same frequency band and constructing spatial domain heterogeneous edges according to the above standards, a complete spatial domain graph can be obtained. The spatial domain graph is an undirected graph, with nodes as heterogeneous graph nodes and edges as spatial domain heterogeneous edges. The spatial domain graph depicts the adjacency relationship of nodes in the same frequency band in the geographic space, reflects the local correlation in space, and provides important prior knowledge for subsequent spectrum map learning.

[0111] By introducing the node distance threshold and the distance calculation between nodes, and considering the influence of the occluder nodes, a heterogeneous subgraph that integrates geographic spatial information can be obtained, laying the foundation for the learning of the spectrum map. At the same time, by adjusting the node distance threshold, the sparsity of the spatial domain graph can be flexibly controlled to adapt to different task requirements.

[0112] Step 302 : For the heterogeneous graph nodes in the heterogeneous graph node set that belong to the same acquisition grid, a frequency domain graph is calculated based on a preset frequency difference threshold and the frequency difference between every two heterogeneous graph nodes.

[0113] Among them, the frequency domain graph reflects the association relationship between heterogeneous graph nodes in different frequency bands at the same position. In this step, only nodes belonging to the same acquisition grid are considered, because the correlation between nodes at different positions in frequency may be weak and has been reflected in the spatial domain graph, so it is no longer considered here. Similar to the spatial domain graph, in order to construct the frequency domain graph, a preset frequency difference threshold is introduced as a criterion for judging whether there is a connection between nodes. The size of the frequency difference threshold determines the sparsity of the frequency domain graph. The larger the threshold, the more edges there are in the graph and the denser the connections between nodes; conversely, the smaller the threshold, the fewer edges there are in the graph and the sparser the connections between nodes. The selection of the threshold needs to be determined according to the actual application scenario and the characteristics of the spectrum data to balance the expression ability of the graph and the computational complexity. After determining the frequency difference threshold, it is also necessary to calculate the frequency difference between each two heterogeneous graph nodes as a basis for judging whether they are connected.

[0114] See also Figure 5 In some embodiments, step 302 may include but is not limited to steps 501 to 502.

[0115] Step 501 : for the heterogeneous graph nodes in the heterogeneous graph node set that belong to the same acquisition grid, calculate the inter-node frequency difference between every two heterogeneous graph nodes.

[0116] Step 502: When the frequency difference between nodes is less than the frequency difference threshold, a frequency domain heterogeneous edge is constructed between two heterogeneous graph nodes to obtain a frequency domain graph.

[0117] Steps 501 to 502 are explained in detail below.

[0118] In some embodiments, the heterogeneous graph node set is first traversed to find all node pairs belonging to the same acquisition grid, and the frequency difference between the nodes is calculated. The frequency difference between each pair of nodes in the same position in the heterogeneous graph node set is obtained. Then, the frequency difference between each pair of nodes obtained in step 501 is judged to determine whether there should be a frequency domain heterogeneous edge between them. The criterion for judgment is whether the frequency difference between the nodes is less than the preset frequency difference threshold. If the frequency difference between the nodes is less than the threshold, it means that the two nodes are close enough in frequency and have a certain correlation, and a frequency domain heterogeneous edge should be constructed between them; conversely, if the frequency difference between the nodes is greater than or equal to the threshold, it means that the two nodes are greatly different in frequency and have a weak correlation, and there is no need to construct a frequency domain heterogeneous edge.

[0119] By traversing all pairs of nodes at the same location and constructing frequency domain heterogeneous edges according to the above standards, a complete frequency domain graph can be obtained. The frequency domain graph is also an undirected graph, with nodes as heterogeneous graph nodes and edges as frequency domain heterogeneous edges. The frequency domain graph depicts the relationship between nodes in different frequency bands at the same location, reflecting the correlation on the spectrum.

[0120] By introducing the frequency difference threshold and calculating the frequency difference between nodes, a heterogeneous subgraph reflecting the spectral correlation is obtained, which complements the spatial domain graph and together constitutes the complete structure of the heterogeneous graph. At the same time, by adjusting the frequency difference threshold, the sparsity of the frequency domain graph can be flexibly controlled to meet different task requirements.

[0121] Similarly, in the frequency domain, the edge connectivity criterion can be defined as:

[0122]

[0123] in Represents nodes of different frequency bands in the same location grid and The edge connection indicator function between represents the frequency difference threshold, Indicates frequency band and The absolute value of the difference between the carrier frequencies of the grid at each location is calculated. , thus obtaining the frequency domain graph (frequency domain adjacency matrix). After completing the modeling of the spatial and frequency domain edges, the heterogeneous graph is modeled.

[0124] Through the above-mentioned dual-domain fusion heterogeneous graph modeling method, the information of both spatial and spectral dimensions is fully utilized, which enables the subsequent graph neural network model to learn a more comprehensive and accurate spectrum map representation.

[0125] In some embodiments, the node distance threshold and the frequency difference threshold may be determined by the following steps:

[0126] Step 1: Get a spectrum map dataset containing multiple frequency bands as the basis for threshold selection. The dataset should contain the complete spectrum map ground truth and the corresponding sparse sampling points.

[0127] Step 2: According to the spectrum map and sampling points in the data set, a series of heterogeneous graphs under different thresholds are constructed according to the methods of steps 101 to 104. Specifically, a candidate set of node distance thresholds {d1, d2, ..., d m｝ and a candidate set of frequency difference thresholds {f1, f2, ..., f n}, then for each pair (d i , f j) to construct a corresponding heterogeneous graph, numbered G ij .

[0128] Step 3: For each constructed heterogeneous graph, number G ij， The pre-trained heterogeneous graph neural network model is used to reconstruct and infer the spectrum map according to the method of step 105. Sparse sampling points are input as labeled data, and the heterogeneous graph neural network will output a complete spectrum map prediction value.

[0129] Step 4: Evaluate the spectrum map reconstruction score of the heterogeneous graph neural network on each heterogeneous graph. Compare the predicted value of the spectrum map with the true value, calculate the corresponding mean square error (MSE), and obtain a performance matrix P, where the element P ij Indicates that in a heterogeneous graph, number G ij The reconstruction score on .

[0130] Step 5: According to the performance matrix P, select the threshold combination with the best reconstruction performance (d i * , f j * ) as the final node distance threshold and frequency difference threshold. Specifically, we can find the element Pij* with the smallest mean square error in the matrix P, and its corresponding row and column index (i * , j * ) is the optimal threshold combination.

[0131] Step 6: After obtaining the preliminary optimal threshold combination, the optimal threshold combination can be locally fine-tuned according to the threshold step size to further improve the reconstruction performance. For example, in (d i * , f j * ), take a smaller threshold step size near the value of ), construct one or more new heterogeneous graphs, and repeat steps 3 to 5 until the mean square error cannot be further reduced.

[0132] Through the above steps, the optimal node distance threshold and frequency difference threshold can be adaptively selected from a data-driven perspective, so that the heterogeneous graph neural network can achieve the best performance in the spectrum map reconstruction task. Traditional threshold selection usually relies on manual experience and prior knowledge, which is subjective and uncertain. The data-driven threshold selection method can minimize manual intervention, automatically learn the optimal threshold from the data, and reduce dependence on prior knowledge. In addition, by optimizing the node distance threshold and frequency difference threshold, the sparsity of the heterogeneous graph can be controlled and the number of redundant edges can be reduced. This can not only reduce the storage and computing overhead of the heterogeneous graph, but also improve the training and reasoning efficiency of the heterogeneous graph neural network.

[0133] Step 105: Take the pre-trained heterogeneous graph neural network, and input the spatial domain graph, the frequency domain graph, and the heterogeneous graph node information of each heterogeneous graph node into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area.

[0134] The following is a detailed explanation of step 105.

[0135] After constructing the spatial domain graph reflecting the spatial correlation of nodes and the frequency domain graph reflecting the spectral correlation of nodes, each heterogeneous graph node contains the location and spectral attribute information. However, the node features of the heterogeneous graph are still relatively independent and local, lacking the perception and utilization of the global structure. Therefore, the heterogeneous graph neural network is used to further learn and fuse the structural information of the heterogeneous graph to generate a complete multi-band spectrum map.

[0136] Heterogeneous graph neural network is a deep learning model specially designed for processing heterogeneous graph data. It can simultaneously model different types of nodes and edges and learn the hidden representation of nodes. By propagating and aggregating the feature information of nodes on heterogeneous graphs, heterogeneous graph neural network can make full use of the structural information of the graph to obtain a more accurate and comprehensive node representation.

[0137] In this step, a pre-trained heterogeneous graph neural network model is obtained, and the constructed spatial domain graph, frequency domain graph, and attribute information of each heterogeneous graph node are input into the model. The heterogeneous graph neural network has learned how to propagate and fuse information on the heterogeneous graph during the training phase, so it can be directly applied to new tasks and data. The output of the heterogeneous graph neural network is the hidden representation of each heterogeneous graph node. These hidden representations integrate the node's location attributes, spectrum attributes, and structural information of the spatial domain graph and frequency domain graph, and can be regarded as a new encoding of the node. By reorganizing these node hidden representations into a grid map of the target area, a complete multi-band spectrum map can be obtained. Each grid corresponds to a heterogeneous graph node, and the value of the grid is the hidden representation of the node.

[0138] See also Figure 6 In some embodiments, step 105 may include but is not limited to steps 601 to 602.

[0139] Step 601, input the spatial domain graph and the heterogeneous graph node information of each heterogeneous graph node into the spatial domain aggregation graph attention network of the heterogeneous graph neural network, so that the spatial domain aggregation graph attention network propagates the heterogeneous graph node information of each heterogeneous graph node according to the spatial domain graph to obtain the intermediate spectrum map data.

[0140] Step 601 is explained in detail below.

[0141] like Figure 7As shown in the figure, the spatial domain aggregation graph attention network is the first submodule of the heterogeneous graph neural network. Its main purpose is to propagate and aggregate the attribute information of nodes on the spatial domain graph and learn node representations that contain spatial correlation. Specifically, the spatial domain aggregation graph attention network takes the spatial domain graph and the heterogeneous graph node attribute information as input. For each node on the spatial domain graph, the graph attention network first calculates the relationship weight between the node and its neighbor nodes through the attention mechanism, and then performs weighted aggregation of the attribute information of the neighbor nodes according to these weights to obtain the new feature representation of the node. This process can be regarded as a message propagation process. Each node receives information from the neighbor nodes and updates its own feature representation according to its connection relationship in the spatial domain graph. Through multiple rounds of message propagation and feature aggregation on the spatial domain graph, the spatial domain aggregation graph attention network can gradually fuse the spatial correlation information of the nodes and learn the node representations that contain spatial structures. These node representations not only contain the attribute information of the nodes themselves, but also fuse the information of the neighbor nodes.

[0142] The output of the spatial domain aggregate graph attention network is a set of new node feature representations, which are defined as intermediate spectrum map data. Compared with the original heterogeneous graph node attribute information, the intermediate spectrum map data already contains spatial correlation information, but has not yet considered the correlation in the frequency domain. Therefore, it needs to be further input into the frequency domain aggregate graph attention network for processing.

[0143] Step 602, input the frequency domain graph and the intermediate spectrum map data into the frequency domain aggregation graph attention network of the heterogeneous graph neural network, so that the frequency domain aggregation graph attention network transmits messages to the intermediate spectrum map data according to the frequency domain graph to obtain a multi-band spectrum map.

[0144] Step 602 is explained in detail below.

[0145] like Figure 7 As shown in the figure, the purpose of the frequency domain aggregate graph attention network is to propagate and aggregate the intermediate feature representations of nodes on the frequency domain graph, learn the node representations that contain spectral correlations, and generate the final multi-band spectrum map. Specifically, the frequency domain aggregate graph attention network takes the frequency domain graph and the intermediate spectrum map data as input. Similar to the spatial domain aggregate graph attention network, the frequency domain aggregate graph attention network also uses the attention mechanism to calculate the relationship weights between nodes, and aggregates the intermediate feature representations of neighboring nodes according to the weights to update the feature representation of the node itself. The difference is that the frequency domain aggregate graph attention network propagates messages on the frequency domain graph, so it aggregates spectral correlation information.

[0146] By performing multiple rounds of message propagation and feature aggregation on the frequency domain graph, the frequency domain aggregation graph attention network can further fuse the spectral correlation information of the nodes, and learn node information that contains both spatial correlation and spectral correlation based on the intermediate spectrum map data. This node information integrates the node information in the spatial domain graph and the frequency domain graph.

[0147] The output of the frequency domain aggregate graph attention network is the multi-band spectrum map data. The output node feature representation is reorganized into a grid in the target area. Each grid corresponds to a heterogeneous graph node, and the grid value is the feature representation of the node. This results in a complete multi-band spectrum map. This spectrum map not only reflects the spectrum distribution of the target area in different frequency bands, but also contains rich spatial correlation and spectrum correlation information, which can be used for subsequent spectrum analysis and application tasks.

[0148] See also Figure 8 In some embodiments, after step 105, steps 801 to 803 are also included.

[0149] Step 801, determining at least one heterogeneous graph node with a true label in a multi-band spectrum map output by a heterogeneous graph neural network.

[0150] The following is a detailed explanation of step 801.

[0151] In step 105, the heterogeneous graph neural network has output a complete multi-band spectrum map, in which each heterogeneous graph node has a predicted spectrum value. However, since the spectrum data actually collected is sparse and incomplete, it is impossible to provide a true spectrum label for each node. Therefore, it is necessary to find those nodes with true labels in the output spectrum map for subsequent loss function calculation and network optimization.

[0152] Specifically, first determine which heterogeneous graph nodes have real spectrum labels. These nodes usually correspond to the acquisition grids that collected spectrum data in step 102, because only when the data is actually collected can the real spectrum value be obtained as a label. These labeled nodes can be identified by the attribute information of the node or additional tags. After determining the labeled nodes, the predicted spectrum values ​​of these nodes are paired with the real label values ​​to form a (predicted value, real value) data pair for the next step of loss function calculation.

[0153] Step 802 , calculating a loss function according to the predicted values ​​of the heterogeneous graph nodes in the multi-band spectrum map and the corresponding true labels.

[0154] Step 802 is explained in detail below.

[0155] Since only the true labels of some nodes can be obtained during network training, and the complete spectrum map cannot be obtained as supervision information, the mask training method under semi-supervised learning is used to calculate the loss function. Specifically, the loss function only calculates the prediction error of labeled nodes and ignores the output of unlabeled nodes. The total number of nodes with labels is , the true label of the labeled sample is , the predicted value is , then the loss function of mask training is Defined as:

[0156]

[0157] Among them, z i is a mask indicator function. When the i-th node has a true label, z i =1, indicating that the node is used for loss function calculation and back propagation; when the i-th node does not have a true label, z i =0, indicating that the node does not participate in the loss function calculation, and its prediction error will not affect the update of network parameters.

[0158] By introducing the mask indicator function, the loss function only focuses on the prediction quality of labeled nodes, avoiding the introduction of unnecessary noise due to unlabeled nodes. At the same time, since the prediction error of each labeled node is squared and summed, the loss function can measure the overall fit of the spectrum map in the labeled area.

[0159] Step 803, performing back propagation training on the heterogeneous graph neural network according to the loss function, so that the heterogeneous graph neural network can correct the multi-band spectrum map according to the loss function.

[0160] Step 803 is explained in detail below.

[0161] After calculating the loss function, the back-propagation algorithm can be used to optimize the parameters of the heterogeneous graph neural network so that its output spectrum map is closer to the true label.

[0162] Specifically, the predicted value of the network output layer is first calculated based on the loss function to obtain the direction in which the predicted value should be adjusted to reduce the loss. Then, this gradient is back-propagated to the previous layers of the network, and the parameters of each layer are updated according to the gradient value, for example, through the gradient descent algorithm.

[0163] During the back propagation process, for nodes with labels, their prediction errors will be back propagated normally and affect the update of network parameters; for nodes without labels, since their mask indicator function is 0, the prediction errors will not be back propagated and will not affect the network parameters. In this way, the network can focus on learning labeled nodes and acquire knowledge from sparse labels without being disturbed by unlabeled nodes.

[0164] Through multiple rounds of iterative training, the heterogeneous graph neural network can gradually adjust its parameters so that the output spectrum map is closer to the true label in the labeled area, while giving a reasonable prediction value in the unlabeled area. This reflects the idea of ​​semi-supervised learning, which is to use a small amount of labeled data and a large amount of unlabeled data to train a robust and generalizable model.

[0165] Through steps 801 to 803, the mask training loss function and semi-supervised learning method enable the heterogeneous graph neural network to learn from sparse and incomplete spectrum labels and infer a complete spectrum map. Through the information fusion of the spatial domain graph and the frequency domain graph, as well as the aggregation and propagation of the attention mechanism, the heterogeneous graph neural network can effectively utilize the spatial correlation and spectrum correlation of the nodes, recover the global spectrum distribution from the local sparse spectrum sampling, and has good reasoning and completion capabilities. At the same time, the mask training method also gives the network strong adaptability, enabling it to handle spectrum sampling of different sparsity levels, and has broad practical application prospects.

[0166] See also Fig. 9 The embodiment of the present application further provides a multi-band spectrum map construction device 900, which can implement the above-mentioned multi-band spectrum map construction method, including:

[0167] The rasterization module 901 is used to obtain the map data of the target area and perform rasterization processing on the map data to obtain a target raster map;

[0168] An acquisition module 902 is used to randomly select a plurality of acquisition grids in the target grid map and acquire the radiation signal strength in different frequency bands in each acquisition grid;

[0169] Modeling module 903, used to use the position information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid as heterogeneous graph node information, and model each acquisition grid in each frequency band as a heterogeneous graph node;

[0170] A construction module 904 is used to construct a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node;

[0171] The input module 905 is used to obtain a pre-trained heterogeneous graph neural network, and input the spatial domain graph, the frequency domain graph and the heterogeneous graph node information of each heterogeneous graph node into the heterogeneous graph neural network, so that the heterogeneous graph neural network outputs a multi-band spectrum map corresponding to the target area.

[0172] A method, device, electronic device and storage medium for constructing a multi-band spectrum map proposed in an embodiment of the present application first obtain map data of a target area, and rasterize the map data to obtain a target grid map; then, randomly select multiple acquisition grids in the target grid map to obtain the radiation signal strength of each acquisition grid in different frequency bands; next, use the position information of each acquisition grid and the radiation signal strength of each acquisition grid in different frequency bands as heterogeneous graph node information, and model each acquisition grid in each frequency band as a heterogeneous graph node; then, construct a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node; finally, obtain a pre-trained heterogeneous graph neural network, and input the spatial domain map, the frequency domain map and the heterogeneous graph node information of each heterogeneous graph node into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area.

[0173] With respect to the problems in the background technology, the multi-band spectrum map construction method provided by the embodiment of the present application first obtains the target grid map by rasterizing the map data of the target area. The target grid map provides a fine spatial framework for the construction of the spectrum map; then, by randomly selecting multiple acquisition grids in the target grid map, the radiation signal strength of each acquisition grid in different frequency bands is obtained. Next, the position information of each acquisition grid and the radiation signal strength of each acquisition grid in different frequency bands are used as heterogeneous graph node information, and each acquisition grid in each frequency band is modeled as a heterogeneous graph node. The spatial domain map and frequency domain map of the target grid map are constructed according to the heterogeneous graph node information of each heterogeneous graph node, and the spatial domain map, frequency domain map and heterogeneous graph node information of each heterogeneous graph node are input into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area. The spatial domain map reflects the spatial correlation between nodes, and the frequency domain map depicts the correlation between different frequencies. By constructing these two heterogeneous graphs, spectrum sensing data can be represented and modeled from different dimensions. Heterogeneous graph neural networks can effectively fuse and utilize heterogeneous information in spatial domain graphs and frequency domain graphs, and aggregate and update node features through the graph attention mechanism. This fusion method of heterogeneous information can capture high-order relationships in spectrum sensing data and improve the accuracy of spectrum map construction. Heterogeneous graph neural networks can complete and infer complete multi-band spectrum maps from sparse spectrum sensing data through iterative updates of node information. The network's information propagation capabilities enable it to utilize the correlation between nodes, fill in missing spectrum information, and generate high-quality multi-band spectrum maps.

[0174] The present application also provides an electronic device, including:

[0175] at least one memory;

[0176] at least one processor;

[0177] at least one program;

[0178] The program is stored in the memory, and the processor executes the at least one program to implement the multi-band spectrum map construction method implemented in the present application. The electronic device can be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a car computer, etc.

[0179] See also Fig.10 , Fig.10 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0180] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0181] The memory 1002 can be implemented in the form of ROM (Read Only Memory), static storage device, dynamic storage device or RAM (Random Access Memory). The memory 1002 can store operating systems and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1002, and the processor 1001 calls and executes the multi-band spectrum map construction method of the embodiment of this application;

[0182] Input / output interface 1003, used to implement information input and output;

[0183] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0184] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0185] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0186] An embodiment of the present application further provides a storage medium, which is a computer-readable storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned multi-band spectrum map construction method is implemented.

[0187] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0188] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0189] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0190] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0191] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0192] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0193] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c. A), can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0194] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0195] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0196] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0197] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store programs.

[0198] The preferred embodiments of the present application are described above with reference to the accompanying drawings, but the scope of the rights of the present application is not limited thereto. Any modification, equivalent substitution and improvement made by those skilled in the art without departing from the scope and essence of the present application should be within the scope of the rights of the present application.

Claims

1. A method for constructing a multi-band spectrum map, characterized in that: The method comprises: Acquire map data of the target area, and perform rasterization processing on the map data to obtain a target raster map; Randomly select multiple acquisition grids in the target grid map to obtain the radiation signal intensity of each acquisition grid in different frequency bands; The location information of each acquisition grid and the radiation signal strength of each acquisition grid in different frequency bands are used as heterogeneous graph node information, and each acquisition grid in each frequency band is modeled as a heterogeneous graph node; Constructing a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node; A pre-trained heterogeneous graph neural network is obtained, and the spatial domain graph, the frequency domain graph, and the heterogeneous graph node information of each heterogeneous graph node are input into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area.

2. The method according to claim 1, characterized in that The method uses the location information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid as heterogeneous graph node information, and models each acquisition grid in each frequency band as a heterogeneous graph node, including: According to the target grid map, each grid in each frequency band is modeled as an original node; According to the position information of each acquisition grid, the occluder nodes in the original nodes are removed to obtain a heterogeneous graph node set; Based on the heterogeneous graph node set, the position information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid are used as heterogeneous graph node information, and each acquisition grid in each frequency band is modeled as a heterogeneous graph node.

3. The method according to claim 2, characterized in that The step of constructing a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node includes: For heterogeneous graph nodes belonging to the same frequency band in the heterogeneous graph node set, the spatial domain graph is calculated based on a preset node distance threshold and a distance between every two heterogeneous graph nodes; For the heterogeneous graph nodes belonging to the same acquisition grid in the heterogeneous graph node set, the frequency domain graph is calculated based on a preset frequency difference threshold and the frequency difference between every two heterogeneous graph nodes.

4. The method according to claim 3, characterized in that The calculating of the spatial domain graph for the heterogeneous graph nodes belonging to the same frequency band in the heterogeneous graph node set based on a preset node distance threshold and the distance between every two heterogeneous graph nodes includes: For the heterogeneous graph nodes belonging to the same frequency band in the heterogeneous graph node set, calculating the inter-node distance between every two heterogeneous graph nodes; When the distance between the nodes is less than the node distance threshold and there is no occluder node between the corresponding two heterogeneous graph nodes, a spatial domain heterogeneous edge is constructed between the two heterogeneous graph nodes to obtain the spatial domain graph.

5. The method according to claim 3, characterized in that: The step of calculating the frequency domain graph for the heterogeneous graph nodes belonging to the same acquisition grid in the heterogeneous graph node set based on a preset frequency difference threshold and a frequency difference between every two heterogeneous graph nodes includes: For the heterogeneous graph nodes in the heterogeneous graph node set that belong to the same acquisition grid, calculating the inter-node frequency difference between every two heterogeneous graph nodes; When the frequency difference between the nodes is less than the frequency difference threshold, a frequency domain heterogeneous edge is constructed between two heterogeneous graph nodes to obtain the frequency domain graph.

6. The method according to claim 1, characterized in that The heterogeneous graph neural network includes a spatial domain aggregate graph attention network and a frequency domain aggregate graph attention network. The spatial domain graph, the frequency domain graph and the heterogeneous graph node information of each heterogeneous graph node are input into the heterogeneous graph neural network to obtain a multi-band spectrum map corresponding to the target area, including: Inputting the spatial domain graph and the heterogeneous graph node information of each heterogeneous graph node into the spatial domain aggregate graph attention network of the heterogeneous graph neural network, so that the spatial domain aggregate graph attention network performs message propagation on the heterogeneous graph node information of each heterogeneous graph node according to the spatial domain graph to obtain intermediate spectrum map data; The frequency domain graph and the intermediate spectrum map data are input into the frequency domain aggregate graph attention network of the heterogeneous graph neural network, so that the frequency domain aggregate graph attention network performs message transmission on the intermediate spectrum map data according to the frequency domain graph to obtain the multi-band spectrum map.

7. The method according to claim 1, characterized in that After obtaining the multi-band spectrum map corresponding to the target area, the method further includes: Determining at least one heterogeneous graph node having a true label in the multi-band spectrum map output by the heterogeneous graph neural network; Calculating a loss function according to the predicted value of the heterogeneous graph node in the multi-band spectrum map and the corresponding true label; The heterogeneous graph neural network is back-propagated and trained according to the loss function, so that the heterogeneous graph neural network modifies the multi-band spectrum map according to the loss function.

8. A multi-band spectrum map construction device, characterized in that: include: A rasterization module is used to obtain map data of a target area and perform rasterization processing on the map data to obtain a target raster map; An acquisition module is used to randomly select a plurality of acquisition grids in the target grid map and acquire the radiation signal strength in different frequency bands in each acquisition grid; A modeling module, used to take the position information of each acquisition grid and the radiation signal strength in different frequency bands in each acquisition grid as heterogeneous graph node information, and model each acquisition grid in each frequency band as a heterogeneous graph node; A construction module, used to construct a spatial domain map and a frequency domain map of the target grid map according to the heterogeneous graph node information of each heterogeneous graph node; An input module is used to obtain a pre-trained heterogeneous graph neural network, and input the spatial domain graph, the frequency domain graph and the heterogeneous graph node information of each heterogeneous graph node into the heterogeneous graph neural network, so that the heterogeneous graph neural network outputs a multi-band spectrum map corresponding to the target area.

9. An electronic device, characterized in that: It comprises a memory and a processor, the memory stores a computer program, and is characterized in that when the processor executes the computer program, it implements the multi-band spectrum map construction method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-band spectrum map construction method according to any one of claims 1 to 7 is implemented.

Images