An electromagnetic spectrum map construction method for a scenario with unprobed areas

Abstract

This invention discloses a method for constructing electromagnetic spectrum maps for scenarios with undetectable areas. The method includes: acquiring spectrum data and geographic environmental information of sensing nodes in the target area; generating an environmental map of the target area; discretizing the target area into a spatial grid to create missing spectrum map images and sampling location images, and creating building map images; modeling the missing spectrum map problem as an image inpainting problem, and establishing a spectrum map completion model based on a partially convolutional UNet network; inputting the three-channel color spectrum map image, sampling location image, and single-channel building map image into the spectrum map completion model to train the model; and processing the output of the spectrum map completion model to output a complete spectrum map. This invention can effectively improve the reconstruction accuracy of spectrum maps.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically to a method for constructing electromagnetic spectrum maps for scenarios with undetectable areas. Background Technology

[0002] With the rapid development of wireless communication technology and the advent of the sixth-generation (6G) mobile communication network era, diverse wireless services are constantly emerging. Against this backdrop, the accurate construction of spectrum maps becomes particularly crucial. These maps reflect the precise distribution of spectrum resources in the radio environment, effectively supporting dynamic sharing and intelligent management of spectrum resources, and are essential for improving the utilization efficiency of spectrum resources.

[0003] Existing spectrum map construction methods can be divided into model-based methods and model-free (data-driven) methods. Model-based spectrum map construction methods rely on prior models, primarily utilizing radio propagation models. However, they are sensitive to the selection of propagation models and their related parameters, and may struggle to account for specific geographical environmental factors such as shadows or obstacles. Model-free spectrum map construction methods can reconstruct radio maps based solely on measurement data, without relying on specific radio propagation models.

[0004] Publication No. CN116578660A discloses a method for constructing electromagnetic target situation maps based on sparse data. This method improves upon model-based methods by considering the influence of unknown environments on spectral signal strength, basis mismatch issues at target locations, and mutual interference between adjacent targets. Publication No. CN115205481A discloses a spectral map construction method based on graph neural networks. This method utilizes graph neural networks, combined with geographical location information and signal strength data, to construct a high-precision spectral map. However, this method assumes uniform distribution of measurement data and the ability to monitor the spectrum at all spatial locations. Most existing spectral map construction techniques consider scenarios where all areas are detectable and sensing data is uniform. However, in real-world scenarios, due to security considerations or physical limitations, unauthorized spectrum equipment is often prohibited from entering certain inaccessible areas, resulting in blank areas on radio maps. Therefore, traditional spectral map reconstruction methods are difficult to apply to spectral map construction in scenarios with non-uniformly distributed measurements and inaccessible areas, necessitating the development of new spectral map construction methods.

[0005] In complex scenarios, such as inaccessible regions, the adaptability of models is limited. Therefore, how to use deep learning methods to construct high-precision spectral maps in situations where sparse data sampling and inaccessible regions coexist is an urgent problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to provide a method for constructing electromagnetic spectrum maps for scenarios with undetectable areas. This method is particularly suitable for complex environments with undetectable regions. It utilizes partial convolution and combines it with dynamic updates of sampled location images to fully leverage the limited spectrum data of accessible areas, gradually updating the spectrum maps of missing portions of accessible areas and inaccessible areas. Furthermore, by fusing geographic environmental information, the reconstruction accuracy of the spectrum map is further improved.

[0007] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:

[0008] This invention discloses a method for constructing electromagnetic spectrum maps for scenarios with undetectable areas, the method comprising the following steps:

[0009] S1, collects spectrum data and geographic environment information of target area sensing nodes in scenarios with undetectable areas;

[0010] S2, preprocesses the collected spectrum data, and generates an environmental map of the target area based on the collected geographic environmental information;

[0011] S3, spatial grid discretize the target area, create the missing spectral map image and sampling location image of the target area based on the preprocessed spectral data, and create the building map image based on the generated environmental map;

[0012] S4 models the missing spectrum map problem as an image inpainting problem. Based on a partially convolutional UNet structure and combined with the dynamic update of the sampled location image, a spectrum map completion model is established. The input of the spectrum map completion model is a three-channel color sampled spectrum map image and a single-channel building map image, and the output is the completed spectrum map.

[0013] S5 collects complete spectrum map images, building map images, and randomly generated sampling location images at different base station locations to form the original dataset. The original dataset is then divided proportionally to obtain the training set, validation set, and test set. The spectrum map completion model is trained using the training set data and the validation set data.

[0014] S6 inputs the test set data into the best-trained spectrum map completion model and outputs the complete spectrum map.

[0015] In step S2, the collected spectrum data is preprocessed by normalization, and the spectrum map measurement values ​​are standardized between 0 and 1.

[0016] Step S3 further includes:

[0017] The target area is divided into discrete grids of size S×N at fixed intervals along the latitude and longitude directions, and the target area is divided into S×N square blocks of the same size.

[0018] Create the missing spectrum map image:

[0019]

[0020] In the formula, Let Ω be the latitude and longitude coordinates, and Ω be the corresponding received signal strength; if grid I = (i,j) contains a sample value, then [Ω] k ] i,j The value is assigned to the received signal strength acquired at the corresponding grid point; if grid I = (i,j) contains M sampled values, then [Ω] k ] i,j The value is assigned as the average of the M received signal strengths; if there is no sampled value within grid I = (i,j), then [Ω] k ] i,j The value is assigned to 0; frequency f k Missing Spectrum Map Represented as:

[0021]

[0022] In the formula, Ψ(f k ,c i,j The signal strength at grid I = (i,j) is represented by , and the subscript m indicates the m-th sample value, where 1 < m ≤ M. Indicates the transmission frequency is f k Missing spectrum map Indicates the transmission frequency is f k Spectral data at time grid I = (i,j).

[0023] Furthermore, in step S3, the sampling location image for:

[0024]

[0025] Building map images for:

[0026]

[0027] In the formula, B i,j This represents the value of the building map at grid I = (i,j).

[0028] Further, in step S4, the spectrum map completion model uses partial convolution combined with sampling position update operations to gradually complete the missing spectrum map, ultimately completing the construction of the electromagnetic spectrum map; the expression of the spectrum map completion model is:

[0029]

[0030] Where g(·) is the spectrum map completion model, To complete the full spectrum map output by the model.

[0031] Furthermore, in step S4, some convolution operations are as follows:

[0032]

[0033] Where W is the convolution kernel, (·) T ⊙ denotes transpose, X is the feature matrix of the current convolution (sliding) window, A is the corresponding binary mask matrix, and ⊙ denotes element-wise multiplication; X in the first layer is a part of the real spectrum map Ψ, A is a part of the sampling location image M; b is the bias term of the convolution operation, 1 is a matrix of the same size as the convolution kernel with all elements being 1, and the output x' is a feature value of the next layer convolution operation.

[0034] Furthermore, in step S4, the dynamic update expression for the sampled location image is as follows:

[0035]

[0036] When the feature matrix of the mask matrix A contains a value of 1, it indicates that the output of the corresponding part of the convolution operation is a valid value, and this position is marked as valid as the input of the next layer. After each convolution operation, the sampled position image is dynamically updated according to the output features, so that the network gradually expands the valid region in subsequent convolutions and fills in the missing regions.

[0037] Furthermore, in step S5, the training process of the spectrum map completion model includes:

[0038] S51, multiply the matrix corresponding to the complete spectrum map and the randomly generated sampling location image element by element to obtain the three-channel sampled spectrum map image, and combine it with the single-channel building map image to form the training set data;

[0039] S52, initialize the network training iteration count epoch equal to 1, initialize the maximum number of iterations, learning rate and training optimizer;

[0040] S53, the training set data is input into the UNet network with partial convolution in batches for training. The training error of each batch is backpropagated to optimize the network parameters. When all batches of training data have been backpropagated, one iteration is completed.

[0041] S54. Determine if the current training error is smaller than the previous one. If so, update the network model and save the network model parameters. If not, do not update the network model.

[0042] S55, Determine if network training is complete: Check if the current epoch has reached the set maximum epoch. If yes, output the completed spectrum map to complete the model. If no, increment the epoch by 1 and return to step S52 to continue network training.

[0043] Furthermore, in step S5, the error loss function l of the spectrum map completion model... total for:

[0044]

[0045] Where λ1, λ2, λ3, λ4, and λ5 are the corresponding loss weights; s and l u-s For the sampled value loss function:

[0046] l u-s =||(1-M)⊙(Ψ) o -Ψ)||1

[0047] l s =||M⊙(Ψ) o -Ψ)‖1

[0048] In the formula, ⊙ represents element-wise multiplication, and Ψ represents element-wise multiplication. o This is a spectrum map generated by the spectrum map completion model, where Ψ is the actual spectrum map and ||·||1 represents the L1 distance.

[0049] l pe For the perceptual loss function:

[0050]

[0051] In the formula, Q is the number of active layers, Y q This represents the activation map of the q-th selected layer extracted from the VGG-16 model pre-trained on ImageNet. The activation maps extracted from the VGG-16 model pre-trained on ImageNet are used to calculate the differences between the input image and the target image on these feature layers.

[0052] and Style loss function:

[0053]

[0054] In the formula, E q It is a factor used to normalize the activation map of the q-th selected layer;

[0055] l tv The total variation loss function is:

[0056]

[0057] In the formula, (x,y) are the location coordinates corresponding to the pixels in the spectrum map, and P represents the region after the missing region is expanded by 1 pixel.

[0058] Furthermore, the output of the spectrum map completion model is processed using the following formula to output the complete spectrum map:

[0059]

[0060] In the formula, For the sampling location image, Ψ o It is a spectrum map generated by the spectrum map completion model.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] This invention provides an electromagnetic spectrum map construction method for scenarios with undetectable regions. Addressing the issue of missing spectrum data, it proposes modeling this as an image inpainting problem. To better distinguish between sampled and unsampled regions, partial convolution replaces ordinary convolution. This invention utilizes a neural network method based on partial convolution and comprehensively considers geographical environmental information to assist in the construction of spectrum maps for target areas. This enables rapid spectrum map construction and significantly improves the accuracy of spectrum map construction. Attached Figure Description

[0063] Figure 1 This is a flowchart of the electromagnetic spectrum map construction method for scenarios with undetectable areas according to the present invention;

[0064] Figure 2 This is a flowchart of the framework of this invention, as well as a schematic diagram of the corresponding network structure G-PcNet and related parameters;

[0065] Figure 3 This is a comparison chart of the reconstruction error of the present invention and other methods in the scenario of increased number of blocks in large missing areas at a 10% sampling rate.

[0066] Figure 4 This is a comparison chart of the reconstruction error of the method of this invention and other methods under the same large missing area with different sampling rates;

[0067] Figure 5 The method of this invention and other methods, at a sampling rate of 10%, cover a large missing area of ​​50×50m. 2 A comparison chart of the reconstructed spectrum map visualization effects;

[0068] Figure 6 yes Figure 5 The corresponding normalized MSE result diagram. Detailed Implementation

[0069] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0070] Although the steps in this invention are arranged with numbers, they are not intended to limit the order of the steps. Unless the order of the steps is explicitly stated or the execution of a certain step requires other steps as a basis, the relative order of the steps can be adjusted.

[0071] Combination Figure 1 The specific steps of the environmental spectrum map construction method for undetectable areas, which takes into account environmental information according to the present invention, are described as follows:

[0072] Step 1: Collect spectrum data of sensing nodes in the target area with undetectable regions, and collect geographic environmental information of the target area;

[0073] Step 1.1: Collect spectral data of the target area with undetectable regions;

[0074] The first step is to identify and define the target area where spectrum data needs to be collected;

[0075] The second step is to collect spectrum data in the target area using high-precision sensing equipment, according to the predetermined sampling ratio requirements.

[0076] The third step is to upload the collected spectrum data to the central database for storage.

[0077] In embodiments of the present invention, due to deployment cost limitations, the number of sensing nodes is often limited and unevenly distributed. Furthermore, due to security considerations and physical limitations, unauthorized spectrum devices are typically prohibited from entering certain inaccessible areas, preventing data collection in these areas. Ultimately, this results in a discrete, unevenly distributed spectrum map with large areas of missing data.

[0078] Step 1.2: Collect geographic environmental information of the target area;

[0079] Detailed geographic environmental information of the area is collected using a Geographic Information System (GIS). Based on the collected data, an environmental map of the target area is generated and uploaded to the central database.

[0080] Step 2: Data preprocessing. The target area is discretized into a spatial grid to obtain the missing spectral map image; the environmental map of the target area is processed to create a building map image.

[0081] Step 2.1: Discretize the target area using a spatial grid to obtain the missing spectral map image;

[0082] The first step is to discretize the target region into a grid;

[0083] The target region is divided into square grids of equal length and width at fixed intervals. The target region is thus divided into S×N discrete grids.

[0084] After discretizing the target area into a grid, each grid may contain one or more sensing nodes. However, due to deployment cost limitations, most grids do not contain sensing nodes. Furthermore, due to security considerations and physical restrictions, unauthorized spectrum devices are typically prohibited from entering certain inaccessible areas, preventing the deployment of sensing nodes in these areas. Therefore, most grids do not contain sensing nodes, with only a few grids containing multiple sensing nodes. Based on this, a missing spectrum map image is created:

[0085] If the grid I = (i,j) contains a sampled value, then [Ω] k ] i,j The value is assigned to the received signal strength acquired at the corresponding grid point; if grid I = (i,j) contains M sampled values, then [Ω] k ] i,j The value is assigned as the average of the M received signal strengths; if there is no sampled value within grid I = (i,j), then [Ω] k ] i,j The value is assigned to 0;

[0086] Correspondingly, frequency f k Missing Spectrum Map Represented as:

[0087]

[0088] Among them, Ψ(f k ,c i,j The signal strength at grid I = (i,j) is represented by , and the subscript m indicates the m-th sample value, where 1 < m ≤ M. Indicates the transmission frequency is f k Missing spectrum map Indicates the transmission frequency is f k Spectral data at time grid I = (i,j).

[0089] Step 2.2: Extract the collected geographic environment information, create building map images, and input them into the network to assist in the construction of the spectrum map;

[0090] The first step is to create a binary building map.

[0091]

[0092] Among them, B i,j This represents the value of the building map at grid point I = (i,j).

[0093] In an embodiment of this invention, considering that unauthorized spectrum devices are typically prohibited from entering certain inaccessible areas due to security considerations and physical limitations, resulting in large blank areas in the spectrum map, this invention incorporates binary building map information to assist in spectrum map construction in order to enable the neural network to perform better reasoning.

[0094] Step 3: Establish a spectrum map completion model;

[0095] Modeling the missing spectrum map problem as an image inpainting problem:

[0096]

[0097] Where g(·) is the spectrum map completion model, To complete the full spectrum map output by the model.

[0098] Step 4: Input the three-channel sampled spectrum map image and the single-channel building map image into the network model, and train the network model using the training set data and validation set data. During network training, the real spectrum map is obtained from the complete spectrum map and is only used to calculate the loss and guide network learning. However, in the inference stage, only the sampled regions have true values. For non-sampled regions, the network must complete these regions through contextual relationships and the model's inference ability to output the complete spectrum map. The training process includes the following steps:

[0099] Step (1), as follows Figure 2 As shown, the three-channel sampled spectrum map image (the sampled spectrum map image during the training phase is obtained by multiplying the matrix corresponding to the complete spectrum map and the randomly generated sampled location image element by element) and the single-channel building map image are input into the UNet network G-PcNet with partial convolution.

[0100] Step (2): Initialize network training parameters, initialize the number of network training iterations (epoch) to 1, the maximum number of iterations to 50, the learning rate to 0.0001, and use the Adam optimization algorithm as the network training optimizer.

[0101] Step (3) involves inputting training data into the network in batches for training, and backpropagating the training error of each batch to optimize the network parameters. The error loss function is... total The calculation formula is as follows:

[0102]

[0103] Among them, l s and l u-s For the sampled value loss function:

[0104] l u-s =||(1-M)⊙(Ψ) o -Ψ)||1 (5)

[0105] l s =||M⊙(Ψ) o -Ψ)||1 (6)

[0106] Where ⊙ represents element-wise multiplication, and Ψ represents element-wise multiplication. o This is a spectrum map generated by the network, where ||·||1 represents the L1 (absolute) distance; l pe For the perceptual loss function:

[0107]

[0108] Y q This represents the activation map of the q-th selected layer extracted from the VGG-16 model pre-trained on ImageNet. and Style loss function:

[0109]

[0110] E q It is a factor used to normalize the activation map of the q-th selected layer. tv The total variation loss function is:

[0111]

[0112] P represents the region after the missing region is expanded by 1 pixel. λ1, λ2, λ3, λ4, and λ5 are the corresponding loss weights, which are 1, 6, 0.1, 0.05, and 120, respectively.

[0113] One epoch is defined as the backpropagation of all batches of data in the training data.

[0114] Step (4), save the optimal network model: In step (3), after one backpropagation, save the network model once, and determine whether the current training error is smaller than the previous one. If so, update the network model and save the network model parameters. If not, do not update the network model.

[0115] Step (5), determine whether the network training is complete: determine whether the current epoch has reached the set maximum epoch. If yes, proceed to step 6. If no, increment the epoch by 1, return to step (2), and continue network training.

[0116] Step 6: Input the test set data into the best-trained network model, process the output results, and output a complete spectrum map.

[0117] The output results are processed to produce a complete spectrum map.

[0118] The effects of the present invention will be further explained below with reference to simulation experiments.

[0119] 1. Simulation conditions and parameter settings:

[0120] The simulation experiments of this invention were conducted on a simulation platform using Python 3.11.3 and PyTorch 2.0.1. The server CPU was an Intel Core i9-13900K processor, equipped with an NVIDIA GeForce RTX4090 dedicated graphics card.

[0121] The maximum number of iterations for network training is 50, the learning rate is 0.001, the Adam optimization algorithm is selected as the network training optimizer, and the batch size is 12.

[0122] 2. Simulation content:

[0123] For convenience, the spectrogram measurements are further standardized to between 0 and 1. Since the original and normalized spectrograms have identical patterns, completion and conversion can be applied to the normalized spectrogram. The completion performance of the spectrogram is measured using the mean squared error (MSE); a smaller MSE indicates better completion.

[0124]

[0125] Where M represents the total number of samples, m represents the sample number, and 1 ≤ m ≤ M; ||·|| F Denotes the Frobenius norm of the tensor, Ψ m This indicates that spectrum map data is missing. represents the actual spectrum map data, and g(·) represents the spectrum map completion model.

[0126] This invention comprehensively compares five methods: PcNet, a partially convolutional neural network method; NW and FDGM, classic algorithms in image inpainting; and Kriging and IDW, representative methods in spectral map completion. To achieve a fair and intuitive comparison, the six MSE results from a single application scenario are normalized to facilitate quantitative analysis of the advantages and disadvantages of different methods. Specifically, the minimum value among the six MSE results is taken as the benchmark and normalized to 0; the maximum value is normalized to 1; and the remaining values ​​are calculated using the following formula:

[0127]

[0128] Among them, MSE norm,i Let MSE be the normalized MSE of the i-th method in the current scenario. i Let MSE be the original MSE of the i-th method in the current scenario. min and MSE max These are the minimum and maximum MSE values ​​for all methods in this scenario, respectively. Through this normalization operation, the MSE results for each scenario are mapped to the [0,1] interval, thereby eliminating the interference of different units and ensuring the fairness and consistency of the comparison.

[0129] Figure 3 This is a comparison chart of reconstruction errors under different sizes of large missing areas, considering environmental information and not considering environmental information. The scenario considered is: target area size of 512×512m. 2 The sampling rate is 10%, and the size of the inaccessible area is 50×50m. 2 Completion performance when the number ranges from 1 to 4. Figure 4 In the diagram, the horizontal axis represents the number of inaccessible areas, and the vertical axis represents the normalized MSE. Figure 4 The comparison shows that the proposed method for completing information that takes into account environmental information is superior to other methods.

[0130] Figure 4 This is a performance comparison chart of the present invention under different sampling rate scenarios, considering environmental information and not considering environmental information. The scenario considered is: the target area size is 512×512m. 2 The inaccessible area is 50×50m in size. 2 The completion performance at sampling rates of 1%, 5%, 10%, and 100%. Figure 3 In the diagram, the horizontal axis represents the sampling rate, and the vertical axis represents the normalized MSE. Figure 3 The comparison shows that the proposed method for completing information that takes into account environmental information is superior to other methods.

[0131] See Figure 5 and Figure 6 As can be seen from the visualization, the deep learning method significantly outperforms the conventional method. Regarding the completion performance for inaccessible regions, considering environmental information results in better performance than not considering it.

[0132] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0133] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0134] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0135] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, causing a series of operational steps to be executed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that run on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0136] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0137] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for constructing electromagnetic spectrum maps for scenarios with undetectable regions, characterized in that, The method includes the following steps: S1, collects spectrum data and geographic environment information of target area sensing nodes in scenarios with undetectable areas; S2, preprocesses the collected spectrum data, and generates an environmental map of the target area based on the collected geographic environmental information; S3, spatial grid discretize the target area, create the missing spectral map image and sampling location image of the target area based on the preprocessed spectral data, and create the building map image based on the generated environmental map; S4 models the missing spectrum map problem as an image inpainting problem. Based on a partially convolutional UNet structure and combined with the dynamic update of the sampled location image, a spectrum map completion model is established. The input of the spectrum map completion model is a three-channel color sampled spectrum map image and a single-channel building map image, and the output is the completed spectrum map. S5 collects complete spectrum map images, building map images, and randomly generated sampling location images at different base station locations to form the original dataset. The original dataset is then divided proportionally to obtain the training set, validation set, and test set. The spectrum map completion model is trained using the training set data and the validation set data. S6: Input the test set data into the best-trained spectrum map completion model and output the complete spectrum map; In step S5, the training process of the spectrum map completion model includes: S51, multiply the matrix corresponding to the complete spectrum map and the randomly generated sampling location image element by element to obtain the three-channel sampled spectrum map image, and combine it with the single-channel building map image to form the training set data; S52, initialize the network training iteration count epoch equal to 1, initialize the maximum number of iterations, learning rate and training optimizer; S53, the training set data is input into the UNet network with partial convolution in batches for training. The training error of each batch is backpropagated to optimize the network parameters. When all batches of training data have been backpropagated, one iteration is completed. S54. Determine if the current training error is smaller than the previous one. If so, update the network model and save the network model parameters. If not, do not update the network model. S55, Determine if the network training is complete: Determine if the current epoch has reached the set maximum epoch. If yes, output the completed spectrum map to complete the model. If no, increment the epoch by 1, return to step S52, and continue network training. In step S5, the error loss function of the spectrum map completion model for: ； in, , , , and For the corresponding loss weights; and For the sampled value loss function: ； ； In the formula, It is element-wise multiplication. It is a spectrum map generated by a spectrum map completion model. It is a real spectrum map. Indicates the L1 distance; For the perceptual loss function: ； In the formula, To activate the number of layers, This indicates the extraction of the first [item] from the VGG-16 model pre-trained on ImageNet. Activation maps of selected layers are used, extracted from a VGG-16 model pre-trained on ImageNet, to compute the differences between the input and target images on these feature layers. To complete the full spectrum map output by the model; and Style loss function: ； ； In the formula, It is used for normalization of the first Factors of the selected layer activation map; The total variation loss function is: ； In the formula, These are the location coordinates corresponding to pixels in the spectrum map. This represents the area after the missing region has been expanded by 1 pixel.

2. The method for constructing an electromagnetic spectrum map for scenarios with undetectable regions according to claim 1, characterized in that, In step S2, the collected spectrum data is preprocessed by normalization, and the spectrum map measurement values ​​are standardized between 0 and 1.

3. The method for constructing an electromagnetic spectrum map for scenarios with undetectable regions according to claim 1, characterized in that, Step S3 further includes: The target area is divided into equal intervals of size along the latitude and longitude directions at fixed intervals. The discrete grid divides the target region into equal parts. A square block of the same size; Create the missing spectrum map image: ； In the formula, Latitude and longitude coordinates The corresponding received signal strength; if the grid If it contains a sample value, then... The value is assigned to the received signal strength acquired at the corresponding grid point; if the grid... Contains If there are 1 sampled value, then The value is assigned to be the average of the M received signal strengths; if the grid... If there are no sampled values, then... Assign a value of 0; frequency Missing Spectrum Map Represented as: ； In the formula, Represents a grid Received signal strength at the location, subscript Indicates the first Each sample value, , Indicates the transmission frequency is Missing spectrum map Indicates the transmission frequency is Time grid Spectral data at the location.

4. The method for constructing an electromagnetic spectrum map for scenarios with undetectable regions according to claim 1, characterized in that, In step S3, the sampling location image for: ； Building map images for: ； In the formula, Represents a grid The value of the building map.

5. The method for constructing an electromagnetic spectrum map for scenarios with undetectable regions according to claim 1, characterized in that, In step S4, the spectrum map completion model uses partial convolution combined with sampling position update operations to gradually complete the missing spectrum map, ultimately constructing the electromagnetic spectrum map; the expression of the spectrum map completion model is: ； in, To complete the model for the spectrum map, To complete the full spectrum map output by the model.

6. The method for constructing an electromagnetic spectrum map for scenarios with undetectable regions according to claim 1, characterized in that, In step S4, some convolution operations are as follows: ； in, It is a convolution kernel. Indicates transpose. It is the feature matrix of the current convolutional sliding window. It is the corresponding binary mask matrix. Represents element-wise multiplication; first level It is a real spectrum map Part of For sampling location image Part of; This refers to the bias term for the convolution operation. It is a matrix with the same size as the convolution kernel, and all its elements are 1. The output is... This is a feature value for the next layer of convolution operation.

7. The method for constructing an electromagnetic spectrum map for scenarios with undetectable regions according to claim 1, characterized in that, In step S4, the dynamic update expression for the sampled location image is as follows: ； Where, when the mask matrix When the feature matrix contains a value of 1, it indicates that the output of the corresponding part of the convolution operation is a valid value, and this position is marked as valid as the input of the next layer. After each convolution operation, the sampled position image is dynamically updated according to the output features, so that the network gradually expands the valid region in subsequent convolutions and fills in the missing regions.

8. The method for constructing an electromagnetic spectrum map for scenarios with undetectable regions according to claim 1, characterized in that, The output of the spectrum map completion model is processed using the following formula to output a complete spectrum map: ； In the formula, For the sampling location image, It is a spectrum map generated by the spectrum map completion model.

Images