A time-frequency map simulation method for electromagnetic signal intelligent detection model training

Abstract

The application provides a time-frequency graph simulation method for electromagnetic signal intelligent detection model training, relates to the field of electromagnetic signal data simulation, and solves the problems of strong dependence on environment and signal acquisition equipment and great difficulty in accurate modeling of existing electromagnetic signal simulation methods; the method comprises the following steps: obtaining a target signal time-frequency graph data set and an electromagnetic environment time-frequency graph data set, labeling and storing target signal blocks; randomly selecting corresponding time-frequency graphs, denoted as labeled target signal time-frequency graph A and electromagnetic environment time-frequency graph B, extracting a first intensity matrix of the target signal block, performing random variation of multiple parameters, obtaining a second intensity matrix of the target signal block, and covering the second intensity matrix to a corresponding position in the electromagnetic environment time-frequency graph B; the electromagnetic environment time-frequency graph B after being covered is a newly generated target signal time-frequency graph; and the application can simulate a large number of high-generalization simulation time-frequency graphs based on limited real signal data and typical environment signal waveforms.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic signal data simulation, and specifically to a time-frequency diagram simulation method for training intelligent electromagnetic signal detection models. Background Technology

[0002] In the development and application of modern communication and related systems, accurate simulation of electromagnetic signals is crucial for training high-performance artificial intelligence models. However, in real-world environments, due to physical limitations, privacy requirements, or security considerations, it is often difficult to obtain a sufficient amount of actual electromagnetic signal data, which greatly limits the effective training of deep learning algorithms such as deep convolutional neural networks.

[0003] Currently, there are two main methods for simulating electromagnetic signals: computer simulation and radio frequency simulation.

[0004] Computer simulation: This simulation method relies on mathematical models of signals and simulates signal generation through software programming. Its advantages include high flexibility, low cost, and ease of rapid iteration. However, this method is highly dependent on the accuracy of the mathematical model; establishing accurate mathematical models for complex systems and channel environments is very challenging. Furthermore, due to the lack of accurate simulation of physical layer characteristics, the data generated by computer simulation may deviate significantly from reality, thus affecting the performance of models trained based on such data.

[0005] Radio frequency (RF) simulation: Compared to computer simulation, RF simulation more closely resembles the physical characteristics of actual signals. It uses RF technology to simulate signals at the physical level, enabling a better reproduction of real-world signal environments. However, this method typically requires complex hardware, such as signal generators and spectrum analyzers, and demands extremely high equipment configurations and environmental control for different simulation scenarios, thus increasing costs and limiting the construction of large-scale simulation datasets.

[0006] It is evident that, although both methods have their advantages, they also have certain limitations: computer simulation, while low in cost, lacks accuracy in complex systems; radio frequency simulation, while closer to reality, is expensive to implement and difficult to achieve large-scale scene coverage.

[0007] Therefore, existing technologies still face severe challenges in acquiring sufficiently diverse and high-quality electromagnetic signal data, which directly restricts the performance of artificial intelligence models in electromagnetic signal recognition and processing tasks. Summary of the Invention

[0008] Based on the current state of the technology, the purpose of this invention is to solve the problems of existing electromagnetic signal simulation methods, such as strong dependence on the environment and signal acquisition equipment and difficulty in accurate modeling. Therefore, a time-frequency diagram simulation method for training intelligent electromagnetic signal detection models is proposed. Based on limited real signal data and typical environmental signal waveforms, this invention can simulate and generate a large number of highly generalizable simulation time-frequency diagrams.

[0009] The present invention employs the following technical solutions to achieve its objective:

[0010] A time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model, the method comprising the following steps:

[0011] S1. Obtain the target signal time-frequency map dataset and the electromagnetic environment time-frequency map dataset;

[0012] S2. In the target signal time-frequency map dataset, label the target signal blocks of all target signal time-frequency maps and store the labeling information;

[0013] S3. Randomly select one time-frequency map data from the target signal time-frequency map dataset and the electromagnetic environment time-frequency map dataset, and denot them as the labeled target signal time-frequency map A and electromagnetic environment time-frequency map B, respectively.

[0014] S4. Based on the target signal time-frequency map A and its annotation information, extract the first intensity matrix of the target signal block in the target signal time-frequency map A, and perform various random parameter changes on the first intensity matrix of the target signal block.

[0015] S5. After the random changes of multiple parameters are completed, the second intensity matrix of the target signal block in the target signal time-frequency map A is obtained; the second intensity matrix of the target signal block is overlaid on the corresponding position in the electromagnetic environment time-frequency map B, and the overlaid electromagnetic environment time-frequency map B is the new target signal time-frequency map generated by simulation.

[0016] S6. Continue the random selection, random variation and coverage process from steps S3 to S5 until the number of new target signal time-frequency diagrams generated by the simulation meets the model training requirements, thereby completing the time-frequency diagram simulation process.

[0017] Specifically, in step S1, the target signal time-frequency map dataset is obtained by: acquiring multiple target signal time-domain waveforms, converting them into multiple target signal time-frequency maps through short-time Fourier transform, thereby forming a target signal time-frequency map dataset; in the target signal time-frequency map dataset, the dimension of the target signal time-frequency map is [N,M], where N is the number of spectral frames and M is the frame length.

[0018] Specifically, in step S1, the electromagnetic environment time-frequency map dataset is obtained by collecting multiple typical electromagnetic environment data without a target signal, converting them into multiple electromagnetic environment time-frequency maps through short-time Fourier transform, thereby forming an electromagnetic environment time-frequency map dataset; in the electromagnetic environment time-frequency map dataset, the dimension of the electromagnetic environment time-frequency map is the same as the dimension of the target signal time-frequency map, which is [N,M].

[0019] Specifically, in step S2, a rectangular bounding box is used to label the target signal block in the target signal time-frequency map, and the coordinates of the upper left and lower right corners of the bounding box are used as labeling information and stored; after the target signal time-frequency map dataset is labeled, the labeling information of each target signal time-frequency map is recorded as: [(x1 min ,y1 min ,x1 max ,y1 max ),(x2 min ,y2 min x2 max ,y2 max ),...].

[0020] Furthermore, in step S4, after the first intensity matrices of multiple target signal blocks are denoted as [M1, M2, M3, ...], six parameters are randomly varied: electromagnetic environment time-frequency diagram frequency variation, target signal block bandwidth variation, target signal block duration variation, random variation of target signal block gap, random variation of target signal block signal-to-noise ratio, and simulated variation of sudden interference enhancement and weakening of target signal block intensity.

[0021] Furthermore, in step S5, when the second intensity matrix of the target signal block is overlaid onto the corresponding position in the electromagnetic environment time-frequency diagram B, if the intensity of the target signal block at the overlaid position is lower than the intensity of the corresponding electromagnetic environment signal, the intensity value at the overlaid position is set to the original electromagnetic environment signal intensity, thereby updating the second intensity matrix of the target signal block in the electromagnetic environment time-frequency diagram B.

[0022] In summary, due to the adoption of this technical solution, the beneficial effects of this invention are as follows:

[0023] This invention utilizes limited real signal data and typical environmental signal waveforms to generate a large number of highly generalizable simulated time-frequency diagrams without significantly increasing data acquisition costs, thereby significantly reducing dependence on expensive signal acquisition equipment and specific environmental conditions.

[0024] The method of this invention extends the simulation of multidimensional parameters of the real signal time-frequency diagram and combines it with the time-frequency diagram of typical environmental signal waveforms for random synthesis, thus avoiding the complex and time-consuming precise modeling steps in traditional simulation methods and greatly reducing the modeling difficulty.

[0025] By employing multi-dimensional parameter extended simulation and random synthesis, this invention can generate a richer and more diverse range of simulation time-frequency diagrams. These time-frequency diagrams can not only cover a variety of practical application scenarios, but also better simulate complex environmental changes in the real world, thereby improving the realism of the simulation data.

[0026] Training a deep model using the highly generalizable simulated time-frequency graph generated by this invention can effectively improve the model's performance on unseen data, making it more adaptable and accurate when facing electromagnetic signals in different environments.

[0027] Compared to traditional electromagnetic signal simulation methods, this invention can quickly generate a large amount of high-quality training data at a lower cost, significantly improving the efficiency of the data preparation stage and shortening the time cycle from data acquisition to model deployment. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0029] Figure 2 This is a schematic diagram of the simulation scheme used in the embodiments of the present invention;

[0030] Figure 3 This is a schematic diagram of the time-frequency graph of the target signal acquired and processed in an embodiment of the present invention;

[0031] Figure 4 A schematic diagram showing the target signal blocks marked on the time-frequency plot of the target signal;

[0032] Figure 5 A schematic diagram for extracting labeled uplink and downlink simulated target signal blocks;

[0033] Figure 6 This is a schematic diagram showing the extracted target signal block after parameter changes;

[0034] Figure 7 This is a schematic diagram of the electromagnetic environment time-frequency diagram in an embodiment of the present invention;

[0035] Figure 8 This is a schematic diagram of the time-frequency diagram of the target signal generated by simulation in an embodiment of the present invention;

[0036] Figure 9 This is a schematic diagram illustrating the effect of applying the present invention to the detection of real target signals. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The parts of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0038] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0039] Example 1

[0040] like Figure 1 As shown, a time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model is presented. The overall process of this method can be summarized as follows:

[0041] S1. Obtain the target signal time-frequency map dataset and the electromagnetic environment time-frequency map dataset;

[0042] S2. In the target signal time-frequency map dataset, label the target signal blocks of all target signal time-frequency maps and store the labeling information;

[0043] S3. Randomly select one time-frequency map data from the target signal time-frequency map dataset and the electromagnetic environment time-frequency map dataset, and denot them as the labeled target signal time-frequency map A and electromagnetic environment time-frequency map B, respectively.

[0044] S4. Based on the target signal time-frequency map A and its annotation information, extract the first intensity matrix of the target signal block in the target signal time-frequency map A, and perform various random parameter changes on the first intensity matrix of the target signal block.

[0045] S5. After the random changes of multiple parameters are completed, the second intensity matrix of the target signal block in the target signal time-frequency map A is obtained; the second intensity matrix of the target signal block is overlaid on the corresponding position in the electromagnetic environment time-frequency map B, and the overlaid electromagnetic environment time-frequency map B is the new target signal time-frequency map generated by simulation.

[0046] S6. Continue the random selection, random variation and coverage process from steps S3 to S5 until the number of new target signal time-frequency diagrams generated by the simulation meets the model training requirements, thereby completing the time-frequency diagram simulation process.

[0047] This embodiment will provide a more detailed and preferred description of the steps described above.

[0048] In step S1, the target signal time-frequency map dataset is obtained by: acquiring multiple target signal time-domain waveforms, converting them into multiple target signal time-frequency maps through short-time Fourier transform (STFT transform), and each target signal time-frequency map may contain multiple target signal blocks, thus forming a target signal time-frequency map dataset; in the target signal time-frequency map dataset, the dimension of the target signal time-frequency map is [N,M], where N is the number of spectral frames and M is the frame length.

[0049] In step S1, the electromagnetic environment time-frequency map dataset is obtained by collecting multiple typical electromagnetic environment data without a target signal, and converting them into multiple electromagnetic environment time-frequency maps through short-time Fourier transform (STFT transform), thereby forming an electromagnetic environment time-frequency map dataset. In the electromagnetic environment time-frequency map dataset, the dimension of the electromagnetic environment time-frequency map is the same as the dimension of the target signal time-frequency map, which is [N,M].

[0050] In step S2, the preferred annotation method in this embodiment is as follows: The target signal blocks in the target signal time-frequency map are annotated with rectangular boxes, and the coordinates of the upper left and lower right corners of the rectangular boxes are used as annotation information and stored. After the target signal time-frequency map dataset is annotated, the annotation information for each target signal time-frequency map is recorded as: [(x1 min ,y1 min ,x1 max ,y1 max ),(x2 min ,y2 min x2 max ,y2 max ),...], which means that multiple target signal blocks are marked in the time-frequency diagram of each target signal.

[0051] After randomly selecting the target signal time-frequency diagram A and the electromagnetic environment time-frequency diagram B, the optimization process in step S4 is as follows:

[0052] In step S4, after denoteing the first intensity matrices of multiple target signal blocks as [M1, M2, M3, ...], six parameters are randomly varied: electromagnetic environment time-frequency diagram frequency variation, target signal block bandwidth variation, target signal block duration variation, random variation of target signal block gap, random variation of target signal block signal-to-noise ratio, and simulated variation of sudden interference enhancement and weakening of target signal block intensity. Furthermore, the random variations of the six parameters are performed sequentially, and are described below:

[0053] For the frequency change of the electromagnetic environment time-frequency diagram, a frequency range [f1,f2] is preset. Based on each target signal block marked in the target signal time-frequency diagram A, a frequency value f within this range is randomly selected in the electromagnetic environment time-frequency diagram B as the center frequency. Then, in the target signal time-frequency diagram A, the corresponding target signal block is shifted to the corresponding position centered on the frequency value f. The center frequencies of the remaining target signal blocks are selected and shifted in the same way.

[0054] For the target signal block bandwidth variation, a bandwidth range [b1, b2] is preset, and a value b is randomly selected from this range. The bandwidth of the corresponding target signal block is transformed to the value b by matrix pruning or matrix size transformation. The bandwidth of the remaining target signal blocks is processed in the same way.

[0055] For the duration variation of the target signal block, a preset time range [t1, t2] is used. The target signal block randomly selects a value t from this range. The duration of the corresponding target signal block is transformed into the value t by matrix shearing or matrix size transformation. The duration of the remaining target signal blocks is processed in the same way.

[0056] For the random variation of the gap between target signal blocks, a time gap range [c1, c2] is preset. When processing each target signal block marked in the target signal time-frequency diagram A in sequence, for each target signal block processed, a value c is randomly selected from this range. The time interval between the target signal block and the previous target signal block is transformed into the value c by matrix shearing, matrix size transformation or shifting the target signal block again. All target signal blocks are processed in the same way.

[0057] After the random changes of the first four parameters, the first intensity matrix [M1,M2,M3,...] of multiple target signal blocks is transformed into multiple intermediate intensity matrices [M1′,M2′,M3′,...], each with a dimension of [b,t]. Furthermore, the position of each target signal block in the target signal time-frequency diagram A is shifted relative to its original position.

[0058] For a target signal block with randomly varying signal-to-noise ratio (SNR), a preset SNR range [d1, d2] is defined. A value d is randomly selected from this range, and the SNR of the corresponding target signal block is transformed into the value d through matrix addition and subtraction operations. The single intermediate intensity matrix is ​​as follows:

[0059] M i ′=M i ′-mean(M i ′)+mean(E i )+d,i=1,2,3,...

[0060] In the formula, E i This indicates that the current location corresponds to the environmental signal intensity matrix in the electromagnetic environment time-frequency diagram B, where mean(M) i ′) represents M i The mean of E', mean(E) i ) represents E i The mean.

[0061] To simulate the sudden interferometric enhancement and weakening changes in the intensity of the target signal block, a Gaussian white noise with a mean of 0 and a standard deviation of σ is superimposed on each target signal block marked in the time-frequency diagram A of the target signal. This yields the second intensity matrix of a single target signal block as follows:

[0062]

[0063] In the formula, norm(...) represents a Gaussian distribution; therefore, after random variations of six parameters, the second intensity matrix of the multiple target signal blocks is denoted as follows:

[0064] Finally, the preferred process of step S5 is as follows: When the second intensity matrix of the target signal block is overlaid onto the corresponding position in the electromagnetic environment time-frequency diagram B, if the intensity of the target signal block at the overlaid position is lower than the corresponding electromagnetic environment signal intensity, the intensity value at the overlaid position is set to the original electromagnetic environment signal intensity, thereby updating the second intensity matrix of the target signal block in the electromagnetic environment time-frequency diagram B, which satisfies the following constraint:

[0065]

[0066] After the update is completed, a time-frequency diagram of the electromagnetic environment signal after being covered by the target signal block is obtained. The intensity matrix of this time-frequency diagram is the updated signal. Matrix; in the above formula, E i Both are intensity matrices of dimension [b,t]. The max operation compares the values ​​of elements at the same position in these two intensity matrices and takes the maximum value. The electromagnetic environment time-frequency map B after being covered by the target signal block is the new target signal time-frequency map generated by simulation.

[0067] This embodiment applies the above method, using a typical data link as the target signal for detection. In a test field, 100 time-frequency maps of the target signal under different working modes are collected and generated. At the same time, 1,000 time-frequency maps of the electromagnetic environment under different environments are collected and generated. After parametric simulation and random synthesis using the above method, a total of 20,000 time-frequency map training datasets for training the intelligent electromagnetic signal detection model are generated.

[0068] Verification has shown that the simulation method in this embodiment is highly efficient. Based on a typical PC environment (8-core Intel i7 processor), the generation time for a single new time-frequency map is less than 0.1 seconds, and the simulation process for each map is independent, thus possessing multi-process parallel acceleration capabilities. The generation time for the aforementioned 20,000 time-frequency maps is approximately 300 seconds. Furthermore, by generalizing the relevant detection model using the time-frequency map training dataset, the resulting inference model possesses the ability to accurately detect and identify the target in the data chain under new conditions, with a detection accuracy of no less than 90%.

[0069] In summary, the method proposed in this invention provides strong technical support for research and application in the field of electromagnetic signal processing. It not only helps reduce R&D costs but also accelerates the transformation of scientific research results into practical applications, thus possessing significant theoretical and practical value.

[0070] Example 2

[0071] Based on Example 1, this example applies its method to verify its effectiveness under specific scenarios and data conditions. For example... Figure 2 As shown, this embodiment uses a general-purpose acquisition receiver to acquire radio frequency signals and generates a time-frequency diagram by acquiring the signals at 50ms intervals; and deploys the simulation method of Embodiment 1 in a general PC environment (8-core Intel i7 processor) to simulate and generate time-frequency diagram data.

[0072] Using a typical data link signal as the target object, and combining various schematic diagrams, the effectiveness of the time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model proposed in this invention is verified as follows:

[0073] First, iterate through each time-frequency plot of the target signal in the target signal time-frequency plot dataset, such as... Figure 3 The image shown is one of the time-frequency diagrams; the target signal blocks are manually labeled on this time-frequency diagram, and the labeling results are as follows. Figure 4 As shown; the labeled signal blocks are used for simulation target signal extraction, thereby obtaining the first intensity matrix of the target signal block. A schematic diagram of the extracted target signal is shown below. Figure 5 As shown.

[0074] Next, six parameters are randomly varied on the first intensity matrix of the target signal block extracted from the simulated target signal to change its frequency hopping signal pattern, increase noise power, and change the signal-to-noise ratio, thereby generating simulated target signal blocks under different conditions; the generation results are as follows. Figure 6 As shown.

[0075] For a time-frequency graph that is also randomly selected from the electromagnetic environment time-frequency graph dataset, such as Figure 7 As shown, after integrating the simulated target signal block with it, the resulting simulated time-frequency diagram is as follows. Figure 8As shown in the figure. By repeating the above process continuously, 20,000 time-frequency maps were finally obtained from the simulation.

[0076] Simulated time-frequency maps were used to train a deep neural network, and the trained inference model was used to detect and identify similar target signals under different environments. Experiments showed that the neural network also achieved good recognition results on real signal time-frequency maps collected from different regions. Figure 9 The rectangle shown in the image.

Claims

1. A time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model, characterized in that, The method includes the following steps: S1. Obtain the target signal time-frequency map dataset and the electromagnetic environment time-frequency map dataset; S2. In the target signal time-frequency map dataset, label the target signal blocks of all target signal time-frequency maps and store the labeling information; S3. Randomly select one time-frequency map data from the target signal time-frequency map dataset and the electromagnetic environment time-frequency map dataset, and denot them as the labeled target signal time-frequency map A and electromagnetic environment time-frequency map B, respectively. S4. Based on the target signal time-frequency map A and its annotation information, extract the first intensity matrix of the target signal block in the target signal time-frequency map A, and perform various random parameter changes on the first intensity matrix of the target signal block. S5. After the random changes of multiple parameters are completed, the second intensity matrix of the target signal block in the target signal time-frequency map A is obtained; the second intensity matrix of the target signal block is overlaid on the corresponding position in the electromagnetic environment time-frequency map B, and the overlaid electromagnetic environment time-frequency map B is the new target signal time-frequency map generated by simulation. S6. Continue the random selection, random variation and coverage process from steps S3 to S5 until the number of new target signal time-frequency diagrams generated by the simulation meets the model training requirements, thereby completing the time-frequency diagram simulation process.

2. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 1, characterized in that, In step S1, the target signal time-frequency map dataset is obtained by: acquiring multiple target signal time-domain waveforms, converting them into multiple target signal time-frequency maps through short-time Fourier transform, thereby forming the target signal time-frequency map dataset; in the target signal time-frequency map dataset, the dimension of the target signal time-frequency map is [N,M], where N is the number of spectral frames and M is the frame length.

3. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 2, characterized in that, In step S1, the electromagnetic environment time-frequency map dataset is obtained by collecting multiple typical electromagnetic environment data without a target signal, converting them into multiple electromagnetic environment time-frequency maps through short-time Fourier transform, thereby forming an electromagnetic environment time-frequency map dataset. In the electromagnetic environment time-frequency map dataset, the dimension of the electromagnetic environment time-frequency map is the same as the dimension of the target signal time-frequency map, which is [N,M].

4. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 1, characterized in that: In step S2, target signal blocks in the target signal time-frequency map are labeled using rectangular boxes. The coordinates of the upper left and lower right corners of the rectangular boxes are used as labeling information and stored. After the target signal time-frequency map dataset is labeled, the labeling information for each target signal time-frequency map is recorded as: [(x1 min ,y1 min ,x1 max ,y1 max ),(x2 min ,y2 min x2 max ,y2 max ),...].

5. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 1, characterized in that: In step S4, after the first intensity matrices of multiple target signal blocks are denoted as [M1, M2, M3, ...], six parameters are randomly varied: electromagnetic environment time-frequency diagram frequency variation, target signal block bandwidth variation, target signal block duration variation, random variation of target signal block gap, random variation of target signal block signal-to-noise ratio, and simulated variation of sudden interference enhancement and weakening of target signal block intensity.

6. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 5, characterized in that: The six parameters are randomly changed in sequence. For the frequency change of the electromagnetic environment time-frequency diagram, a frequency range [f1, f2] is preset. According to each target signal block marked in the target signal time-frequency diagram A, a frequency value f within this range is randomly selected in the electromagnetic environment time-frequency diagram B as the center frequency. Then, in the target signal time-frequency diagram A, the corresponding target signal block is translated to the corresponding position centered on the frequency value f. The center frequency of the remaining target signal blocks is selected and translated in the same way. For the target signal block bandwidth variation, a bandwidth range [b1, b2] is preset, and a value b is randomly selected from this range. The bandwidth of the corresponding target signal block is transformed to the value b by matrix pruning or matrix size transformation. The bandwidth of the remaining target signal blocks is processed in the same way.

7. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 6, characterized in that: For the duration variation of the target signal block, a time range [t1, t2] is preset. The target signal block randomly selects a value t from this range. The duration of the corresponding target signal block is transformed into the value t by matrix shearing or matrix size transformation. The duration of the remaining target signal blocks is processed in the same way. For the random variation of the gap between target signal blocks, a time gap range [c1, c2] is preset. When processing each target signal block marked in the target signal time-frequency diagram A in sequence, for each target signal block processed, a value c is randomly selected from this range. The time interval between the target signal block and the previous target signal block is transformed into the value c by matrix shearing, matrix size transformation or shifting the target signal block again. All target signal blocks are processed in the same way. After the random changes of the first four parameters, the first intensity matrix [M1,M2,M3,...] of multiple target signal blocks is transformed into multiple intermediate intensity matrices [M1′,M2′,M3′,...], each with a dimension of [b,t]. Furthermore, the position of each target signal block in the target signal time-frequency diagram A is shifted relative to its original position.

8. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 7, characterized in that: For a target signal block with randomly varying signal-to-noise ratio (SNR), a preset SNR range [d1, d2] is defined. A value d is randomly selected from this range, and the SNR of the corresponding target signal block is transformed into the value d through matrix addition and subtraction operations. The single intermediate intensity matrix is ​​as follows: M i ′=M i ′-mean(M i ′)+mean(E i )+d,i=1,2,3,... In the formula, E i This indicates that the current location corresponds to the environmental signal intensity matrix in the electromagnetic environment time-frequency diagram B, where mean(M) i ′) represents M i The mean of E', mean(E) i ) represents E i The mean.

9. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 8, characterized in that: To simulate the sudden interferometric enhancement and weakening changes in the intensity of the target signal block, a Gaussian white noise with a mean of 0 and a standard deviation of σ is superimposed on each target signal block marked in the time-frequency diagram A of the target signal. This yields the second intensity matrix of a single target signal block as follows: In the formula, norm(...) represents a Gaussian distribution; therefore, after random variations of six parameters, the second intensity matrix of the multiple target signal blocks is denoted as follows:

10. The time-frequency diagram simulation method for training an intelligent electromagnetic signal detection model according to claim 9, characterized in that: In step S5, when the second intensity matrix of the target signal block is overlaid onto the corresponding position in the electromagnetic environment time-frequency diagram B, if the intensity of the target signal block at the overlaid position is lower than the corresponding electromagnetic environment signal intensity, the intensity value at that overlaid position is set to the original electromagnetic environment signal intensity, thereby updating the second intensity matrix of the target signal block in the electromagnetic environment time-frequency diagram B, which satisfies the following constraint: After the update is completed, a time-frequency map of the electromagnetic environment signal after being covered by the target signal block is obtained. The intensity matrix of this time-frequency map is the updated signal. Matrix; in the above formula, E i Both are intensity matrices of dimension [b,t]. The max operation compares the values ​​of elements at the same position in these two intensity matrices and takes the maximum value. The electromagnetic environment time-frequency map B after being covered by the target signal block is the new target signal time-frequency map generated by simulation.

Images