A model training and signal detection method, device, medium and electronic equipment

Abstract

Embodiments of the present application disclose a model training and signal detection method and device, a medium and an electronic device, relating to the technical field of signal processing. The present application trains a signal detection model based on deep learning to realize signal detection. The detection is performed through image recognition to avoid the influence of electromagnetic environment and signal-to-noise ratio, so as to improve the correctness of signal detection and recognition. Data describing signals I / Q, i.e. in-phase component data and quadrature component data describing signal phase and amplitude, are collected. The data are segmented into multiple frames, so that each segmented signal is an approximately stationary signal, facilitating spectrum analysis. Then, the data are converted into frequency domain data by using fast Fourier transform and sequentially stored in a buffer, a data matrix is obtained, and a spectrum diagram is generated by using the data matrix. The spectrum diagram obtained by linear mapping is trained. Since signal information is labeled on the spectrum diagram, the signal detection model obtained by training can extract image features and realize fast and accurate recognition of signals based on images in the application stage. After the features of the signals are recognized on the image, the features are mapped into signal parameters, so that the detection and recognition of the signals are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal processing, in particular to a model training and signal detection method and device, medium and electronic equipment. BACKGROUND

[0002] Radar, short for Radio Detection and Ranging, is an electronic device that uses electromagnetic waves to detect targets. It transmits electromagnetic waves to illuminate targets and receives their echoes, thereby obtaining information such as the distance, distance rate of change (radial velocity), azimuth, and altitude of the target from the electromagnetic wave transmission point. The typical frequency band occupied by radar signals is from 500 megahertz to 18 gigahertz. It is a technology signal that uses electromagnetic waves for target detection and tracking. Traditional radar signal analysis is usually based on intra-pulse modulation recognition algorithms in the time and frequency domains to identify and estimate parameters, which can solve most intra-pulse modulation recognition problems. However, the accuracy of detection still has room for improvement. For example, in a complex electromagnetic environment, electromagnetic interference can affect the accuracy of recognition. When the signal-to-noise ratio is low, the correctness of parameter demodulation will also decrease. SUMMARY

[0003] The main purpose of the present application is to provide a model training and signal detection method, device, medium and electronic equipment, aiming to solve the problem of reduced correctness of radar signal detection and recognition in the prior art.

[0004] To achieve the above-mentioned purpose, the technical solutions adopted by the embodiments of the present application are as follows:

[0005] In a first aspect, the embodiments of the present application provide a model training method, comprising the following steps:

[0006] The sample in-phase component data and the sample quadrature component data are segmented into multiple frames to obtain first sample in-phase component data and first sample quadrature component data;

[0007] The first sample in-phase component data and the first sample quadrature component data are subjected to fast Fourier transform to obtain frequency domain data;

[0008] The frequency domain data is sequentially stored in a data buffer to obtain a data matrix;

[0009] The points of the data matrix are linearly mapped to color information to obtain a spectrogram;

[0010] Signal information is labeled on the spectrogram to obtain a training sample image;

[0011] Based on the training sample image, a signal detection model based on deep learning is trained.

[0012] In a possible implementation manner of the first aspect, the sample in-phase component data and the sample quadrature component data are subjected to multi-frame segmentation to obtain first sample in-phase component data and first sample quadrature component data, including:

[0013] The sample in-phase component data and the sample quadrature component data are subjected to multi-frame segmentation and overlap processing to obtain the first sample in-phase component data and the first sample quadrature component data.

[0014] In a possible implementation manner of the first aspect, after the sample in-phase component data and the sample quadrature component data are subjected to multi-frame segmentation to obtain the first sample in-phase component data and the first sample quadrature component data, the model training method further includes:

[0015] The first sample in-phase component data and the first sample quadrature component data are subjected to windowing processing to obtain second sample in-phase component data and second sample quadrature component data.

[0016] The first sample in-phase component data and the first sample quadrature component data are subjected to fast Fourier transform to obtain frequency domain data, including:

[0017] The second sample in-phase component data and the second sample quadrature component data are subjected to fast Fourier transform to obtain frequency domain data.

[0018] In a possible implementation manner of the first aspect, before the points of the data matrix are linearly mapped to color information to obtain the spectrum diagram, the model training method further includes:

[0019] A judgment result is obtained by judging a size relationship between a height of the data matrix and a height threshold.

[0020] The points of the data matrix are linearly mapped to color information to obtain the spectrum diagram, including:

[0021] The points of the data matrix are linearly mapped to color information, and the spectrum diagram is obtained according to the judgment result.

[0022] In a second aspect, an embodiment of the present application provides a signal detection method, including the following steps:

[0023] The in-phase component data and the quadrature component data of the target signal are subjected to multi-frame segmentation to obtain first in-phase component data and first quadrature component data.

[0024] The first in-phase component data and the first quadrature component data are subjected to fast Fourier transform to obtain target frequency domain data.

[0025] The target frequency domain data is sequentially stored in a data buffer to obtain a target data matrix.

[0026] The points of the target data matrix are linearly mapped to color information to obtain a target spectrum diagram.

[0027] The target spectral image is input into the signal detection model to obtain the output result for calculating the signal parameters of the target signal; wherein the signal detection model is obtained according to the model training method provided in any of the first aspects above.

[0028] In one possible implementation of the second aspect, the target spectral image is input into a signal detection model to obtain the output result for calculating the signal parameters of the target signal, including:

[0029] Input the target spectrum into the signal detection model to obtain the output result;

[0030] The output is mapped to time and frequency information, and the signal parameters of the target signal are calculated based on the time and frequency information.

[0031] Thirdly, embodiments of this application provide a model training apparatus, comprising:

[0032] The first segmentation module is used to segment the in-phase component data and the quadrature component data of the samples into multiple frames to obtain the first in-phase component data and the first quadrature component data of the samples.

[0033] The first transformation module is used to perform a fast Fourier transform on the first sample in-phase component data and the first sample quadrature component data to obtain frequency domain data.

[0034] The first matrix module is used to sequentially store frequency domain data into a data buffer to obtain a data matrix.

[0035] The first mapping module is used to linearly map the points of the data matrix into color information to obtain a spectral map;

[0036] The annotation module is used to annotate signal information on the spectral image to obtain training sample images;

[0037] The training module is used to train a deep learning-based signal detection model based on training sample images.

[0038] Fourthly, embodiments of this application provide a signal detection device, comprising:

[0039] The second segmentation module is used to segment the in-phase component data and quadrature component data of the target signal into multiple frames to obtain the first in-phase component data and the first quadrature component data.

[0040] The second transformation module is used to perform a fast Fourier transform on the first in-phase component data and the first quadrature component data to obtain the target frequency domain data.

[0041] The second matrix module is used to sequentially store the target frequency domain data into the data buffer to obtain the target data matrix.

[0042] The second mapping module is used to linearly map the points of the target data matrix into color information to obtain the target spectral map;

[0043] The detection module is used to input the target spectrum into the signal detection model and obtain the output result to calculate the signal parameters of the target signal; wherein the signal detection model is obtained according to the model training method provided in any of the first aspects above.

[0044] Fifthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when loaded and executed by a processor, implements a model training method as provided in any of the first aspects above or a signal detection method as provided in any of the second aspects above.

[0045] Sixthly, embodiments of this application provide an electronic device, including a processor and a memory, wherein,

[0046] Memory is used to store computer programs;

[0047] The processor is used to load and execute computer programs to cause the electronic device to perform a model training method as provided in any of the first aspects above or a signal detection method as provided in any of the second aspects above.

[0048] Compared with the prior art, the beneficial effects of this application are:

[0049] This application proposes a model training and signal detection method, apparatus, medium, and electronic device. The training method includes: segmenting sample in-phase component data and sample quadrature component data into multiple frames to obtain first sample in-phase component data and first sample quadrature component data; performing a fast Fourier transform on the first sample in-phase component data and the first sample quadrature component data to obtain frequency domain data; storing the frequency domain data sequentially into a data buffer to obtain a data matrix; linearly mapping the points of the data matrix to color information to obtain a spectrum; labeling signal information on the spectrum to obtain training sample images; and training a deep learning-based signal detection model based on the training sample images. This application trains a deep learning-based signal detection model to detect signals. It uses image recognition to avoid the influence of electromagnetic environment and signal-to-noise ratio, thereby improving the accuracy of signal detection and recognition. By collecting data describing the signal's I / Q, namely the in-phase and quadrature components describing the signal's phase and amplitude, this data is segmented into multiple frames, making each frame approximately a stationary signal for easier spectral analysis. Then, a fast Fourier transform is used to convert the data into frequency domain data, which is sequentially stored in a buffer to obtain a data matrix. This matrix is ​​then used to generate a spectrum. The spectrum obtained under linear mapping is used for training. Since the spectrum is labeled with signal information, the trained signal detection model can extract image features and achieve fast and accurate signal recognition based on images in the application stage. After identifying signal features in the image, these features are mapped back to signal parameters to achieve signal detection and recognition. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of this application;

[0051] Figure 2 A schematic flowchart illustrating the model training method provided in this application embodiment;

[0052] Figure 3 A schematic diagram illustrating multi-frame segmentation of data in the model training method provided in this application embodiment;

[0053] Figure 4 A schematic diagram illustrating the mapping via a data matrix in the model training method provided in this application embodiment;

[0054] Figure 5 A flowchart illustrating one implementation of the model training method provided in this application;

[0055] Figure 6 A schematic flowchart illustrating the signal detection method provided in an embodiment of this application;

[0056] Figure 7 This is a schematic diagram illustrating the mapping and acquisition of parameter information in the signal detection method provided in the embodiments of this application.

[0057] Figure 8 A flowchart illustrating one implementation of the signal detection method provided in this application;

[0058] Figure 9 A schematic diagram of the module of the model training device provided in the embodiments of this application;

[0059] Figure 10 This is a schematic diagram of the module of the signal detection device provided in the embodiments of this application;

[0060] The diagram is labeled as follows: 101-Processor, 102-Communication bus, 103-Network interface, 104-User interface, 105-Memory. Detailed Implementation

[0061] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0062] See attached document Figure 1 , attached Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of this application. The electronic device may include: a processor 101, such as a central processing unit (CPU), a communication bus 102, a user interface 104, a network interface 103, and a memory 105. The communication bus 102 is used to realize the connection and communication between these components. The user interface 104 may include a display screen and an input unit such as a keyboard. Optionally, the user interface 104 may also include a standard wired interface and a wireless interface. The network interface 103 may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory 105 may be a storage device independent of the aforementioned processor 101. The memory 105 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as at least one disk storage device. The processor 101 may be a general-purpose processor, including a central processing unit, a network processor, etc., or it may be a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component.

[0063] Those skilled in the art will understand that the appendix Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0064] As attached Figure 1 As shown, the memory 105, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a model training device or a signal detection device.

[0065] In the appendix Figure 1 In the electronic device shown, the network interface 103 is mainly used for data communication with the network server; the user interface 104 is mainly used for data interaction with the user; the processor 101 and the memory 105 in this application can be set in the electronic device. The electronic device calls the model training device or signal detection device stored in the memory 105 through the processor 101 and executes the model training method or signal detection method provided in the embodiments of this application.

[0066] See attached document Figure 2 Based on the hardware device described in the foregoing embodiments, embodiments of this application provide a model training method, including the following steps:

[0067] S10: Perform multi-frame segmentation on the in-phase component data and the quadrature component data of the samples to obtain the first in-phase component data and the first quadrature component data of the samples.

[0068] In the specific implementation process, for non-stationary signals, the signal spectrum may be unstable over a long period. After dividing the sample data into equal multi-frame segments, the signal within each frame approximates a stationary signal, facilitating spectrum analysis. The sample data consists of the I and Q data of the sample baseband, which can be considered as the original digital signal data, where I represents the in-phase component and Q represents the quadrature component. These are used as the sample in-phase component data and sample quadrature component data, and divided into multiple frames. The segmented data are denoted as the first sample in-phase component data and the first sample quadrature component data, respectively. Specifically: the sample in-phase component data and sample quadrature component data are divided into multiple frames to obtain the first sample in-phase component data and the first sample quadrature component data, including:

[0069] The in-phase component data and the quadrature component data of the samples are segmented into multiple frames and overlapped to obtain the first in-phase component data and the first quadrature component data of the samples.

[0070] In the specific implementation process, the data is segmented into identical f-frames. To improve frequency resolution and reduce spectral leakage, overlap processing is performed during data segmentation, as shown in the attached figure. Figure 3The diagram shows a 50% overlap processing method, where the starting frame of the segmented data is the middle frame of the previous segment, and the ending frame is the middle frame of the next segment. In other embodiments, the overlap ratio can be adjusted according to actual needs. The data after overlap processing are denoted as the first sample in-phase component data and the first sample quadrature component data, respectively. The signal is divided into multiple overlapping segments, and each time-domain segment can be independently transformed by FFT to obtain the frequency domain information corresponding to that segment. This captures the signal's time and frequency variation characteristics to a certain extent, allowing for better observation of the dynamic evolution of the signal's frequency components over time.

[0071] S20: Perform a fast Fourier transform on the in-phase component data and the quadrature component data of the first sample to obtain the frequency domain data.

[0072] In practical implementation, the Fast Fourier Transform (FFT) is an efficient algorithm for the Discrete Fourier Transform (DFT) and one of the most fundamental methods in time-domain to frequency-domain transformation analysis. By converting the signal from the time domain to the frequency domain using the Fast Fourier Transform, it decomposes the signal into its constituent frequencies, thereby enabling better analysis, synthesis, and processing of the signal.

[0073] In one embodiment, after performing multi-frame segmentation on the in-phase component data and the quadrature component data to obtain the first in-phase component data and the first quadrature component data, the model training method further includes:

[0074] Windowing is applied to the first sample in-phase component data and the first sample orthogonal component data to obtain the second sample in-phase component data and the second sample orthogonal component data.

[0075] In practical implementation, dividing the data into multiple segments and directly performing FFT transformation may result in spectral leakage. Windowing can mitigate the truncation effect and suppress spectral leakage. Windowing is a signal processing technique that, by selecting appropriate window functions and parameters, can effectively extract feature information from signals, improving the accuracy and effectiveness of signal processing. In practical applications, the windowing method should be flexibly selected and used based on specific signal characteristics and processing requirements. Its main functions are to alleviate the Gibbs effect, reduce leakage, and improve the accuracy of processing results. The windowed data is denoted as the second sample in-phase component data and the second sample quadrature component data.

[0076] Based on the aforementioned steps, a Fast Fourier Transform is performed on the in-phase component data and the quadrature component data of the first sample to obtain frequency domain data, including:

[0077] Perform a Fast Fourier Transform on the in-phase component data of the second sample and the quadrature component data of the second sample to obtain the frequency domain data.

[0078] S30: Store the frequency domain data into the data buffer sequentially to obtain the data matrix.

[0079] In the specific implementation process, the aforementioned steps involve performing an m-point Fast Fourier Transform, which means converting the m data points into frequency domain data through hyperspectral analysis. This data is then sequentially stored in a data buffer, which is a data matrix of size f*m. This data matrix in the buffer is used to generate a waterfall plot, also called a spectral plot. A waterfall plot is an intuitive way to display how signal frequency changes over time. It helps to quickly identify frequency drift, modulation patterns, or other frequency characteristics, extracting meaningful frequency information from time-domain signals to support signal analysis and detection. It also provides visualization tools to aid in understanding signal characteristics.

[0080] S40: Linearly map the points of the data matrix to color information to obtain a spectral map.

[0081] In the specific implementation process, each point of the data matrix is ​​linearly mapped to color information, resulting in one or more spectral maps, also known as time-frequency maps. The number of spectral maps obtained through mapping can be determined based on the height of the matrix and a preset height threshold. That is, before linearly mapping the points of the data matrix to color information to obtain the spectral maps, the model training method also includes:

[0082] Determine the relationship between the height of the data matrix and the height threshold to obtain the determination result.

[0083] The points in the data matrix are linearly mapped to color information to obtain a spectral image, including:

[0084] The points in the data matrix are linearly mapped to color information, and a spectral image is obtained based on the judgment result.

[0085] In this implementation, the height threshold is set to 992. If the height of the data matrix is ​​not greater than 992, the data matrix is ​​directly mapped to a single spectral image. If the height is greater than 992, the data matrix is ​​split into multiple spectral images. Adjacent spectral images have a certain degree of overlap in the Y-axis direction, as shown in the attached figure. Figure 4 The diagram illustrates one method for mapping to obtain a spectral image. Specifically, when the height of the data matrix is ​​greater than 992, the overlap is set to 30% to obtain multiple spectral images through mapping.

[0086] S50: Mark signal information on the spectrum to obtain training sample images.

[0087] In the specific implementation process, the spectral image is saved as a bitmap as training samples. For example, in the implementation method described above, the spectral image is saved as a 992*992 bitmap, which is an image that deep learning can recognize. N sets of sample data are collected as needed and saved as images, where the N sets of sample data contain radar signals with different modulation types. Signal information is labeled on the spectral image, that is, after the spectral image is saved as a bitmap, labeling is performed. The labeling information includes signal type, center coordinates, and signal width and height. After labeling, the training sample images can be divided into training and test sets proportionally, for example, a 9:1 ratio.

[0088] S60: Based on training sample images, a deep learning-based signal detection model is trained.

[0089] In the specific implementation process, deep learning is used for training. Existing model architectures can be utilized, such as the YOLOv8s model, a typical object detection model. Its overall architecture consists of a base network, a feature extraction layer, a prediction layer, and an output layer. The base network is mainly responsible for extracting low-level image features. The feature extraction layer is used to further extract higher-level semantic features. The prediction layer is responsible for predicting the location and category of the target. The output layer outputs the final detection result. Based on the training results, parameters such as the training cycle and pre-training weights are adjusted until the model converges on the training samples, resulting in the trained model YOLOModel, which is the signal detection model.

[0090] In this embodiment, a deep learning-based signal detection model is trained to detect signals. Image recognition is used to avoid the influence of electromagnetic environment and signal-to-noise ratio, thereby improving the accuracy of signal detection and recognition. Data describing the I / Q of the signal, namely the in-phase component data and quadrature component data describing the phase and amplitude of the signal, are collected. This data is divided into multiple frames, making each frame signal approximately a stationary signal, which facilitates spectrum analysis. Then, the data is converted into frequency domain data using fast Fourier transform and stored sequentially in a buffer to obtain a data matrix. The matrix is ​​then used to generate a spectrum. The spectrum obtained under linear mapping is used for training. Since the signal information is marked on the spectrum, the trained signal detection model can extract image features and achieve fast and accurate signal recognition based on images in the application stage. After identifying the signal features on the image, they are mapped back to signal parameters to achieve signal detection and recognition.

[0091] See attached document Figure 5 In the appendix Figure 5 The model training method provided in this application will be further explained in the following embodiments:

[0092] First, sample I and Q data are obtained. Then, they are segmented into f frames and converted into frequency domain data through FFT. The data is then stored in the data buffer to obtain a data matrix. The points of the data matrix are then linearly mapped to color information to obtain a spectral map. The spectral map is saved as a bitmap and then labeled and segmented to form training sample images. Finally, these images are input into YOLOv8s for iterative training. After training is completed, the training model is exported, and the signal detection model can be used to implement signal detection.

[0093] See attached document Figure 6 Based on the same inventive concept as in the foregoing embodiments, this application also provides a signal detection method, including the following steps:

[0094] S100: Divide the in-phase component data and quadrature component data of the target signal into multiple frames to obtain the first in-phase component data and the first quadrature component data;

[0095] S200: Perform a fast Fourier transform on the first in-phase component data and the first quadrature component data to obtain the target frequency domain data;

[0096] S300: Store the target frequency domain data into the data buffer sequentially to obtain the target data matrix;

[0097] S400: Linearly map the points of the target data matrix to color information to obtain the target spectral map;

[0098] S500: Input the target spectrum into the signal detection model to obtain the output result and calculate the signal parameters of the target signal; wherein, the signal detection model is obtained according to the model training method provided in the embodiments of this application.

[0099] In specific implementation, the beneficial effects of the embodiments of this application can be referred to the foregoing embodiments. In the application stage of the model, the signal detection process before inputting the target spectrum into the signal detection model is similar to the model training process. The target signal is the signal to be detected, and the IQ signal to be detected with a collection time length of Ts, namely the in-phase component data and the quadrature component data, is divided into f frames of data according to the model training process. After fast Fourier transform, the data is stored in the data buffer and mapped into a spectrum, namely the target spectrum. During this process, the data can also be overlapped and windowed according to the implementation method of model training. At the same time, it is necessary to record the start time tf of each frame of data in the segmented f frames of data. The calculation method is tf = i*Ts / fi = 0, 1, 2...f.

[0100] The signal detection model obtained by calling the training method provided in the embodiments of this application is used for inference to obtain a prediction result set. It should be noted that the target spectral image here also needs to be saved as a bitmap that the model can recognize for input inference. The result set includes the category of each detected signal, the width and height of the signal box, the pixel offset of the center of the signal box on the x and y axes, and the prediction confidence. The pixel information of each prediction result in the result set is mapped to the parameter information of the signal, and the mapping method is as shown in the appendix. Figure 7 As shown. The parameter information includes center frequency, signal bandwidth, signal pulse, and pulse interval. Given the center frequency centFreq of the acquired IQ data, the analysis bandwidth span, and the acquisition time Tm during the measurement process, the center frequency SigcentFreq of the signal can be calculated using the following formula:

[0101]

[0102] Where x is the appended Figure 7 The x-axis coordinates of the center point of the signal box 1 (the entire image can be regarded as a coordinate axis region, the upper left corner of the image represents the origin of the coordinate axis, and the coordinates of the lower right corner of the image are (img.Width, img.Height), the unit of the coordinate axis is pixels), img.Width represents the image width of the target spectral map, the unit is pixels.

[0103] The signal start time SigStartTime can be obtained using the following formula:

[0104]

[0105] Where y is the appendix Figure 7 The y-axis coordinate of the center point of the signal box 1 is given by result.Height represents the height of the signal box 1, which is SigHeight.img.Height represents the image height of the target spectral map, in pixels.

[0106] Similarly, the frequency and time information corresponding to the four corners of the signal frame can be obtained, and then the signal parameters can be calculated. That is, the target spectrum is input into the signal detection model, and the output results are used to calculate the signal parameters of the target signal, including:

[0107] Input the target spectrum into the signal detection model to obtain the output result;

[0108] The output is mapped to time and frequency information, and the signal parameters of the target signal are calculated based on the time and frequency information.

[0109] As attached Figure 7As shown, the signal pulse width is the frequency difference corresponding to the width of the signal frame, the signal duration is the time difference corresponding to the height of the signal frame, the signal interval is the time difference corresponding to the edges of two adjacent signal frames, and (x, y) are the coordinates of the center point of signal frame 1, which can be output by the detection model. SigWidth and WigHeight represent the width and height of signal frame 1, respectively. Width and Height represent the width and height of the target spectral map, which are img.Width and img.Height in the formula above.

[0110] See attached document Figure 8 In the appendix Figure 8 The signal detection method provided in this application will be further described in the embodiments shown below:

[0111] First, the I and Q data are measured. Following a similar processing method to model training, the data is divided into f frames and converted into frequency domain data using FFT. These frames are then stored sequentially in the data buffer to obtain a data matrix. The start time of each frame is recorded. The matrix height, i.e., the number of rows, is then determined. A threshold of 992 is set, and the number of rows is checked to see if it is greater than 992. If not, the data matrix is ​​directly mapped to a single spectral image. If it is, the matrix is ​​split and mapped to multiple spectral images. Then, the YOLO model is used for inference and prediction, which is the signal detection model obtained by the training method provided in this application embodiment. The output result is obtained and mapped to time and frequency. Signal parameters are calculated to complete the recognition. Finally, the recognition result is displayed in real time using a waterfall plot.

[0112] See attached document Figure 9 Based on the same inventive concept as in the foregoing embodiments, this application also provides a model training apparatus, including:

[0113] The first segmentation module is used to segment the in-phase component data and the quadrature component data of the samples into multiple frames to obtain the first in-phase component data and the first quadrature component data of the samples.

[0114] The first transformation module is used to perform a fast Fourier transform on the first sample in-phase component data and the first sample quadrature component data to obtain frequency domain data.

[0115] The first matrix module is used to sequentially store frequency domain data into a data buffer to obtain a data matrix.

[0116] The first mapping module is used to linearly map the points of the data matrix into color information to obtain a spectral map;

[0117] The annotation module is used to annotate signal information on the spectral image to obtain training sample images;

[0118] The training module is used to train a deep learning-based signal detection model based on training sample images.

[0119] See attached document Figure 10 Based on the same inventive concept as in the foregoing embodiments, this application also provides a signal detection device, including:

[0120] The second segmentation module is used to segment the in-phase component data and quadrature component data of the target signal into multiple frames to obtain the first in-phase component data and the first quadrature component data.

[0121] The second transformation module is used to perform a fast Fourier transform on the first in-phase component data and the first quadrature component data to obtain the target frequency domain data.

[0122] The second matrix module is used to sequentially store the target frequency domain data into the data buffer to obtain the target data matrix.

[0123] The second mapping module is used to linearly map the points of the target data matrix into color information to obtain the target spectral map;

[0124] The detection module is used to input the target spectrum into the signal detection model and obtain the output result to calculate the signal parameters of the target signal; wherein, the signal detection model is obtained according to the model training method provided in the embodiments of this application.

[0125] Those skilled in the art should understand that the division of the various modules in the embodiments is merely a logical functional division. In actual applications, they can be fully or partially integrated onto one or more actual carriers. These modules can be implemented entirely in software through processing unit calls, entirely in hardware, or a combination of software and hardware. It should be noted that each module in the model training device and signal detection device in this embodiment corresponds one-to-one with each step in the model training method and signal detection method in the aforementioned embodiments. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned embodiments, and will not be repeated here.

[0126] Based on the same inventive concept as in the foregoing embodiments, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when loaded and executed by a processor, implements the model training method or signal detection method provided in the embodiments of this application.

[0127] Based on the same inventive concept as in the foregoing embodiments, embodiments of this application also provide an electronic device, including a processor and a memory, wherein,

[0128] Memory is used to store computer programs;

[0129] The processor is used to load and execute computer programs to enable the electronic device to perform model training methods or signal detection methods as provided in the embodiments of this application.

[0130] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a device including one or any combination of the above-mentioned memories. The computer may be a variety of computing devices, including smart terminals and servers.

[0131] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0132] As an example, executable instructions may, but do not necessarily, correspond to files in the file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0133] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0134] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0135] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0136] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a multimedia terminal device (which may be a mobile phone, computer, television receiver, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0137] In summary, this application provides a model training and signal detection method, apparatus, medium, and electronic device. The training method includes: segmenting sample in-phase component data and sample quadrature component data into multiple frames to obtain first sample in-phase component data and first sample quadrature component data; performing a fast Fourier transform on the first sample in-phase component data and the first sample quadrature component data to obtain frequency domain data; storing the frequency domain data sequentially into a data buffer to obtain a data matrix; linearly mapping the points of the data matrix to color information to obtain a spectrum; labeling signal information on the spectrum to obtain training sample images; and training a deep learning-based signal detection model based on the training sample images. This application trains a deep learning-based signal detection model to detect signals. It uses image recognition to avoid the influence of electromagnetic environment and signal-to-noise ratio, thereby improving the accuracy of signal detection and recognition. By collecting data describing the signal's I / Q, namely the in-phase and quadrature components describing the signal's phase and amplitude, this data is segmented into multiple frames, making each frame approximately a stationary signal for easier spectral analysis. Then, a fast Fourier transform is used to convert the data into frequency domain data, which is sequentially stored in a buffer to obtain a data matrix. This matrix is ​​then used to generate a spectrum. The spectrum obtained under linear mapping is used for training. Since the spectrum is labeled with signal information, the trained signal detection model can extract image features and achieve fast and accurate signal recognition based on images in the application stage. After identifying signal features in the image, these features are mapped back to signal parameters to achieve signal detection and recognition.

[0138] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A model training method, characterized in that, For the scene where electromagnetic interference or signal-to-noise effect exists, the following steps are included: The sample in-phase component data and the sample quadrature component data are subjected to multi-frame segmentation to obtain first sample in-phase component data and first sample quadrature component data; The first sample in-phase component data and the first sample quadrature component data are subjected to fast Fourier transform to obtain frequency domain data; The frequency domain data is sequentially stored in a data buffer to obtain a data matrix; The points of the data matrix are linearly mapped to color information to obtain a spectrum graph; Signal information is labeled on the spectrum graph to obtain a training sample image; the labeling of signal information on the spectrum graph to obtain a training sample image includes: The spectrum graph is saved as a bitmap and signal information is labeled on the bitmap to obtain a training sample image; Based on the training sample image, a signal detection model based on deep learning is trained; the signal detection model can extract image features of an input bitmap and output a signal recognition result.

2. The model training method of claim 1, wherein, The sample in-phase component data and the sample quadrature component data are subjected to multi-frame segmentation to obtain first sample in-phase component data and first sample quadrature component data, which includes: The sample in-phase component data and the sample quadrature component data are subjected to multi-frame segmentation and overlap processing to obtain first sample in-phase component data and first sample quadrature component data.

3. The model training method of claim 1, wherein, After the sample in-phase component data and the sample quadrature component data are subjected to multi-frame segmentation to obtain first sample in-phase component data and first sample quadrature component data, the model training method further includes: The first sample in-phase component data and the first sample quadrature component data are subjected to windowing processing to obtain second sample in-phase component data and second sample quadrature component data; The first sample in-phase component data and the first sample quadrature component data are subjected to fast Fourier transform to obtain frequency domain data, which includes: The second sample in-phase component data and the second sample quadrature component data are subjected to fast Fourier transform to obtain frequency domain data.

4. The model training method of claim 1, wherein, Before the points of the data matrix are linearly mapped to color information to obtain a spectrum graph, the model training method further includes: A judgment result is obtained by judging the size relationship between the height of the data matrix and a height threshold; The points of the data matrix are linearly mapped to color information to obtain a spectrum graph, which includes: The points of the data matrix are linearly mapped to color information, and a spectrum graph is obtained according to the judgment result.

5. A signal detection method characterized by, The following steps are included: The in-phase component data and the quadrature component data of the target signal are subjected to multi-frame segmentation to obtain first in-phase component data and first quadrature component data; The first in-phase component data and the first quadrature component data are subjected to fast Fourier transform to obtain target frequency domain data; The target frequency domain data is sequentially stored in a data buffer to obtain a target data matrix; The points of the target data matrix are linearly mapped to color information to obtain a target spectrum graph; The target spectrum graph is input into a signal detection model to obtain an output result to calculate a signal parameter of the target signal; wherein the signal detection model is obtained according to the model training method of any one of claims 1-4.

6. The signal detection method of claim 5, wherein, The target spectrum image input signal is input into the signal detection model to obtain an output result to calculate a signal parameter of the target signal, including: The target spectrum image input signal is input into the signal detection model to obtain an output result; The output result is mapped into time information and frequency information, and the signal parameter of the target signal is calculated according to the time information and the frequency information.

7. A model training apparatus characterized by comprising: For a scene where electromagnetic interference or signal-to-noise effect exists, including: A first segmentation module is configured to segment sample in-phase component data and sample quadrature component data in multiple frames to obtain first sample in-phase component data and first sample quadrature component data; A first transformation module is configured to perform fast Fourier transform on the first sample in-phase component data and the first sample quadrature component data to obtain frequency domain data; A first matrix module is configured to sequentially store the frequency domain data into a data buffer to obtain a data matrix; A first mapping module is configured to linearly map points of the data matrix into color information to obtain a spectrum image; A labeling module is configured to label signal information on the spectrum image to obtain a training sample image; the labeling of signal information on the spectrum image to obtain the training sample image includes: Saving the spectrum image as a bitmap and labeling signal information on the bitmap to obtain the training sample image; A training module is configured to train a signal detection model based on deep learning based on the training sample image; the signal detection model can extract image features of an input bitmap and output a signal recognition result.

8. A signal detection device, characterized by Including: A second segmentation module is configured to segment in-phase component data and quadrature component data of a target signal in multiple frames to obtain first in-phase component data and first quadrature component data; A second transformation module is configured to perform fast Fourier transform on the first in-phase component data and the first quadrature component data to obtain target frequency domain data; A second matrix module is configured to sequentially store the target frequency domain data into a data buffer to obtain a target data matrix; A second mapping module is configured to linearly map points of the target data matrix into color information to obtain a target spectrum image; A detection module is configured to input the target spectrum image into a signal detection model to obtain an output result to calculate a signal parameter of the target signal; wherein the signal detection model is obtained according to the model training method in any one of claims 1-4.

9. A computer-readable storage medium storing a computer program, characterized in that, The computer program is loaded and executed by the processor to implement the model training method in any one of claims 1-4 or the signal detection method in any one of claims 5-6.

10. An electronic device, comprising: Including a processor and a memory, wherein, The memory is configured to store a computer program; The processor is configured to load and execute the computer program to enable the electronic device to perform the model training method in any one of claims 1-4 or the signal detection method in any one of claims 5-6.

Images