Radio signal detection identification method, device, medium and product

Abstract

The embodiment of the application relates to the technical field of communication, and discloses a radio signal detection and identification method, equipment, medium and product; the method comprises the following steps: acquiring time-frequency distribution data of a radio signal, and obtaining a frequency domain feature matrix based on the time-frequency distribution data; determining a target signal region in the time-frequency distribution data, and extracting complex spectrum data corresponding to the target signal region; performing inverse transformation on the complex spectrum data, and reconstructing to obtain a time domain signal component corresponding to the target signal region; mapping the time domain signal component into a time domain feature matrix which is adapted to the dimension of the frequency domain feature matrix; fusing the frequency domain feature matrix and the time domain feature matrix in a channel dimension, constructing multi-channel image data, and inputting the multi-channel image data into a target detection model for identification, so that the identification precision of the radio signal in a multi-signal scene is improved.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a method, device, medium and product for detecting and identifying radio signals. Background Technology

[0002] Currently, in order to monitor radio signal transmission within a region and investigate illegal applications, radio monitoring systems typically employ signal identification technology to confirm the attributes of target users. Existing technologies mainly fall into two categories: direct identification based on IQ data and graphical identification based on time-frequency maps.

[0003] Direct recognition methods based on IQ data extract features directly from the acquired complex signals. However, this approach suffers from high computational complexity, as the model needs to handle the complex amplitude and phase relationships between the real and imaginary parts, resulting in a large parameter size and long training period. Furthermore, IQ data is highly sensitive to the acquisition environment, such as hardware sampling rate and center frequency, leading to poor model portability across different devices. Additionally, the waveforms themselves lack intuitive visualization, making manual verification difficult.

[0004] On the other hand, graphical recognition methods based on time-frequency maps convert signals into time-frequency maps using Fourier transforms, and then process them using target detection models in computer vision. While this method is intuitive, the resulting spectrogram often only retains the energy distribution, losing the phase information of the original signal, thus limiting its ability to distinguish phase-sensitive signals. More importantly, mainstream target detection models (such as the YOLO series) typically require three-channel (RGB) input, while the time-frequency map is essentially a single-channel grayscale image. Existing technologies often simply copy the single-channel image three times to fit the model, which not only wastes computational resources and GPU memory, but also results in complete duplication of information across the three channels, failing to provide the model with additional feature dimensions. This leads to significant bottlenecks in recognition accuracy and efficiency when facing complex electromagnetic environments or narrowband signals. Summary of the Invention

[0005] One objective of this application is to provide a method, device, medium, and product for radio signal detection and identification, at least to solve the problems of low feature utilization, redundant model input, and insufficient identification accuracy in low signal-to-noise ratio or multi-signal scenarios in existing radio signal identification technologies.

[0006] To achieve the above objectives, some embodiments of this application provide the following aspects:

[0007] In a first aspect, some embodiments of this application provide a method for detecting and identifying radio signals, the method comprising:

[0008] Acquire time-frequency distribution data of radio signals, and obtain a frequency domain feature matrix based on the time-frequency distribution data;

[0009] The target signal region is determined from the time-frequency distribution data, and the complex spectrum data corresponding to the target signal region is extracted.

[0010] Perform an inverse transform on the complex spectrum data to reconstruct the time-domain signal components corresponding to the target signal region;

[0011] The time-domain signal components are mapped to a time-domain feature matrix that is adapted to the dimensions of the frequency-domain feature matrix;

[0012] The frequency domain feature matrix and the time domain feature matrix are fused along the channel dimension to construct multi-channel image data, which is then input into the target detection model for recognition.

[0013] Secondly, some embodiments of this application also provide an electronic device, the electronic device comprising: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method described above.

[0014] Thirdly, some embodiments of this application also provide a computer-readable medium having computer program instructions stored thereon, which can be executed by a processor to implement the method described above.

[0015] Fourthly, some embodiments of this application also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the method described above.

[0016] Compared with related technologies, the solution provided in this application, firstly, by constructing a multi-channel image including the frequency domain power spectrum and the reconstructed time domain IQ / phase components, breaks through the limitation of single-channel information replication in traditional methods. The model can simultaneously capture signal features from multiple complementary dimensions such as energy distribution, real waveform, and imaginary phase, thus exhibiting stronger feature extraction and classification capabilities when facing different standards, narrowband signals, or complex modulation methods, effectively solving the problem of phase information loss.

[0017] Secondly, since this method directly generates a three-channel feature map that highly matches the input structure of mainstream object detection models, it eliminates the need for additional image copying or format conversion, avoiding unnecessary redundant computation and memory usage. This allows the model to learn with higher information density, reducing training time and hardware resource consumption while improving the overall training cost-effectiveness, making it easier to deploy in resource-constrained scenarios such as edge computing devices.

[0018] Finally, by introducing a signal region localization and inverse transform reconstruction mechanism based on energy detection, a method for extracting effective signals is achieved. This "localization first, reconstruction second, and fusion third" processing flow can effectively suppress background noise and out-of-band interference, resulting in a purer temporal feature matrix. In complex scenarios with low signal-to-noise ratio, frequency drift, or multiple overlapping signals, this method can better separate different signal patterns and maintain the stability of detection and recognition. Attached Figure Description

[0019] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0020] Figure 1 An exemplary flowchart of a radio signal detection and identification method provided in some embodiments of this application;

[0021] Figure 2 An exemplary flowchart of another radio signal detection and identification method provided in some embodiments of this application;

[0022] Figure 3 An exemplary structural diagram of the electronic device provided for some embodiments of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Figure 1 An exemplary flowchart of a radio signal detection and identification method provided in some embodiments of this application is shown. The method includes:

[0025] S101. Obtain the time-frequency distribution data of the radio signal, and obtain the frequency domain feature matrix based on the time-frequency distribution data.

[0026] Specifically, the system first acquires the raw IQ data of the radio signal (containing the in-phase component I and the quadrature component Q) and performs a short-time Fourier transform (STFT) on it. This transformation converts the one-dimensional time-domain signal into a two-dimensional time-frequency complex matrix, which contains both the complex amplitude and phase information of the signal. To obtain a fundamental frequency-domain view for subsequent processing, the system creates a copy of this time-frequency complex matrix and takes the logarithmic magnitude of the copy, thereby obtaining the power spectrum matrix characterizing the energy intensity distribution of the signal in the time-frequency domain, i.e., the frequency-domain characteristic matrix.

[0027] S102. Determine the target signal region from the time-frequency distribution data and extract the complex spectrum data corresponding to the target signal region.

[0028] Specifically, to separate effective signals from complex electromagnetic environments, this step utilizes an energy detection algorithm to analyze the frequency domain feature matrix. First, the power spectrum matrix is ​​normalized to eliminate background noise inhomogeneity. Next, appropriate threshold conditions are set to judge the normalized data, marking strong signal points and candidate signal points. Then, using the connectivity characteristics of the signal in time and frequency, constraint analysis is performed on the candidate signal points to eliminate isolated noise points and determine the effective signal region. Finally, this mask is used to extract complex spectral data containing only the target signal region from the original complex time-frequency matrix.

[0029] S103. Perform an inverse transform on the complex spectrum data to reconstruct the time-domain signal component corresponding to the target signal region.

[0030] Specifically, for the complex spectrum data within the specific frequency band and time window extracted in step S102, an inverse Fourier transform (IFFT) is performed frame by frame using the same parameters as the STFT. Through this inverse transform process, the target signal, originally mixed in the full-band data, is restored to an IQ signal segment in the time domain. This ensures that the reconstructed time-domain signal component retains the independent amplitude and phase change characteristics of the target signal, while eliminating background noise and interference outside the mask range.

[0031] S104. Map the time-domain signal components to a time-domain feature matrix that is adapted to the dimension of the frequency-domain feature matrix.

[0032] Specifically, since the reconstructed time-domain IQ signal segments are essentially one-dimensional time series, while the model input requires a two-dimensional image structure, mapping processing is necessary. Specifically, a blank matrix plane (e.g., an all-zero matrix) with the exact same size as the power spectrum matrix obtained in step S101 is constructed. Based on the time position and frequency band information of the target signal in the original time-frequency matrix, the reconstructed in-phase components (I-path signals) and quadrature components (Q-path signals) are mapped back to their corresponding coordinate regions in the blank matrix. In this way, I-path and Q-path matrices that maintain amplitude and phase independence are generated as time-domain feature matrices, ensuring strict spatial semantic alignment between time-domain and frequency-domain features.

[0033] S105. The frequency domain feature matrix and the time domain feature matrix are fused in the channel dimension to construct multi-channel image data and input into the target detection model for recognition.

[0034] Specifically, in this step, the power spectrum matrix reflecting the signal energy distribution, the I-path matrix reflecting the real waveform characteristics, and the Q-path matrix reflecting the imaginary phase characteristics are superimposed along the channel dimension. This superposition constructs a three-channel composite image data containing complementary information, directly adapting to the input requirements of mainstream target detection models (such as YOLO or RT-DETR). When this multi-channel image data is input into a pre-trained model, the model can simultaneously utilize the energy texture in the frequency domain and the waveform phase characteristics in the time domain to output the radio signal's standard type, start and end times, and frequency range, thus completing the recognition task.

[0035] In this embodiment, by fusing frequency domain energy features with reconstructed time domain waveform features, the accuracy and robustness of signal recognition are significantly improved. This overcomes the problems of lost phase information in traditional time-frequency diagrams and the high computational complexity of directly processing IQ data. Utilizing IFFT reconstruction and matrix mapping techniques, rich three-dimensional features, including energy, real and imaginary parts, are provided to the target detection model without increasing model input redundancy (no need to copy single-channel images). This not only enables the model to more effectively capture subtle differences in narrowband signals or complex modulation signals, improving detection stability in low signal-to-noise ratio environments, but also directly adapts to the input structure of mainstream visual models, avoiding additional data conversion overhead, thereby significantly improving the system's training efficiency and inference speed.

[0036] In one embodiment, mapping the time-domain signal component to a time-domain feature matrix adapted to the dimensions of the frequency-domain feature matrix specifically includes:

[0037] Obtain in-phase component data and quadrature component data from the reconstructed time-domain signal components;

[0038] Based on the location information of the target signal region in the time-frequency distribution data, the in-phase component data and the quadrature component data are respectively constructed into I-path matrices and Q-path matrices as the time-domain feature matrices.

[0039] Specifically, such as Figure 2 As shown, mapping the time-domain signal components to a time-domain feature matrix adapted to the dimensions of the frequency-domain feature matrix specifically includes the following process: First, the system separates the in-phase component data (I component) and the quadrature component data (Q component) from the time-domain IQ signal segment reconstructed in step S103. These two components represent the real and imaginary parts of the signal, respectively, and are the basis for describing the complex characteristics of radio signals. Next, in order to adapt the one-dimensional time-domain sequence to the two-dimensional image model input space, the system uses the position information of the target signal region in the original time-frequency distribution data (specifically including the time window index range and the frequency band position occupied by the signal in the STFT transformation process) to construct two blank matrices with the same size as the frequency-domain feature matrix (power spectrum matrix) in step S101, which are defined as the initial I-path matrix and the initial Q-path matrix, respectively. Typically, all elements of these two initial matrices are initialized to zero to characterize that the non-signal region (i.e., the background noise region) has no effective signal component in the time domain. Subsequently, the separated in-phase and quadrature component data are filled into the corresponding positions of the initial I-path and initial Q-path matrices, strictly according to their corresponding time and frequency coordinates. Through this mapping operation, the original one-dimensional time-domain waveform is embedded into a specific region of a two-dimensional matrix, thereby generating the final I-path and Q-path matrices. These two matrices maintain strict semantic alignment with the power spectrum matrix at spatial pixel positions; that is, the same coordinate point in the image corresponds to the energy intensity, real part, and imaginary part of the time domain at that frequency at that moment, respectively, on different channels.

[0040] In this embodiment, by mapping the reconstructed I / Q components to independent two-dimensional matrices, the model's ability to capture subtle signal features is significantly enhanced. Compared to traditional grayscale time-frequency maps that only use magnitude information, this processing method fully preserves the original phase information of the signal (implied in the ratio of I / Q data) and the time-domain waveform envelope features. The combination of the I-path matrix, Q-path matrix, and power spectrum matrix enables the target detection model to simultaneously extract three-dimensional features from the signal from three complementary dimensions: "energy intensity," "real part evolution," and "imaginary part evolution." This greatly improves the accuracy and discriminative power for phase-sensitive systems (such as PSK, QAM, etc.) and signals with aliasing at the same frequency.

[0041] In one embodiment, mapping the time-domain signal component to a time-domain feature matrix adapted to the dimensions of the frequency-domain feature matrix specifically includes:

[0042] Amplitude and phase data are calculated from the reconstructed time-domain signal components;

[0043] Based on the location information of the target signal region in the time-frequency distribution data, the amplitude data and the phase data are respectively constructed into an amplitude matrix and a phase matrix, which are used as the time-domain feature matrix.

[0044] Specifically, firstly, the system performs coordinate transformation on the time-domain IQ signal segment reconstructed in step S103. Although the reconstructed signal exists in complex form (I+jQ), this embodiment does not directly use the I / Q components. Instead, it calculates the magnitude and argument of each sampling point and converts them into amplitude and phase data. Next, dimension adaptation and matrix construction are performed. The system reads the position index of the target signal region in the original time-frequency distribution data (i.e., the time window sequence number and frequency channel range corresponding to the signal segment) and constructs two blank matrices with the same size as the frequency domain feature matrix (power spectrum matrix), which are defined as the amplitude matrix and the phase matrix, respectively. Subsequently, the calculated amplitude data sequence and phase data sequence are accurately mapped and filled into the corresponding regions of these two matrices according to their time window correspondence, while the remaining non-signal regions remain zero or background values. The final generated amplitude matrix intuitively displays the envelope fluctuations of the signal in the time domain, while the phase matrix records the phase jump or continuous evolution process of the signal. Together, they constitute the time-domain feature matrix.

[0045] In this embodiment, a time-domain feature matrix is ​​constructed using an "amplitude + phase" mapping method, explicitly decoupling signal features into envelope and phase information. This more physically meaningful feature representation is particularly suitable for identifying modulation schemes that are sensitive to amplitude (e.g., ASK, AM) or phase (e.g., PSK, PM). By mapping one-dimensional phase changes and amplitude fluctuations to two-dimensional image channels, the target detection model can more directly learn key decision features such as "constant envelope" or "phase abrupt change," reducing the difficulty for the model to implicitly extract these features from mixed I / Q data, thereby further improving classification accuracy and convergence speed under complex modulation types.

[0046] In one embodiment, mapping the time-domain signal component to a time-domain feature matrix adapted to the dimensions of the frequency-domain feature matrix specifically includes:

[0047] Phase data is calculated from the reconstructed time-domain signal components, and a phase matrix is ​​constructed based on the position information of the target signal region in the time-frequency distribution data.

[0048] The frequency domain feature matrix is ​​subjected to pooling to obtain a pooling matrix that fits the size of the frequency domain feature matrix.

[0049] The phase matrix and the pooling matrix are combined to form the temporal feature matrix.

[0050] Specifically, firstly, the system extracts phase features from the time-domain IQ signal segment reconstructed in step S103. The phase value of each sampling point is calculated using inverse trigonometric functions, converting the complex-form time-domain signal into a one-dimensional phase change sequence. This sequence can keenly reflect the phase jump or continuous rotation characteristics during signal modulation.

[0051] Next, based on the location information of the target signal region in the time-frequency distribution data (i.e., the time window index and frequency range), a blank matrix with the same size as the frequency domain feature matrix (power spectrum matrix) is constructed and defined as the phase matrix. The calculated phase data sequence is then precisely mapped and filled into the corresponding coordinate region of the phase matrix according to its time correspondence in the original signal, thereby graphically representing the one-dimensional phase information as a two-dimensional feature map.

[0052] Simultaneously, the frequency domain feature matrix (i.e., the power spectrum matrix) obtained in step S101 is subjected to pooling processing. Specifically, max pooling or average pooling strategies can be used to downsample or aggregate the power spectrum. To ensure that the generated pooled data can be fused with other matrices as an independent channel of the multi-channel image, it is necessary to control the stride and padding of the pooling layer, or to perform interpolation amplification after pooling, so that the final pooling matrix is ​​adapted to the frequency domain feature matrix in terms of size (i.e., consistent length and width). The pooling matrix obtained in this step highlights the main energy skeleton or significant feature regions of the signal in the time-frequency domain.

[0053] Finally, the constructed phase matrix and pooling matrix are combined to serve as two components of the temporal feature matrix (e.g., corresponding to the G and B channels of the image, respectively), which together with the original frequency domain feature matrix (R channel) constitute the final multi-channel input data.

[0054] In this embodiment, on the one hand, the introduction of a phase matrix allows the model to directly utilize the phase modulation information of the signal. This is crucial for recognizing phase-sensitive signals such as PSK (Phase Shift Keying) and QAM (Quadrature Amplitude Modulation), overcoming the limitation of traditional spectrograms which only contain amplitude information. On the other hand, using a pooling matrix instead of the original IQ or amplitude information effectively extracts the "backbone" features of the signal energy distribution, serving as a form of spatial filtering and feature enhancement. Max pooling is translation-invariant, tolerating small jitters or frequency offsets in the time-frequency plot, thereby enhancing the model's robustness to non-stationary signals or multipath interference. This three-channel configuration further enriches the model's input dimensions, contributing to improved signal recognition accuracy in complex electromagnetic environments.

[0055] It should be noted that the three implementation methods described above for mapping the reconstructed time-domain signal components to time-domain feature matrices (i.e., constructing I-path and Q-path matrices, constructing amplitude and phase matrices, and constructing phase and pooling matrices) are parallel and interchangeable schemes under the technical concept of this application. In practical applications, technicians can flexibly choose any one of these methods to construct the time-domain feature matrix based on specific hardware computing resource limitations, the main modulation characteristics of the signal to be identified, and differentiated requirements for noise resistance robustness.

[0056] In one embodiment, determining the target signal region in the time-frequency distribution data specifically includes:

[0057] The frequency domain feature matrix is ​​subjected to noise normalization.

[0058] The normalized data is filtered according to the preset threshold conditions, and candidate signal points are marked.

[0059] The candidate signal points are constrained and determined based on time-frequency connectivity to identify the target signal region.

[0060] Specifically, firstly, the frequency domain feature matrix undergoes refined noise normalization. Considering that the background noise floor in different frequency bands of the radio environment is often uneven (i.e., colored noise or fixed-frequency interference exists), the system traverses each row of the power spectrum matrix (corresponding to each frequency channel), calculating the lower quantiles (e.g., 5% to 20% quantiles) of the energy values ​​at all time points in that row, and using this as the noise benchmark for that frequency channel. Then, this noise benchmark is used to normalize all the energy data in that row. This step effectively flattens the noise floor across the entire frequency band, ensuring that subsequent threshold decisions are no longer affected by differences in frequency band noise.

[0061] Next, the normalized data is subjected to multi-level filtering based on preset threshold conditions. This embodiment employs a dual-threshold decision mechanism to balance detection sensitivity and false alarm rate. First, a target signal-to-noise ratio (SNR) is set based on experience or system metrics, and then the formula is used... A high threshold is calculated. Simultaneously, to prevent missed detections due to energy attenuation at signal edges, a low threshold is set, calculated using the following formula: Where the coefficient The value typically ranges from 0.3 to 0.6. The energy value of each element in the normalized time-frequency matrix is ​​compared to these two thresholds: values ​​higher than... Points marked as strong signal points; between and Points between these points are marked as candidate signal points; points below these points are marked as candidate signal points. The points are then considered background noise and set to zero.

[0062] Subsequently, candidate signal points are constrained and determined based on time-frequency connectivity, and a final mask is generated. Since radio signals are typically continuous in the time-frequency domain, while noise is often random and isolated, this embodiment stipulates that a candidate signal point is only considered part of the valid signal if it has a direct or indirect connection with at least one strong signal point in the time-frequency space; otherwise, it is discarded. Through this connectivity analysis, the signal region naturally grows from the high-energy core to the surrounding weak-energy edges, while suppressing isolated noise spikes.

[0063] Finally, to obtain a more regular target signal region, morphological post-processing can be performed on the above decision results. The area and duration of the initially generated binary mask are statistically analyzed to remove fragmented regions that are significantly smaller than normal signal characteristics (e.g., those with too narrow a bandwidth or too short a duration). Furthermore, morphological opening operations (erosion followed by dilation) are used to remove glitches, or closing operations (dilation followed by erosion) are used to fill holes within the signal, ultimately obtaining a mask with smooth edges and a complete structure for subsequent accurate extraction of complex spectral data.

[0064] In this embodiment, a region determination method based on dual thresholds and connectivity constraints is employed, significantly improving the accuracy of signal extraction in complex electromagnetic environments. Line-by-line noise normalization effectively overcomes the interference of colored noise on broadband detection. A high-low dual threshold mechanism is introduced, utilizing the high threshold to ensure detection accuracy (low false alarms) while using the low threshold to protect the integrity of weak signal edges (low false negatives). Furthermore, time-frequency connectivity constraints and morphological processing further utilize the physical continuity of the signal to "clean" the mask, ensuring that the final extracted and reconstructed time-domain signal is continuous and pure, providing high-quality input features for subsequent model recognition.

[0065] In one embodiment, the noise normalization processing of the frequency domain feature matrix includes:

[0066] Statistically analyze the energy distribution of the frequency domain feature matrix in different frequency rows;

[0067] The low quantile value of each frequency row is calculated as a noise reference, and the frequency row is normalized using the noise reference.

[0068] Specifically, firstly, the system scans the frequency domain feature matrix (i.e., the power spectrum matrix) generated in step S101 row by row. Since each row of the frequency domain feature matrix corresponds to a specific frequency channel, and each column corresponds to a time sampling point, the system needs to statistically analyze the energy value distribution of each row at all time points. Next, the lowest quantile value of each frequency row is calculated as the noise baseline for that frequency. Specifically, for each row in the matrix, all energy values ​​in that row are sorted, and the values ​​at the lowest quantile (e.g., 5% to 20% quantile) are selected as the background noise baseline estimate for that frequency channel. The reason for selecting the lowest quantile rather than the minimum value is to avoid the impact of occasional extremely low value measurement errors on the stability of the baseline; the reason for selecting the lowest quantile rather than the mean is that the presence of a signal will raise the mean, while the lowest quantile is more representative of the pure noise level. Finally, the frequency row is normalized using this noise baseline. The specific operation involves dividing the energy value at each time point in the frequency row by the noise baseline value of that row (or performing a subtraction operation in the logarithmic domain), thereby flattening the noise floor of the row to a unit level (or 0 dB level), thus completing the standardization process for the entire matrix.

[0069] In this embodiment, a row-level noise normalization method based on low quantiles is employed, which significantly improves the system's detection performance in non-stationary noise environments. In real-world radio environments, background noise intensity typically varies greatly across different frequency bands. Directly using a globally uniform threshold for detection can lead to numerous false alarms in high-noise bands or missed detections of weak signals in low-noise bands. This embodiment effectively eliminates the noise gain differences between different frequency channels by calculating and normalizing the noise baseline row by row, allowing subsequent energy detection algorithms to use a uniform signal-to-noise ratio threshold for fair and accurate signal decision-making across the entire frequency band.

[0070] In one embodiment, after the step of constraining the candidate signal points based on time-frequency connectivity to determine the target signal region, the method further includes:

[0071] The initially identified areas are filtered using thresholds based on area or duration.

[0072] The filtered regions are smoothed to generate the final mask used for data extraction.

[0073] Specifically, firstly, the binary mask initially generated using the energy detection algorithm and connectivity determination is subjected to geometric feature statistics. The system traverses each connected region in the mask, calculating the total number of pixels it covers (i.e., the region area) and the length it spans on the time axis (i.e., the duration). These statistical values ​​are compared with preset minimum area thresholds and minimum duration thresholds. In real-world radio environments, effective communication signals typically possess a certain duration and bandwidth, while sudden impulse noise or narrowband interference often manifests as extremely short or narrow isolated spots. Therefore, the system classifies regions with an area smaller than the preset area threshold or a duration smaller than the preset time threshold (e.g., less than a few milliseconds or several time sampling points) as invalid interference and sets their pixel values ​​to zero in the mask, thereby eliminating fragmented noise.

[0074] Next, the selected effective regions are morphologically smoothed to generate the final mask. Due to fading effects or transient noise interference in the wireless channel, the edges of the initially detected signal region may have jagged edges, or there may be voids formed by energy breaks within the signal. To correct these defects, this embodiment uses morphological operations to modify the mask. Specifically, a "closing operation" (dilation followed by erosion) can be used to fill the small voids inside the signal region with structuring elements and connect adjacent broken signal segments to ensure the compactness of the signal mask; or a "opening operation" (erosion followed by dilation) can be combined to remove the fine jagged edges of the region and smooth the signal boundaries. After the above processing, a final mask with smooth edges and complete internal filling is generated, and the target signal data is accurately extracted from the original complex spectrum data using this mask for subsequent IFFT reconstruction.

[0075] In this embodiment, by introducing a geometric feature-based filtering and morphological smoothing mechanism, the integrity and signal-to-noise ratio of signal extraction are significantly improved. Specifically, the threshold filtering strategy based on region area and duration effectively identifies and filters out short-duration pulses and narrowband interference, thereby significantly reducing the false alarm rate. Morphological smoothing utilizes image morphology principles to repair discontinuous or incomplete signal time-frequency distribution caused by factors such as channel fading, ensuring the connectivity of the extracted signal segments in both the time and frequency domains. This post-processing strategy, which integrates interference removal and structural restoration, ensures that the signal samples input to the subsequent neural network model have high quality and structural integrity, effectively avoiding recognition errors caused by data fragmentation or residual noise.

[0076] Furthermore, in one embodiment, constructing the multi-channel image data includes:

[0077] The frequency domain feature matrix is ​​used as the first channel of the image;

[0078] Different component matrices in the temporal feature matrix are used as subsequent channels of the image;

[0079] Each channel matrix is ​​normalized and encoded to generate a three-channel image that meets the input requirements of the target detection model.

[0080] Specifically, firstly, the system establishes channel mapping rules for multi-channel images, designating the frequency domain feature matrix (i.e., power spectrum matrix) obtained in step S101 as the first channel data of the image (usually corresponding to the R channel in the RGB color model). This channel carries information about the strength of the signal's energy distribution in the time-frequency domain. Simultaneously, two different component matrices from the time-domain feature matrix generated in step S104 (e.g., the I-path matrix and Q-path matrix, or the amplitude matrix and phase matrix) are designated as subsequent channel data of the image (usually corresponding to the G channel and B channel in the RGB color model). This mapping method ensures that each pixel in the image simultaneously represents the physical properties of that time-frequency unit, such as energy, real waveform, and imaginary phase, in the channel dimension. Finally, to adapt to the input layer requirements of the target detection model, normalization encoding is performed on each of the above channel matrices. Specifically, the system statistically analyzes the dynamic range of each matrix value and uses linear or nonlinear mapping algorithms to uniformly map the power value (dB) of the frequency domain features and the amplitude or phase value of the time domain features to a standard pixel value range (e.g., the 0-255 range for an 8-bit image or the 0.0-1.0 range for a floating-point input). This step eliminates the numerical differences between different physical dimensions, generating a three-channel composite image that meets the input requirements of the target detection model in terms of both data format and numerical distribution.

[0081] In this embodiment, by mapping heterogeneous signal features to standard image channels and performing normalized encoding, seamless adaptation of radio signal data to the input architecture of mainstream computer vision models is achieved. This processing method not only eliminates the dimensional differences between frequency domain energy and time domain waveform, ensuring numerical stability and convergence speed during model training, but also enables direct transfer learning using existing object detection models (such as YOLO and ResNet) pre-trained on massive amounts of optical images, avoiding modifications to the underlying network structure. This significantly reduces model development costs while maximizing the model's efficiency in parsing and utilizing multidimensional signal features.

[0082] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this application.

[0083] Furthermore, some embodiments of this application also provide an electronic device. The electronic device can be various forms of digital computer, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, etc. The electronic device can also be various forms of mobile devices, such as cellular phones, smartphones, wearable devices, and other similar computing devices.

[0084] The electronic device includes: one or more processors; and a memory storing computer program instructions that, when executed, cause the processor to perform the steps of the methods provided in any one or more of the above embodiments. Figure 3 An exemplary structural diagram of the electronic device is disclosed. The electronic device includes one or more processors 1101, a memory 1102, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple electronic devices can be connected, each providing some of the necessary operations. The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0085] The electronic device may further include an input device 1103 and an output device 1104. The processor 1101, memory 1102, input device 1103 and output device 1104 may be connected by a bus or other means, as shown in the figure, which is connected by a bus.

[0086] Input device 1103 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the electronic device, such as a touch screen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 1104 may include a display device, auxiliary lighting device (e.g., LED), and haptic feedback device (e.g., vibration motor). The display device may include, but is not limited to, a liquid crystal display, a light-emitting diode display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0087] To provide interaction with the user, the electronic device can be a computer. The computer has: a display device (e.g., a cathode ray tube or LCD monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback); and input from the user can be received in any form (e.g., voice input or tactile input).

[0088] In this embodiment, a computer-readable medium stores a computer program / instructions that, when executed by a processor, implement the steps of the methods provided in any one or more of the above embodiments. This computer-readable medium may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into that device. The aforementioned computer-readable medium carries one or more computer-readable instructions.

[0089] The memory 1102 can serve as a non-transitory computer-readable storage medium, used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor 1101 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions, and modules stored in the memory 1102, thereby implementing the program instructions / modules corresponding to the methods provided in any one or more of the embodiments described above in this application.

[0090] The memory 1102 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 1102 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 1102 may optionally include memory remotely located relative to the processor 1101, and these remote memories can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0091] It should be noted that the computer-readable medium described in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. Computer-readable media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections having one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0092] Computer-readable media include permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, read-only optical discs, digital versatile optical discs or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0093] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including local area networks (LANs) or wide area networks (WANs), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0094] In the above embodiments, all or part of the implementation can be achieved through software, hardware, firmware, or any combination thereof. For example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In some embodiments, the software program of this application can be executed by a processor to implement the above steps or functions. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, magnetic or optical drives, floppy disks, and similar devices. In addition, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

[0095] The computer program product provided in this application includes one or more computer programs / instructions. When executed by a processor, these computer programs / instructions generate, in whole or in part, the processes or functions described in this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0096] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0097] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device in software or hardware. Terms such as "first," "second," etc., are used only for distinguishing descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.

[0098] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily made by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.

Claims

1. A method for detecting and identifying radio signals, characterized in that, The method includes: Acquire time-frequency distribution data of radio signals, and obtain a frequency domain feature matrix based on the time-frequency distribution data; The target signal region is determined from the time-frequency distribution data, and the complex spectrum data corresponding to the target signal region is extracted. Determining the target signal region in the time-frequency distribution data specifically includes: The frequency domain feature matrix is ​​subjected to noise normalization. The normalized data is filtered according to the preset threshold conditions, and candidate signal points are marked. The candidate signal points are constrained and determined based on time-frequency connectivity to identify the target signal region. The noise normalization process for the frequency domain feature matrix includes: Statistically analyze the energy distribution of the frequency domain feature matrix in different frequency rows; The low quantile value of each frequency row is calculated as a noise reference, and the frequency row is normalized using the noise reference; an inverse transform is performed on the complex spectrum data to reconstruct the time-domain signal components corresponding to the target signal region; The time-domain signal components are mapped to a time-domain feature matrix that is adapted to the dimensions of the frequency-domain feature matrix; The frequency domain feature matrix and the time domain feature matrix are fused along the channel dimension to construct multi-channel image data, which is then input into the target detection model for recognition.

2. The method according to claim 1, characterized in that, The step of mapping the time-domain signal components into a time-domain feature matrix adapted to the dimensions of the frequency-domain feature matrix specifically includes: Obtain in-phase component data and quadrature component data from the reconstructed time-domain signal components; Based on the location information of the target signal region in the time-frequency distribution data, the in-phase component data and the quadrature component data are respectively constructed into I-path matrices and Q-path matrices as the time-domain feature matrices.

3. The method according to claim 1, characterized in that, The step of mapping the time-domain signal components into a time-domain feature matrix adapted to the dimensions of the frequency-domain feature matrix specifically includes: Amplitude and phase data are calculated from the reconstructed time-domain signal components; Based on the location information of the target signal region in the time-frequency distribution data, the amplitude data and the phase data are respectively constructed into an amplitude matrix and a phase matrix, which are used as the time-domain feature matrix.

4. The method according to claim 1, characterized in that, The step of mapping the time-domain signal components into a time-domain feature matrix adapted to the dimensions of the frequency-domain feature matrix specifically includes: Phase data is calculated from the reconstructed time-domain signal components, and a phase matrix is ​​constructed based on the position information of the target signal region in the time-frequency distribution data. The frequency domain feature matrix is ​​subjected to pooling to obtain a pooling matrix that fits the size of the frequency domain feature matrix. The phase matrix and the pooling matrix are combined to form the temporal feature matrix.

5. The method according to claim 4, characterized in that, After the step of constraining and determining the target signal region based on time-frequency connectivity of the candidate signal points, the method further includes: The initially identified areas are filtered using thresholds based on area or duration. The filtered regions are smoothed to generate the final mask used for data extraction.

6. An electronic device, characterized in that, The electronic device includes: One or more processors; and A memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method as described in any one of claims 1 to 5.

7. A computer-readable medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 5.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 5.

Images