A communication signal detection method and system based on multi-model cooperation and a medium

Abstract

This invention discloses a communication signal detection method, system, and medium based on multi-model collaboration, belonging to the field of signal communication technology. The method includes the following steps: receiving a communication radio frequency signal and converting it into broadband time-frequency data; scanning the broadband time-frequency data using a lightweight signal detection model to generate at least one candidate region, encoded in a tuple format containing frequency and time intervals; extracting corresponding time-frequency sub-segments from the broadband time-frequency data based on the encoding of the candidate regions, and inputting them into a deep signal detection model; performing local detection on the time-frequency sub-segments using the deep signal detection model, and outputting the signal existence determination result and detection confidence level. The method, system, and medium of this invention can achieve multi-model collaboration, ensuring both the comprehensiveness of signal scanning and the accuracy of signal detection, meeting the rapid detection requirements of communication signals under broadband conditions, and improving the intelligence level and engineering application value of communication signal detection systems.

Description

Technical Field

[0001] This invention belongs to the field of communication signal detection technology, specifically relating to a communication signal detection method, system and medium based on multi-model collaboration, which is particularly suitable for signal presence detection and blind monitoring in wireless communication scenarios such as shortwave and ultra-shortwave. Background Technology

[0002] Communication signal detection is a fundamental function in wireless communication systems, spectrum monitoring systems, and electromagnetic environment sensing systems. Its task is to determine the presence of communication signals within a bandwidth. With the diversification of wireless communication standards and the increasing complexity of electromagnetic environments, communication signals exhibit characteristics such as wide bandwidth, rapid time-domain changes, and large signal-to-noise ratio fluctuations in the spectrum. Against this backdrop, how to detect the existence of communication signals of unknown standards or with unknown parameters within a wide broadband spectrum has become a focus of technological research.

[0003] Currently, common communication signal detection methods mainly include energy detection methods, feature detection methods, and model-based detection methods. Among them, energy detection methods achieve rapid determination by setting an energy threshold, but their performance is greatly affected by noise estimation errors and channel conditions; feature detection methods rely on specific signal structure features for determination, and their applicability is limited in unknown standard scenarios; model-based detection methods use deep learning models to determine time-frequency features, and although they can achieve high detection capabilities in complex scenarios, directly performing deep model inference on broadband spectrum will result in high computational resource consumption and bandwidth processing overhead.

[0004] Furthermore, when detecting communication signals under broadband conditions, it is usually necessary to balance real-time performance and detection accuracy. On the one hand, the broadband spectrum contains a large number of signal-free areas, and if the deep model performs point-by-point inference across the entire bandwidth, it will lead to redundant calculations. On the other hand, burst signals or weak signals may exist in different frequency bands, and a single model cannot meet the needs of broadband coverage and local fine detection under full-band conditions, thus failing to fully meet the requirements of practical applications. Summary of the Invention

[0005] In response to one or more of the above-mentioned defects or improvement needs of existing technologies, this invention provides a communication signal detection method, system, and medium based on multi-model collaboration. It aims to overcome the shortcomings of existing communication signal detection technologies, such as the difficulty in balancing detection accuracy and efficiency, instability in detecting low signal-to-noise ratio signals, and lack of adaptability in manually adjusting detection thresholds and strategies. This invention achieves efficient, reliable, and adaptive signal detection capabilities for broadband communication time-frequency data, thereby improving the intelligence level and engineering application value of communication signal detection systems.

[0006] To achieve the above objectives, one aspect of the present invention provides a communication signal detection method based on multi-model collaboration, comprising the following steps: S1: Receives communication radio frequency signals and converts them into broadband time-frequency data; S2: A lightweight signal detection model is used to scan the broadband time-frequency data to generate at least one candidate region that is suspected of containing communication signals; wherein, the candidate region is encoded in a tuple format that includes frequency intervals and time intervals; S3: Based on the coding of the candidate region, extract the corresponding time-frequency sub-segments from the broadband time-frequency data and input them into the depth signal detection model; S4: Use the depth signal detection model to perform local detection on the time-frequency sub-segment, and output the signal existence determination result and detection confidence level.

[0007] As a further improvement of the present invention, in step S2, the lightweight signal detection model generates candidate regions using a candidate region generation mode based on an energy statistical threshold. This candidate region generation mode includes: Based on a preset energy threshold, broadband time-frequency data is scanned frame by frame, and frequency intervals where the energy continuously exceeds the energy threshold and time frame intervals that appear consecutively are encoded into tuple format containing frequency intervals and time intervals.

[0008] As a further improvement of the present invention, in step S2, the lightweight signal detection model generates candidate regions using a candidate region generation mode based on a lightweight neural network, the candidate region generation mode including: A shallow convolutional neural network is used to classify local windows of broadband time-frequency data, and time-frequency ranges with classification probabilities greater than a preset threshold are recorded as candidate regions.

[0009] As a further improvement of the present invention, the method also includes an adaptive detection and adjustment process: S5: Generate feedback control parameters based on the detection confidence output by the depth signal detection model; and use the feedback control parameters to update the lightweight signal detection model, forming a closed-loop adaptive detection adjustment process.

[0010] As a further improvement of the present invention, the lightweight signal detection model is updated using feedback control parameters, including: Adjust at least one of the following in the lightweight signal detection model: detection threshold, scan step size, or frequency band priority, through online or offline updates.

[0011] As a further improvement of the present invention, the adaptive detection adjustment process includes a detection threshold adjustment process, which includes: S51: Obtain the average detection confidence score output by the depth signal detection model; S52: Generate feedback control parameters for adjusting the detection threshold based on the average detection confidence level; where: If the average detection confidence is lower than the lower limit of the preset confidence threshold range, feedback control parameters are generated to reduce the detection threshold of the lightweight signal detection model. If the average detection confidence level is higher than the upper limit of the preset confidence threshold range, feedback control parameters are generated to improve the detection threshold of the lightweight signal detection model.

[0012] As a further improvement of the present invention, the adaptive detection and adjustment process includes: In response to the number of candidate regions generated within a preset period exceeding the processing capacity threshold, feedback control parameters are generated to increase the temporal scan step size. and / or Based on the cumulative detection confidence of each frequency range within a preset period, the scanning priority of the corresponding frequency range is adjusted.

[0013] Another aspect of the present invention provides a communication signal detection system based on multi-model collaboration, comprising: The signal receiving and preprocessing module is used to receive communication radio frequency signals and convert them into broadband time-frequency data. A lightweight signal detection module is used to scan broadband time-frequency data to generate at least one candidate region suspected of containing communication signals, wherein the candidate region is encoded in a tuple format containing frequency intervals and time intervals; The depth signal detection module is used to extract the corresponding time-frequency sub-segments from the broadband time-frequency data according to the encoding of the candidate region, and to perform local detection on the time-frequency sub-segments to output the signal existence determination result and detection confidence.

[0014] As a further improvement of the present invention, the system further includes an adaptive update module configured with a feedback control subunit and a module update subunit, wherein: The feedback control subunit is configured to communicate with the depth signal detection module to generate feedback control parameters based on the detection confidence level output by the depth signal detection module. The module update subunit is used to receive feedback control parameters and apply them to update at least one of the detection threshold, scan step size, or frequency band interval scan priority of the lightweight signal detection module.

[0015] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned multi-model cooperative communication signal detection method.

[0016] The aforementioned improved technical features can be combined with each other as long as they do not conflict with each other.

[0017] In summary, the beneficial effects of the above-described technical solutions conceived by this invention compared with the prior art include: The communication signal detection method based on multi-model collaboration of the present invention includes the following steps: receiving a communication radio frequency signal and converting it into broadband time-frequency data; scanning the broadband time-frequency data using a lightweight signal detection model to generate at least one candidate region suspected of containing a communication signal; wherein, the candidate region is encoded in a tuple format containing a frequency interval and a time interval; extracting the corresponding time-frequency sub-segments from the broadband time-frequency data according to the encoding of the candidate region, and inputting them into a deep signal detection model; performing local detection on the time-frequency sub-segments using the deep signal detection model, and outputting the signal existence determination result and detection confidence, thereby completing the rapid and accurate detection of the communication signal. Using the aforementioned method, a two-level collaborative architecture of "lightweight coarse scanning + deep fine-grained confirmation" is constructed. The lightweight signal detection model is responsible for quickly locking the suspected region, while the deep signal detection model only performs fine analysis on the locked local region, significantly reducing the overall computational complexity and achieving a balance between detection efficiency and detection accuracy.

[0018] The communication signal detection method, system, and medium based on multi-model collaboration in this invention have simple steps and a concise system architecture. They can effectively overcome the shortcomings of existing communication signal detection technologies, such as difficulty in balancing detection accuracy and efficiency, instability in detecting low signal-to-noise ratio signals, and lack of adaptability in manually adjusting detection thresholds and strategies. They significantly improve the scanning rate and recognition accuracy of communication signals under broadband conditions, realize efficient, reliable, and adaptive signal detection capabilities for broadband communication time-frequency data, and enhance the intelligence level and engineering application value of communication signal detection systems. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the communication signal detection method based on multi-model collaboration in Embodiment 1 of the present invention; Figure 2 This is a time-frequency domain schematic diagram of the candidate region output by the lightweight signal detection model in Embodiment 1 of the present invention; Figure 3 This is a flowchart of the adaptive adjustment of the detection threshold of the lightweight signal detection model in Embodiment 1 of the present invention; Figure 4This is a module architecture diagram of the communication signal detection system based on multi-model collaboration in Embodiment 2 of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0022] In the description of this invention, it should be understood that, unless otherwise expressly specified and limited, the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," "circumferential," etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0023] Furthermore, unless otherwise expressly defined, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise expressly and specifically defined.

[0024] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0025] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0026] Below, for reference Figures 1-4 This paper describes a communication signal detection method, system, and storage medium based on multi-model collaboration according to a preferred embodiment of the present invention, and elaborates on the technical solutions of the preferred embodiment of the present invention through the following three specific embodiments.

[0027] Example 1: As a first aspect of the present invention, this embodiment proposes a communication signal detection method based on multi-model collaboration, the process of which is as follows: Figure 1 The steps shown are as follows: S1: Receives communication radio frequency signals and converts them into broadband time-frequency data; Specifically, in the preferred embodiment, a radio frequency front-end receiving device is used to receive communication radio frequency signals and perform analog-to-digital conversion on them to form a complex time-domain IQ data stream.

[0028] More specifically, the aforementioned time-domain IQ data is stored in memory in the form of a buffer queue.

[0029] Furthermore, regarding the aforementioned IQ data, in a preferred embodiment, at least one of the following methods is used for processing: Method 1: The Short-Time Fourier Transform (STFT) module is invoked to process the aforementioned IQ data. The window length L, step size H, and FFT point count K are set to obtain a value of... A two-dimensional time-frequency matrix; where: For time frames; This represents the number of frequency sampling points.

[0030] Meanwhile, the resulting matrix is ​​an amplitude spectrum or power spectrum, stored in a two-dimensional floating-point array for subsequent detection.

[0031] Method 2: By replacing the STFT with a multi-band filter bank, the broadband is decomposed into M sub-band energies, forming a group of sizes. The energy time-frequency matrix.

[0032] S2: Use a lightweight signal detection model to scan broadband time-frequency data to generate at least one candidate region that is suspected of containing communication signals; wherein, the candidate region is encoded in a tuple format that includes frequency intervals and time intervals.

[0033] Specifically, based on the broadband time-frequency data obtained in step S1, step S2 further employs a lightweight signal detection model to perform a full-band sliding window scan on the broadband time-frequency data, calculates the energy statistical characteristics or shallow texture characteristics within the window, and generates several candidate regions suspected of containing communication signals based on preset judgment conditions. A time-frequency domain schematic diagram of these candidate regions is shown below. Figure 2 As shown in the image.

[0034] In practical design, the lightweight signal detection model in the preferred embodiment preferably generates candidate regions through one of the following two methods: Method 1: Candidate Region Generation Mode Based on Energy Statistical Threshold Specifically, this method scans broadband time-frequency data frame by frame based on a preset energy threshold, and encodes frequency intervals where the energy continuously exceeds the energy threshold and time frame intervals where they occur consecutively into a tuple format containing the frequency interval and the time interval.

[0035] More specifically, it includes the following steps: (1) Obtaining background noise power And set an energy threshold. ;in, It is a multiplier; (2) Calculate the energy spectrum for each time frame, and mark the frequency ranges with energy exceeding the threshold as candidate regions, and then use the obtained candidate regions as... Formal storage.

[0036] in, Frequency start; This marks the end of the frequency range; The starting point of the time; The time has ended; The frequency range in which continuous energy exceeds the energy threshold; It represents the time frame interval where the events occur consecutively.

[0037] Method 2: Candidate Region Generation Mode Based on Lightweight Neural Networks Specifically, this method uses a shallow convolutional neural network to classify local windows of broadband time-frequency data and records the time-frequency ranges with a classification probability greater than a preset threshold as candidate regions.

[0038] For example, in one specific implementation, a shallow convolutional neural network is used to take a 256×32 submatrix as input and output a binary classification probability. Accordingly, the time-frequency ranges with a probability greater than a preset threshold are recorded as candidate regions.

[0039] In actual detection, regardless of the method used, the candidate regions are ultimately stored as linked lists or arrays, and each record includes a frequency range and a time range.

[0040] S3: Based on the encoding of the candidate region, extract the corresponding time-frequency sub-segments from the broadband time-frequency data and input them into the depth signal detection model.

[0041] Specifically, for each candidate region, a corresponding submatrix is ​​extracted from the global time-frequency matrix (i.e., the complete broadband time-frequency data) based on its tuple coordinates, namely: The aforementioned submatrix is ​​then fed into the depth signal detection model for fine-grained detection.

[0042] S4: Use a depth signal detection model to perform local detection on time-frequency sub-segments and output the signal existence determination result and detection confidence.

[0043] In a specific preferred embodiment, the aforementioned depth signal detection model is a time-frequency feature fusion network, whose input is a three-dimensional tensor:

[0044] in, The number of channels, and ;when When the input only contains the amplitude spectrum channel; when At the same time, the input includes both amplitude spectrum channel and power spectrum channel, and the combination of the two can enhance the model's ability to perceive weak signals.

[0045] Accordingly, the output of the depth signal detection model is stored in the candidate result queue, which includes: existence determination label (i.e., existence / non-existence) and detection confidence (continuous value).

[0046] In more detail, the communication signal detection method in the preferred embodiment further includes an adaptive detection adjustment process: S5: Generate feedback control parameters based on the detection confidence output by the depth signal detection model; and use the feedback control parameters to update the lightweight signal detection model, forming a closed-loop adaptive detection adjustment process.

[0047] Specifically, the adaptive detection adjustment process in the preferred embodiment preferably includes adaptive adjustment of the detection threshold and adaptive adjustment of the scanning strategy. Wherein: In order to achieve a dynamic balance between the false alarm rate and the false alarm rate, the system corrects the energy detection threshold of the lightweight signal detection model in real time based on the detection confidence deviation.

[0048] More specifically, the feedback control parameters for the corresponding detection threshold are generated using the detection confidence level, and the specific process is as follows: Figure 3 As shown, preferably including: S51: Obtain the average detection confidence score output by the depth signal detection model; S52: Generate feedback control parameters for adjusting the detection threshold based on the average detection confidence level; where: If the average detection confidence is lower than the lower limit of the preset confidence threshold range, feedback control parameters are generated to reduce the detection threshold of the lightweight signal detection model. If the average detection confidence level is higher than the upper limit of the preset confidence threshold range, feedback control parameters are generated to improve the detection threshold of the lightweight signal detection model.

[0049] For example, suppose the current moment of the lightweight signal detection model... t The detection threshold is The average detection confidence level output by the depth signal detection model is The preset confidence threshold range is .

[0050] At this moment, record this instant. t The detection threshold of +1 is Then its preferred value is updated according to the following formula:

[0051] In the formula, This is the threshold adjustment step size factor, whose value can be selected according to actual needs (such as the requirement for adjustment accuracy), for example, 0.05; For the confidence bias term, its value is preferably selected between -1, +1, and 0, and the preferred logic for value selection is as follows: When the average detection confidence Below the preset lower limit When the threshold is set too high, it indicates that the depth signal detection model has low overall confidence in the candidate regions generated by the lightweight signal detection model, meaning that the current detection threshold is set too high, resulting in some potential signals not being effectively detected. At this point, feedback control parameters (i.e., ...) are generated to lower the detection threshold. This expands the search scope and improves the detection capability of candidate regions.

[0052] When the average detection confidence Higher than the preset range upper limit When the threshold is set too low, it indicates that the depth signal detection model has a high confidence level in its judgment of the candidate region, suggesting that the current detection threshold setting is too low, resulting in redundancy in the candidate region. At this point, feedback control parameters (i.e., ...) are generated to improve the detection threshold. This reduces invalid candidate regions and improves detection efficiency.

[0053] Of course, if the average detection confidence level If the value is within the preset confidence threshold range, it means that the detection threshold can continue to be maintained. Take 0.

[0054] Furthermore, in addition to the adaptive adjustment of the detection threshold, the adaptive adjustment process for the scanning strategy of the lightweight signal detection model in the preferred embodiment is preferably as follows: (1) Adaptive adjustment of scanning step size Specifically, in response to the number of candidate regions generated within a preset period exceeding the processing capacity threshold, feedback control parameters are generated to increase the temporal scan step size, thereby reducing temporal resolution in exchange for processing speed and preventing system blockage.

[0055] (2) Adjust the scanning priority of the corresponding frequency range according to the cumulative detection confidence of each frequency range within the preset period.

[0056] More specifically, a frequency band activity table is configured and maintained; if the sum of the cumulative detection confidence of a certain frequency range in the most recent period exceeds a preset threshold, the frequency band is marked as a "high priority frequency band" and is given priority in the next round of scanning or the sampling weight of the frequency band is increased.

[0057] Using the aforementioned adaptive detection and adjustment process design, during actual operation, at least one of the detection threshold, scan step size, or frequency band priority of the lightweight signal detection model can be adjusted through online or offline updates.

[0058] Furthermore, for the lightweight signal detection model and the deep signal detection model in the preferred embodiments, both can be built using mature modeling techniques in the prior art, according to the functional requirements of the preferred embodiments. For example, the lightweight signal detection model can be built using the traditional energy statistical method or a lightweight neural network; while the deep signal detection model can be built using at least one of the following: a hybrid architecture of CNN (Convolutional Neural Network) and RNN / LSTM (Recurrent Neural Network / Long Short-Term Memory Network), an architecture based on Residual Network (ResNet), a Transformer architecture, or an architecture based on object detection. These will not be elaborated further here.

[0059] Based on the design of the communication signal detection method in this embodiment, a two-level collaborative architecture of "lightweight coarse scanning + deep-level fine-grained analysis" is constructed. The lightweight signal detection model is responsible for quickly locating suspected areas, while the deep signal detection model performs fine analysis only on the located local areas, significantly reducing the overall computational complexity and achieving a balance between detection efficiency and accuracy. Moreover, by adopting a tuple format encoding that includes frequency and time intervals, the deep signal detection model can accurately extract the local sub-segments to be analyzed from the broadband data, avoiding redundant calculations on invalid background areas.

[0060] Furthermore, by utilizing feedback control parameters that reflect the detection threshold, scan step size, and scan priority of the lightweight signal detection model, the output of the depth signal detection model is not only used for the final judgment but also as the basis for dynamically adjusting the current workflow. This enables the system to self-adjust and optimize based on the current environment (such as noise fluctuations and signal strength distribution), significantly improving the weak signal detection rate and the robustness of the system.

[0061] In practical design, both models can be trained using supervised training based on simulated or measured communication signal time-frequency datasets. Positive samples are time-frequency segments containing communication signals, while negative samples are pure noise time-frequency segments. The training objective is to minimize the binary classification cross-entropy loss. The lightweight signal detection model prioritizes high recall to ensure complete candidate coverage, while the deep signal detection model prioritizes high precision and high confidence calibration to ensure reliable final judgments. In actual deployment, both models can be continuously updated through online fine-tuning to adapt to dynamic changes in the channel environment.

[0062] Example 2: As another aspect of the present invention, a communication signal detection system based on multi-model collaboration is also provided, which corresponds to the communication signal detection method in Embodiment 1, with a focus on protecting the corresponding module architecture.

[0063] Specifically, the communication signal detection system in the preferred embodiment includes: The signal receiving and preprocessing module is used to receive communication radio frequency signals and convert them into broadband time-frequency data. A lightweight signal detection module is used to scan broadband time-frequency data to generate at least one candidate region suspected of containing communication signals, wherein the candidate region is encoded in a tuple format containing frequency intervals and time intervals; The depth signal detection module is used to extract the corresponding time-frequency sub-segments from the broadband time-frequency data according to the encoding of the candidate region, and to perform local detection on the time-frequency sub-segments to output the signal existence determination result and detection confidence.

[0064] By utilizing the combined configuration of the aforementioned modules, the synergistic cooperation between lightweight and advanced modules can be fully realized, enabling rapid location and detailed analysis of communication signals of unknown format or parameters within a wide bandwidth, improving the detection capability of communication signals in complex electromagnetic environments, and ensuring the real-time performance and accuracy of signal detection.

[0065] Furthermore, in actual implementation, the communication signal detection system in the preferred embodiment preferably includes an adaptive update module, which includes a feedback control subunit and a module update subunit, such as... Figure 4 As shown in the figure. Wherein: The feedback control subunit communicates with the depth signal detection module to generate feedback control parameters based on the detection confidence level output by the depth signal detection module.

[0066] The module update subunit is used to receive feedback control parameters and apply them to update at least one of the detection threshold, scan step size, or frequency band interval scan priority of the lightweight signal detection module.

[0067] More specifically, in actual setup, data interaction between the lightweight signal detection module and the depth signal detection module is preferably implemented through shared memory, internal bus, or cache pointers.

[0068] Meanwhile, in an optional example, the lightweight signal detection module and the depth signal detection module preferably run on different processing units, for example: the lightweight signal detection module runs on a general-purpose processor, and the depth signal detection module runs on a graphics processor or a dedicated accelerator.

[0069] In addition, the adaptive update module can update the lightweight signal detection module either online or offline.

[0070] Example 3: For the method in Embodiment 1, the method steps can be implemented by an electronic device executing program instructions, or by software, hardware, or a combination of both.

[0071] In this embodiment, a computer-readable storage medium is further provided, on which a computer program is stored, wherein when the program is executed by a processor, it implements the communication signal detection method based on multi-model collaboration as described in Embodiment 1.

[0072] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A communication signal detection method based on multi-model collaboration, characterized in that, Includes the following steps: S1: Receives communication radio frequency signals and converts them into broadband time-frequency data; S2: A lightweight signal detection model is used to scan the broadband time-frequency data to generate at least one candidate region that is suspected of containing communication signals; wherein, the candidate region is encoded in a tuple format that includes frequency intervals and time intervals; S3: Based on the coding of the candidate region, extract the corresponding time-frequency sub-segments from the broadband time-frequency data and input them into the depth signal detection model; S4: Use the depth signal detection model to perform local detection on the time-frequency sub-segment, and output the signal existence determination result and detection confidence level.

2. The communication signal detection method based on multi-model collaboration according to claim 1, characterized in that, In step S2, the lightweight signal detection model generates candidate regions using a candidate region generation mode based on an energy statistical threshold. This candidate region generation mode includes: Based on a preset energy threshold, broadband time-frequency data is scanned frame by frame, and frequency intervals where the energy continuously exceeds the energy threshold and time frame intervals that appear consecutively are encoded into tuple format containing frequency intervals and time intervals.

3. The communication signal detection method based on multi-model collaboration according to claim 1, characterized in that, In step S2, the lightweight signal detection model generates candidate regions using a candidate region generation mode based on a lightweight neural network. This candidate region generation mode includes: A shallow convolutional neural network is used to classify local windows of broadband time-frequency data, and time-frequency ranges with classification probabilities greater than a preset threshold are recorded as candidate regions.

4. The communication signal detection method based on multi-model collaboration according to any one of claims 1 to 3, characterized in that, The method also includes an adaptive detection and adjustment process: S5: Generate feedback control parameters based on the detection confidence output by the depth signal detection model; and use the feedback control parameters to update the lightweight signal detection model, forming a closed-loop adaptive detection adjustment process.

5. The communication signal detection method based on multi-model collaboration according to claim 4, characterized in that, Updating a lightweight signal detection model using feedback control parameters includes: Adjust at least one of the following in the lightweight signal detection model: detection threshold, scan step size, or frequency band priority, through online or offline updates.

6. The communication signal detection method based on multi-model collaboration according to claim 4, characterized in that, The adaptive detection adjustment process includes a detection threshold adjustment process, which includes: S51: Obtain the average detection confidence score output by the depth signal detection model; S52: Generate feedback control parameters for adjusting the detection threshold based on the average detection confidence level; where: If the average detection confidence is lower than the lower limit of the preset confidence threshold range, feedback control parameters are generated to reduce the detection threshold of the lightweight signal detection model. If the average detection confidence level is higher than the upper limit of the preset confidence threshold range, feedback control parameters are generated to improve the detection threshold of the lightweight signal detection model.

7. The communication signal detection method based on multi-model collaboration according to claim 4, characterized in that, The adaptive detection and adjustment process includes: In response to the number of candidate regions generated within a preset period exceeding the processing capacity threshold, feedback control parameters are generated to increase the temporal scan step size. and / or Based on the cumulative detection confidence of each frequency range within a preset period, the scanning priority of the corresponding frequency range is adjusted.

8. A communication signal detection system based on multi-model collaboration, characterized in that, include: The signal receiving and preprocessing module is used to receive communication radio frequency signals and convert them into broadband time-frequency data. A lightweight signal detection module is used to scan broadband time-frequency data to generate at least one candidate region suspected of containing communication signals, wherein the candidate region is encoded in a tuple format containing frequency intervals and time intervals; The depth signal detection module is used to extract the corresponding time-frequency sub-segments from the broadband time-frequency data according to the encoding of the candidate region, and to perform local detection on the time-frequency sub-segments to output the signal existence determination result and detection confidence.

9. The communication signal detection system based on multi-model collaboration according to claim 8, characterized in that, The system also includes an adaptive update module configured with a feedback control subunit and a module update subunit, wherein: The feedback control subunit is configured to communicate with the depth signal detection module to generate feedback control parameters based on the detection confidence level output by the depth signal detection module. The module update subunit is used to receive feedback control parameters and apply them to update at least one of the detection threshold, scan step size, or frequency band interval scan priority of the lightweight signal detection module.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the communication signal detection method based on multi-model collaboration as described in any one of claims 1 to 7.

Images