Deep learning-based shortwave signal detection, identification and training method, device and equipment, and storage medium

Abstract

The invention discloses a short wave signal detection, identification and training method based on deep learning. The method comprises the following steps: extracting IQ data of a narrowband signal from target signal data through a deep learning model; the modulation features and protocol features of the narrowband IQ data are analyzed through a deep learning model, and signal modulation mode identification and signal protocol type identification are carried out; storing all processed signal data and related label information; labeling the signal data based on a deep learning model and a preset rule, and storing a finally confirmed label as a final labeling result in a database; and collecting labeled data from the database, performing learning model training and parameter adjustment based on a deep learning framework, and performing retraining by taking a first time interval as a period. The invention also discloses a short wave signal detection, identification and training device based on deep learning, corresponding equipment and a storage medium. According to the invention, the short-wave signal can be efficiently and accurately detected and identified.

Description

Technical Field

[0001] The present application relates to the field of signal detection technology, and more specifically, to a shortwave signal detection, recognition and training method, device, equipment and storage medium based on deep learning. Background Art

[0002] Traditional shortwave signal detection and recognition systems usually rely on preset algorithms and models for signal processing and recognition. These systems often lack flexibility and efficiency when faced with new or changing shortwave signal types. Especially when the shortwave signal environment is complex or constantly changing, it is difficult for existing systems to adapt and update in a timely manner, thus affecting overall performance and reliability. With the continuous development of wireless communication technology, the types and characteristics of shortwave signals are becoming more and more diverse and complex. Traditional detection and recognition methods can no longer meet the needs of actual applications. Therefore, a signal detection and recognition system that can adaptively adjust and continuously optimize is needed. Summary of the invention

[0003] In response to at least one defect or improvement need in the prior art, the present invention provides a shortwave signal detection, recognition and training method, device, equipment and storage medium based on deep learning, which can solve at least one of the problems existing in the above-mentioned background technology.

[0004] To achieve the above objectives, according to a first aspect of the present invention, a shortwave signal detection, recognition and training method based on deep learning is provided, the method comprising:

[0005] The frequency, bandwidth, start time and end time information of the input target signal are obtained through the deep learning model, and the IQ data of the narrowband signal is extracted from the target signal data;

[0006] Perform signal modulation mode identification and signal protocol type identification, analyze the modulation characteristics of narrowband IQ data through a deep learning model, output the probability that the signal belongs to each modulation mode, and use the modulation mode with the highest probability as the prediction result; analyze the protocol characteristics of narrowband IQ data through a deep learning model, output the probability distribution of the signal belonging to each protocol type, and use the protocol type with the highest probability as the prediction result;

[0007] storing all processed signal data and related label information, wherein the processed signal data and related label information include generated time-frequency data and extracted narrowband signal data;

[0008] Label the signal data based on the deep learning model and preset rules, and store the final confirmed labels as the final labeling results in the database;

[0009] Collect labeled data from the database, perform learning model training and parameter adjustment based on the deep learning framework, and retrain at a first time interval as a period.

[0010] Furthermore, the above-mentioned shortwave signal detection, recognition and training method based on deep learning, wherein the frequency, bandwidth, start time and end time information of the input target signal are obtained through the deep learning model, and the IQ data of the narrowband signal is extracted from the target signal data, specifically includes:

[0011] Based on the input target signal IQ data, a time-frequency graph is generated by fast Fourier transform, and the generated time-frequency graph is preprocessed by overlapping and segmenting in frequency and time. The preprocessed time-frequency graph is input into a pre-trained deep learning target detection model, and the detected bounding box of the target signal area and its confidence score are output, and non-maximum resistance is used to eliminate duplicate bounding boxes;

[0012] Calculate the frequency, bandwidth, start time and end time information of the target signal for the detected bounding box of the target signal;

[0013] Based on the frequency, bandwidth, start time and end time information of the target signal, narrowband IQ data of the target signal is extracted from the IQ data of the target signal.

[0014] Furthermore, the above-mentioned shortwave signal detection, identification and training method based on deep learning, the signal modulation mode identification and signal protocol type identification specifically include:

[0015] Extracting modulation features from the extracted narrowband IQ data, where the extracted features are the mode angle sequence of the narrowband IQ data, performing standardization processing on the mode angle sequence, inputting the processed mode angle sequence into a deep learning model for inference prediction, and outputting a probability distribution of the modulation mode;

[0016] According to the probability distribution of the modulation mode output by the model, the modulation mode of the signal is classified, and the modulation mode with the highest probability is taken as the prediction result;

[0017] Extracting protocol features from the extracted narrowband IQ data, where the extracted features are the mode angle sequence of the narrowband IQ data, performing standardization processing on the mode angle sequence, inputting the processed mode angle sequence into a deep learning model for inference prediction, and outputting a probability distribution of the protocol type;

[0018] According to the probability distribution of the protocol type output by the model, the signal is classified into protocol types, and the protocol type with the highest probability is taken as the prediction result.

[0019] Furthermore, in the above-mentioned shortwave signal detection, recognition and training method based on deep learning, the processed signal data and related label information include generated time-frequency data and extracted narrowband signal data, specifically including:

[0020] The generated time-frequency data includes time-frequency graph image data, a detection frame and confidence score of the time-frequency graph, specific parameters of the generated time-frequency graph, and signal receiving environment information;

[0021] The extracted narrowband signal data includes IQ data of the narrowband signal, identified signal protocol type, identified signal modulation mode, center frequency, bandwidth, start and end time, duration and signal receiving environment information of the signal.

[0022] Furthermore, the above-mentioned shortwave signal detection, recognition and training method based on deep learning, wherein the signal data is labeled based on the deep learning model and preset rules, specifically includes:

[0023] Generate preliminary labels, including label mapping and confidence scores;

[0024] Processing and confirming the labels based on an optimization strategy, the optimization strategy including context analysis, application of prior knowledge, multi-model fusion, and manual correction;

[0025] Final confirmation is made on the labels, and the final confirmed labels are stored in the database as final labeling results;

[0026] The tag database is updated periodically at a second time interval, and tags are supplemented and updated based on new signal types and modulation modes.

[0027] Furthermore, the above shortwave signal detection, recognition and training method based on deep learning also includes data collection and preprocessing before learning model training and parameter adjustment based on the deep learning framework:

[0028] Collect annotated time-frequency diagrams and narrowband data from the database;

[0029] The collected time-frequency diagram and narrowband data are preprocessed, wherein the preprocessing of the time-frequency diagram includes cropping and scaling, spectrum standardization and image enhancement, and the preprocessing of the narrowband data includes filtering, normalization and feature extraction.

[0030] Furthermore, the above-mentioned shortwave signal detection, recognition and training method based on deep learning, the learning model training and parameter adjustment based on the deep learning framework specifically include:

[0031] The learning model includes a signal detection model, a modulation recognition model and a protocol recognition model. A corresponding deep learning network is selected for each of the learning models, a data set is constructed, model training is performed in a deep learning framework, and model parameters are adjusted to respectively minimize the error between the detection frame and the true label, improve the recognition accuracy of the modulation mode, and improve the recognition accuracy of the protocol type;

[0032] The trained models are evaluated and cross-validated to select the best model.

[0033] According to a second aspect of the present invention, there is also provided a shortwave signal detection, recognition and training device based on deep learning, which comprises:

[0034] A signal detection and extraction module, used to obtain the frequency, bandwidth, start time and end time information of the input target signal through a deep learning model, and extract the IQ data of the narrowband signal from the target signal data;

[0035] The signal recognition module is used to identify the signal modulation mode and signal protocol type. The modulation characteristics of the narrowband IQ data are analyzed through a deep learning model, and the probability of the output signal belonging to each modulation mode is used. The modulation mode with the highest probability is used as the prediction result. The protocol characteristics of the narrowband IQ data are analyzed through a deep learning model, and the probability distribution of the output signal belonging to each protocol type is used. The protocol type with the highest probability is used as the prediction result.

[0036] A data storage module, used to store all processed signal data and related label information, wherein the processed signal data and related label information include generated time-frequency data and extracted narrowband signal data;

[0037] Intelligent labeling module, used to label signal data based on deep learning models and preset rules, and store the final confirmed labels as final labeling results in the database;

[0038] The model training and updating module is used to collect labeled data from the database, perform learning model training and parameter adjustment based on the deep learning framework, and perform retraining at a first time interval as a period.

[0039] According to the third aspect of the present invention, there is also provided a shortwave signal detection, identification and training device based on deep learning, which comprises at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit executes the steps of any one of the above methods.

[0040] According to the fourth aspect of the present invention, a storage medium is also provided, which stores a computer program that can be executed by a shortwave signal detection, recognition and training device based on deep learning. When the computer program runs on the shortwave signal detection, recognition and training device based on deep learning, the shortwave signal detection, recognition and training device based on deep learning executes the steps of any one of the above methods.

[0041] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0042] The shortwave signal detection, recognition and training method based on deep learning provided by the present invention continuously improves system performance through intelligent labeling and iterative training, is suitable for use in radio communication intelligent reconnaissance systems, can improve the intelligence level of the system, and can efficiently and accurately detect and recognize shortwave signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A flowchart of a shortwave signal detection, recognition and training method based on deep learning provided in an embodiment of the present application;

[0045] Figure 2 A schematic diagram of a specific process of a shortwave signal detection, recognition and training method based on deep learning provided in an embodiment of the present application;

[0046] Figure 3 A schematic diagram of the structure of a shortwave signal detection, recognition and training device based on deep learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0048] The terms "first", "second", "third", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0049] Figure 1 A flowchart of a shortwave signal detection, recognition and training method based on deep learning provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the embodiment of the present application provides a shortwave signal detection, recognition and training method based on deep learning, including:

[0050] 101. Obtain the frequency, bandwidth, start time, and end time information of the input target signal through the deep learning model, and extract the IQ data of the narrowband signal from the target signal data;

[0051] 102. Perform signal modulation mode identification and signal protocol type identification, analyze the modulation characteristics of narrowband IQ data through a deep learning model, output the probability that the signal belongs to each modulation mode, and use the modulation mode with the highest probability as the prediction result, analyze the protocol characteristics of narrowband IQ data through a deep learning model, output the probability distribution of the signal belonging to each protocol type, and use the protocol type with the highest probability as the prediction result;

[0052] 103. Storing all processed signal data and related label information, where the processed signal data and related label information include generated time-frequency data and extracted narrowband signal data;

[0053] 104. Label the signal data based on the deep learning model and preset rules, and store the finally confirmed labels as the final labeling results in the database;

[0054] 105. Collect labeled data from the database, perform learning model training and parameter adjustment based on the deep learning framework, and perform retraining at the first time interval as a period.

[0055] Specifically, the target signal in the embodiment of the present application is a shortwave time-domain broadband signal. The input shortwave time-domain broadband signal data is analyzed and processed by a deep learning model, a time-frequency diagram is generated by a fast Fourier transform, and the time-frequency diagram is preprocessed to prevent the image from being truncate and causing the signal feature to be lost. The preprocessed time-frequency diagram is input into a pre-trained deep learning target detection model, and the model outputs the bounding box of the detected signal area and its confidence score. The bounding box and confidence score output by the model are post-processed, and non-maximum resistance is used to detect repeated messages and retain the optimal detection results. Subsequently, the frequency, bandwidth, start time and end time information of the target signal are accurately calculated, and the IQ data of the target narrowband signal is extracted from the original signal data.

[0056] Signal modulation mode identification and signal protocol type identification are performed. Modulation mode identification uses a deep learning model to analyze the modulation characteristics (mode angle sequence) of narrowband IQ data. The modulation characteristic data is input into the model after standardization. The probability of the output signal belonging to each modulation mode is selected as the prediction result. Protocol type identification uses a deep learning model to analyze the protocol characteristics (mode angle sequence) of narrowband IQ data. The protocol characteristic data is input into the model after standardization. The probability distribution of the output signal belonging to each protocol type is selected as the prediction result.

[0057] Stores all processed signal data and related label information, including generated time-frequency diagram data and extracted narrowband signal data. The stored content includes image data, detection labels, IQ data, modulation mode, protocol type and other related information (such as the center frequency, bandwidth, start and end time of the signal and the signal receiving environment).

[0058] The signal data is annotated using deep learning models and preset rules. Preliminary labels are first generated and confidence scores are generated for each label. Then, labels are optimized and confirmed through strategies such as context analysis, application of prior knowledge, multi-model fusion, and manual correction to ensure the accuracy and consistency of labels. The final confirmed labels will be stored as the final annotation results.

[0059] Retraining is performed periodically at the first time interval to realize the training and updating of signal detection, modulation mode recognition and protocol recognition models. The labeled data is collected from the database and preprocessed to narrow the band and ensure data consistency and quality. Model training includes network selection, data set construction and training process, and a deep learning framework is used for model training and parameter adjustment. The model evaluation and cross-validation steps evaluate the model parameters through evaluation indicators such as accuracy, recall rate and mAP, and select the best model. The model update step continuously updates the model parameters through incremental training to adapt to new signal patterns and features, and performs comprehensive retraining regularly to ensure that the model maintains optimal performance. The first time interval can be determined according to actual engineering needs, and this application does not make specific limitations here.

[0060] The shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application continuously improves system performance through intelligent labeling and iterative training, is applicable to radio communication intelligent reconnaissance systems, can improve the intelligence level of the system, and can efficiently and accurately detect and recognize shortwave signals.

[0061] Optionally, the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application obtains the frequency, bandwidth, start time and end time information of the input target signal through the deep learning model, and extracts the IQ data of the narrowband signal from the target signal data, specifically including:

[0062] Based on the input target signal IQ data, a time-frequency graph is generated through fast Fourier transform, and the generated time-frequency graph is preprocessed by overlapping and segmenting in frequency and time. The preprocessed time-frequency graph is input into the pre-trained deep learning target detection model, and the bounding box of the detected target signal area and its confidence score are output. Non-maximum resistance is used to eliminate duplicate bounding boxes.

[0063] Calculate the frequency, bandwidth, start time and end time information of the target signal for the bounding box of the detected target signal;

[0064] Based on the frequency, bandwidth, start time and end time information of the target signal, narrowband IQ data of the target signal is extracted from the IQ data of the target signal.

[0065] Specifically, Figure 2 A specific flow chart of the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application is as follows: Figure 2As shown, using the input shortwave time-domain broadband signal IQ data (IQ data is expressed in complex form, where I represents the in-phase component In-phase, and Q represents the orthogonal component Quadrature), preferably, the input shortwave time-domain broadband signal IQ data is real-time signal data or offline file data, and a time-frequency diagram is generated by fast Fourier transform to show the distribution of the signal in time and frequency. The generated time-frequency diagram is preprocessed by overlapping segmentation in frequency and time to prevent the loss of important signal features caused by image truncation. The preprocessed time-frequency diagram is input into the pre-trained deep learning target detection model, and the model outputs the bounding box of the detected target signal area and its confidence score. The bounding box and confidence score output by the model are post-processed, and non-maximum resistance is used to eliminate duplicate detection boxes and retain the optimal detection results.

[0066] The detected bounding box of the target signal area is further processed to accurately calculate the frequency, bandwidth, start time and end time information of the target signal through the detection box.

[0067] According to the calculated frequency, bandwidth, sampling rate, start time and end time information of the target signal, the IQ data of the narrowband part of the target signal is extracted from the shortwave time domain broadband signal IQ data. Preferably, the specific steps of the above extraction method are: converting a time domain broadband signal of a certain length into a frequency domain signal sequence by fast Fourier transform, extracting the frequency domain signal data of the target signal from the above frequency domain signal sequence according to the frequency, bandwidth and sampling rate of the target signal, and performing windowing filtering to filter other frequency information outside the target signal frequency, and the frequency domain signal sequence after windowing filtering is subjected to inverse fast Fourier transform to obtain the narrowband IQ data of the target signal.

[0068] Optionally, the shortwave signal detection, identification and training method based on deep learning provided in the embodiment of the present application performs signal modulation mode identification and signal protocol type identification, specifically including:

[0069] Extract modulation features from the extracted narrowband IQ data. The extracted features are the mode angle sequence of the narrowband IQ data. The mode angle sequence is standardized and input into the deep learning model for inference and prediction, and the probability distribution of the modulation mode is output.

[0070] According to the probability distribution of the modulation mode output by the model, the modulation mode of the signal is classified, and the modulation mode with the highest probability is taken as the prediction result;

[0071] Extract protocol features from the extracted narrowband IQ data. The extracted features are the mode angle sequence of the narrowband IQ data. The mode angle sequence is standardized and input into the deep learning model for inference and prediction, and the probability distribution of the protocol type is output.

[0072] According to the probability distribution of the protocol type output by the model, the signal is classified into protocol types, and the protocol type with the highest probability is taken as the prediction result.

[0073] Specifically, Figure 2 A specific flow chart of the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application is as follows: Figure 2 As shown, the signal modulation mode recognition is realized by using the modulation recognition model to analyze the modulation mode of the signal and obtain the modulation information, including the following steps:

[0074] The modulation feature is extracted from the extracted narrowband IQ data, and the feature is the mode angle sequence of the narrowband IQ data, that is, each complex number is represented as its amplitude (mode) and phase (angle). Preferably, the mode angle sequence is adjusted to a fixed length, and zero padding is performed if the signal length is insufficient, and truncation is performed if the signal length exceeds.

[0075] The mode angle sequence is normalized to ensure that its characteristics are consistent with the training data. Preferably, the normalization is performed by normalizing to a certain range or subtracting the mean and dividing by the standard deviation, consistent with the processing method used in the model training process.

[0076] The standardized mode angle sequence is input into the deep learning model for inference and prediction. The model outputs a probability distribution, which indicates the probability of the input signal belonging to each modulation mode. The number of types contained in the output probability distribution is the same as the number of modulation modes used in the model training process. Assuming that the model uses N different modulation modes during training, the output probability distribution will contain N values, each corresponding to the predicted probability of one of the modulation modes.

[0077] According to the output of the model, the signal is classified into modulation modes, and the modulation mode with the highest probability is selected as the prediction result to determine the modulation mode used by the signal (such as AM, FM, QAM, PSK, etc.). Preferably, a confidence threshold can be set, and if the highest probability is lower than the threshold, the prediction is considered unreliable.

[0078] The signal protocol type identification process is basically the same as the signal modulation method identification. The mode angle sequence after data standardization is input into the deep learning model for inference and prediction. The model outputs a probability distribution, which represents the probability that the input signal belongs to each protocol type. The number of types contained in the output probability distribution is the same as the number of protocol types used in the model training process. Assuming that the model uses M different protocol types during training, the output probability distribution will contain M values, each corresponding to the predicted probability of one of the protocols.

[0079] According to the output of the model, the signal is classified into protocol types, and the protocol type with the highest probability is selected as the prediction result to determine which communication protocol the signal uses (such as 2G-ALE, 3G-ALE, ANDVT, etc.). Preferably, a confidence threshold can be set, and if the highest probability is lower than the threshold, the prediction is considered unreliable.

[0080] Optionally, the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application, the processed signal data and related label information include generated time-frequency data and extracted narrowband signal data, specifically including:

[0081] The generated time-frequency data includes the time-frequency graph image data, the detection frame and confidence score of the time-frequency graph, the specific parameters of the generated time-frequency graph and the receiving environment information of the signal;

[0082] The extracted narrowband signal data includes IQ data of the narrowband signal, identified signal protocol type, identified signal modulation mode, center frequency, bandwidth, start and end time, duration and signal receiving environment information of the signal.

[0083] Specifically, Figure 2 A specific flow chart of the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application is as follows: Figure 2 As shown in the figure, the frequency-time graph data storage realizes storing the frequency-time graph and related label information generated by the signal detection part into the database, including: image data, the frequency-time graph image data generated by the signal detection part. Detection label, detection box and confidence score related to the frequency-time graph, used to identify the frequency point and start and end time of the signal. Other related information, including specific parameters for generating the frequency-time graph and environmental information for signal reception, etc.

[0084] The narrowband data storage realizes storing the narrowband data extracted by the signal detection and extraction part and related label information into the database, including: IQ data, IQ data of the narrowband signal extracted from the broadband shortwave signal data. Protocol type, the identified signal protocol type. Modulation mode, the identified signal modulation mode. Other related information, the center frequency, bandwidth, start and end time, duration and signal receiving environment information of the signal.

[0085] Optionally, the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application labels the signal data based on the deep learning model and preset rules, specifically including:

[0086] Generate preliminary labels, including label mapping and confidence scores;

[0087] Process and confirm labels based on optimization strategies, including context analysis, application of prior knowledge, multi-model fusion, and manual correction;

[0088] Final confirmation is made on the labels, and the final confirmed labels are stored in the database as final labeling results;

[0089] The tag database is updated periodically at a second time interval, and tags are supplemented and updated based on new signal types and modulation modes.

[0090] Specifically, Figure 2 A specific flow chart of the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application is as follows: Figure 2 As shown, the following steps are included: preliminary label formation, including label mapping and confidence scoring. Label mapping, based on the signal detection results, uses the label database or the preset label set to automatically generate preliminary labels for the time-frequency graph, including detection box information; combines the signal recognition results and the preset label set to generate preliminary labels for narrowband data, including protocol type and modulation mode. For example, the typical modulation mode of a certain type of signal can be directly used for preliminary label generation. Confidence scoring, generates a corresponding confidence score for each preliminary label for subsequent optimization and confirmation. The confidence score reflects the degree of certainty about the label.

[0091] Label optimization: After the initial label is generated, the label is further processed and confirmed through a variety of optimization strategies. Including: context analysis: Combine the context information of time and frequency of the time-frequency diagram and narrowband data to comprehensively analyze and optimize the initial label. Preferably, combined with the confidence fusion method, if the signal recognition results of the same frequency point have different protocol types or modulation methods at adjacent times, the part with high confidence may be more likely. When optimizing the label, more reference is made to the high-confidence label. Prior knowledge application: Optimize the preliminary label using preset label rules and prior knowledge. If the modulation method of the signal does not match the known communication protocol, the label needs to be further reviewed. Multi-model fusion: Use the prediction results of multiple deep learning models to fuse and improve the accuracy and stability of the label. By combining the advantages of different models, the errors that may be caused by a single model are reduced. Preferably, when multiple deep learning models give different prediction results, the confidence score is used as a weighting factor for fusion, and the prediction results of the model with high confidence have a greater weight in the final result. Manual correction, based on automatic labeling, allows experts to manually correct the labels to further improve the accuracy and reliability of the labeling. Preferably, experts prioritize low-confidence labels based on confidence scores to improve overall labeling efficiency and quality.

[0092] Label confirmation, after multiple optimization strategies, the labels are finally confirmed to ensure their accuracy and consistency. The confirmed labels will be stored as the final annotation results.

[0093] Tag management ensures the completeness and accuracy of the tag set by regularly updating the tag database. With the emergence of new signal types and modulation methods, the tags are supplemented and updated in a timely manner. The second time interval can be determined according to engineering requirements, and this application does not make specific limitations here.

[0094] Optionally, the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application further includes data collection and preprocessing before performing learning model training and parameter adjustment based on the deep learning framework:

[0095] Collect annotated time-frequency diagrams and narrowband data from the database;

[0096] The collected time-frequency diagram and narrowband data are preprocessed. The preprocessing of the time-frequency diagram includes cropping and scaling, spectrum normalization and image enhancement, and the preprocessing of the narrowband data includes filtering, normalization and feature extraction.

[0097] Specifically, Figure 2 A specific flow chart of the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application is as follows: Figure 2 As shown in Figure 1, the labeled time-frequency diagram and narrowband data are collected from the database as samples for model training. The collected data are preprocessed to ensure the consistency and quality of the data.

[0098] The preprocessing operations of the time-frequency graph include the following aspects: cropping and scaling, cropping the time-frequency graph according to the detection frame, and scaling as needed to unify the image size and ratio, ensuring that the essence of the image is not changed to avoid losing signal features. Spectral normalization, standardizing the time-frequency graph to ensure that all signal features are compared at the same scale. Image enhancement, applying random cropping, translation and other techniques to increase the diversity of training data and improve the robustness of the model.

[0099] The preprocessing operations of narrowband data include the following aspects: filtering, applying a bandpass filter to the narrowband data to remove interference from the signal and retain the useful signal part. Normalization, normalizing the narrowband data to unify the amplitude of the data. Feature extraction, extracting the mode angle sequence of IQ data from the narrowband data.

[0100] Optionally, the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application performs learning model training and parameter adjustment based on a deep learning framework, specifically including:

[0101] The learning model includes a signal detection model, a modulation recognition model, and a protocol recognition model. The corresponding deep learning network is selected for each learning model, a data set is constructed, and the model is trained in a deep learning framework. The model parameters are adjusted to minimize the error between the detection box and the true label, improve the recognition accuracy of the modulation mode, and improve the recognition accuracy of the protocol type.

[0102] The trained models are evaluated and cross-validated to select the best model.

[0103] Specifically, Figure 2 A specific flow chart of the shortwave signal detection, recognition and training method based on deep learning provided in the embodiment of the present application is as follows: Figure 2 As shown, the model is trained to realize the training of signal detection, modulation mode recognition model and protocol type recognition model.

[0104] The training of the signal detection model includes the following steps: Network selection, selecting a deep learning network suitable for signal detection, preferably using a YOLO (You Only Look Once) series network or a DETR (Detection Transformer) network. Dataset construction, using the preprocessed time-frequency graph and its detection box labels as the model training data set and validation data set. Training process, training the model in a deep learning framework, adjusting the model parameters to minimize the error between the detection box and the true label. Preferably, the deep learning framework is PyTorch.

[0105] The training of the modulation recognition model includes the following steps: network selection, selecting a deep learning network suitable for modulation recognition, preferably using a recurrent neural network (RNN) or a long short-term memory network (LSTM). Data set construction, using preprocessed narrowband data and its modulation mode labels as model training data sets and verification data sets. Training process, model training is performed in a deep learning framework, and model parameters are optimized to improve the recognition accuracy of the modulation mode. Preferably, the deep learning framework is PyTorch or TensorFlow.

[0106] The training of the protocol identification model includes the following steps: network selection, selecting a deep learning network suitable for protocol identification, preferably using a convolutional neural network (CNN) or a transfomer network. Data set construction, using preprocessed narrowband data and its protocol type as a model training data set and a validation data set. Training process, model training in a deep learning framework, optimizing model parameters to improve the recognition accuracy of protocol types, preferably, the deep learning framework is PyTorch.

[0107] Model evaluation and cross-validation, use accuracy, recall, mAP (mean of Average Precision) and other indicators to evaluate the performance of the model and select the best model. Use cross-validation method to evaluate the performance of the model and ensure the generalization ability of the model through multiple training and testing.

[0108] Model update: Use newly collected and annotated data for incremental model training, and continuously update model parameters to adapt to new signal patterns and features. Regularly retrain the model, using all the latest data for comprehensive training to ensure that the model maintains optimal performance. Feed the trained model back to the signal detection and recognition system to replace the old model, achieving real-time model updates and continuous optimization of system performance.

[0109] Model management manages and records different versions of models, saves model training parameters, performance indicators, and training data set information for easy traceability and review. In actual applications, the running status of the reference model is monitored to detect changes in model performance and promptly identify and handle potential problems.

[0110] Through the functional implementation of the above modules, the signal detection, recognition and training system of this intellectual achievement can detect and recognize signals efficiently and accurately, and continuously improve system performance through intelligent labeling and iterative training.

[0111] The present application also provides a shortwave signal detection, recognition and training device based on deep learning, such as Figure 3 As shown, including:

[0112] The signal detection and extraction module is used to obtain the frequency, bandwidth, start time and end time information of the input target signal through the deep learning model, and extract the IQ data of the narrowband signal from the target signal data;

[0113] The signal recognition module is used to identify the signal modulation mode and signal protocol type. The modulation characteristics of the narrowband IQ data are analyzed through a deep learning model, and the probability of the output signal belonging to each modulation mode is used. The modulation mode with the highest probability is used as the prediction result. The protocol characteristics of the narrowband IQ data are analyzed through a deep learning model, and the probability distribution of the output signal belonging to each protocol type is used. The protocol type with the highest probability is used as the prediction result.

[0114] A data storage module, used to store all processed signal data and related label information, the processed signal data and related label information including generated time-frequency data and extracted narrowband signal data;

[0115] Intelligent labeling module, used to label signal data based on deep learning models and preset rules, and store the final confirmed labels as final labeling results in the database;

[0116] The model training and updating module is used to collect labeled data from the database, perform learning model training and parameter adjustment based on the deep learning framework, and perform retraining at a first time interval as a period.

[0117] The present application also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented. The computer-readable storage medium may include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive, and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.

[0118] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0119] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0120] In the several embodiments provided in the present application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0121] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0122] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0123] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a memory, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned memory includes: U disk, read-only memory (ROM), random access memory (RAM), mobile hard disk, disk or optical disk and other media that can store program code.

[0124] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable memory, which can include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0125] The above is only an exemplary embodiment of the present disclosure, and the scope of the present disclosure cannot be limited thereto. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the specification and practicing the disclosure here, those skilled in the art will easily think of the implementation scheme of the present disclosure. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the technical field not recorded in the present disclosure. The description and examples are regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

[0126] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0127] It will be easily understood by those skilled in the art that the above description is only 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 in the protection scope of the present invention.

Claims

1. A shortwave signal detection, recognition and training method based on deep learning, characterized in that: The steps include: The frequency, bandwidth, start time and end time information of the input target signal are obtained through the deep learning model, and the IQ data of the narrowband signal is extracted from the target signal data; Perform signal modulation mode identification and signal protocol type identification, analyze the modulation characteristics of narrowband IQ data through a deep learning model, output the probability that the signal belongs to each modulation mode, and take the modulation mode with the highest probability as the prediction result; analyze the protocol characteristics of narrowband IQ data through a deep learning model, output the probability distribution of the signal belonging to each protocol type, and take the protocol type with the highest probability as the prediction result; storing all processed signal data and related label information, wherein the processed signal data and related label information include generated time-frequency data and extracted narrowband signal data; Label the signal data based on the deep learning model and preset rules, and store the final confirmed labels as the final labeling results in the database; Collect labeled data from the database, perform learning model training and parameter adjustment based on the deep learning framework, and retrain at a first time interval as a period.

2. The shortwave signal detection, recognition and training method based on deep learning as claimed in claim 1, characterized in that: The step of obtaining the frequency, bandwidth, start time, and end time information of the input target signal through the deep learning model, and extracting the IQ data of the narrowband signal from the target signal data, specifically includes: Based on the input target signal IQ data, a time-frequency graph is generated by fast Fourier transform, and the generated time-frequency graph is preprocessed by overlapping and segmenting in frequency and time. The preprocessed time-frequency graph is input into a pre-trained deep learning target detection model, and the detected bounding box of the target signal area and its confidence score are output, and non-maximum resistance is used to eliminate duplicate bounding boxes; Calculate the frequency, bandwidth, start time and end time information of the target signal for the detected bounding box of the target signal; Based on the frequency, bandwidth, start time and end time information of the target signal, narrowband IQ data of the target signal is extracted from the IQ data of the target signal.

3. The shortwave signal detection, recognition and training method based on deep learning as claimed in claim 1, characterized in that: The signal modulation mode identification and signal protocol type identification specifically include: Extracting modulation features from the extracted narrowband IQ data, where the extracted features are the mode angle sequence of the narrowband IQ data, performing standardization processing on the mode angle sequence, inputting the processed mode angle sequence into a deep learning model for inference prediction, and outputting a probability distribution of the modulation mode; According to the probability distribution of the modulation mode output by the model, the modulation mode of the signal is classified, and the modulation mode with the highest probability is taken as the prediction result; Extracting protocol features from the extracted narrowband IQ data, where the extracted features are the mode angle sequence of the narrowband IQ data, performing standardization processing on the mode angle sequence, inputting the processed mode angle sequence into a deep learning model for inference prediction, and outputting a probability distribution of the protocol type; According to the probability distribution of the protocol type output by the model, the signal is classified into protocol types, and the protocol type with the highest probability is taken as the prediction result.

4. The shortwave signal detection, recognition and training method based on deep learning as claimed in claim 1, characterized in that: The processed signal data and related label information include generated time-frequency data and extracted narrowband signal data, specifically including: The generated time-frequency data includes time-frequency graph image data, a detection frame and confidence score of the time-frequency graph, specific parameters of the generated time-frequency graph, and signal receiving environment information; The extracted narrowband signal data includes IQ data of the narrowband signal, identified signal protocol type, identified signal modulation mode, center frequency, bandwidth, start and end time, duration and signal receiving environment information of the signal.

5. The shortwave signal detection, recognition and training method based on deep learning as claimed in claim 1, characterized in that: The labeling of signal data based on the deep learning model and preset rules specifically includes: Generate preliminary labels, including label mapping and confidence scores; Processing and confirming the labels based on an optimization strategy, the optimization strategy including context analysis, application of prior knowledge, multi-model fusion, and manual correction; Final confirmation is made on the labels, and the final confirmed labels are stored in the database as final labeling results; The tag database is updated periodically at a second time interval, and tags are supplemented and updated based on new signal types and modulation modes.

6. The shortwave signal detection, recognition and training method based on deep learning as claimed in claim 1, characterized in that: Before learning model training and parameter adjustment based on the deep learning framework, data collection and preprocessing are also included: Collect annotated time-frequency diagrams and narrowband data from the database; The collected time-frequency diagram and narrowband data are preprocessed, wherein the preprocessing of the time-frequency diagram includes cropping and scaling, spectrum standardization and image enhancement, and the preprocessing of the narrowband data includes filtering, normalization and feature extraction.

7. The shortwave signal detection, recognition and training method based on deep learning as claimed in claim 1, characterized in that: The learning model training and parameter adjustment based on the deep learning framework specifically include: The learning model includes a signal detection model, a modulation recognition model and a protocol recognition model. A corresponding deep learning network is selected for each of the learning models, a data set is constructed, model training is performed in a deep learning framework, and model parameters are adjusted to respectively minimize the error between the detection frame and the true label, improve the recognition accuracy of the modulation mode, and improve the recognition accuracy of the protocol type; The trained models are evaluated and cross-validated to select the best model.

8. A shortwave signal detection, recognition and training device based on deep learning, characterized in that: include: A signal detection and extraction module, used to obtain the frequency, bandwidth, start time and end time information of the input target signal through a deep learning model, and extract the IQ data of the narrowband signal from the target signal data; The signal recognition module is used to identify the signal modulation mode and signal protocol type. The modulation characteristics of the narrowband IQ data are analyzed through a deep learning model, and the probability of the output signal belonging to each modulation mode is used. The modulation mode with the highest probability is used as the prediction result. The protocol characteristics of the narrowband IQ data are analyzed through a deep learning model, and the probability distribution of the output signal belonging to each protocol type is used. The protocol type with the highest probability is used as the prediction result. A data storage module, used to store all processed signal data and related label information, wherein the processed signal data and related label information include generated time-frequency data and extracted narrowband signal data; Intelligent labeling module, used to label signal data based on deep learning models and preset rules, and store the final confirmed labels as final labeling results in the database; The model training and updating module is used to collect labeled data from the database, perform learning model training and parameter adjustment based on the deep learning framework, and perform retraining at a first time interval as a period.

9. A shortwave signal detection, recognition and training device based on deep learning, characterized in that: The method comprises at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit executes the steps of the method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: It stores a computer program that can be executed by a shortwave signal detection, recognition and training device based on deep learning. When the computer program runs on the shortwave signal detection, recognition and training device based on deep learning, the shortwave signal detection, recognition and training device based on deep learning executes the steps of the method described in any one of claims 1 to 7.

Images