A deep learning method and system for electromagnetic signal classification and recognition

Abstract

This invention discloses a deep learning method and system for electromagnetic signal classification and recognition. It involves constructing an electromagnetic signal dataset, where each dataset contains multi-dimensional electromagnetic signal features and corresponding label information; selecting and configuring a corresponding deep learning model; labeling the obtained electromagnetic signal dataset with contamination, including Y-level individual contamination and M-matrix level contamination. Y-level contamination involves randomly selecting a specific proportion of data points and modifying their labels, while M-matrix level contamination involves systematically modifying the labels of all or part of the dataset by applying a predefined contamination matrix; training the deep learning model using the contaminated electromagnetic signal dataset; and inputting the uncontaminated electromagnetic signal dataset into the deep learning model to output the identified electromagnetic signals. This method effectively addresses different levels of data contamination, ensuring optimal classification performance under various environments.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, specifically relating to a deep learning method and system for electromagnetic signal classification and recognition. Background Technology

[0002] Electromagnetic signal classification is a crucial technical field with wide applications in communications, radar monitoring, and wireless spectrum management. The core of this technology lies in its ability to accurately identify and classify electromagnetic signals collected from different sources. Traditional signal classification methods typically rely on feature engineering and machine learning algorithms, such as Support Vector Machines (SVMs) and traditional neural networks. These methods perform well when signal characteristics are readily apparent and the environment is relatively stable. However, as the complexity of the signal environment increases, especially when interference and noise are present, the performance of these traditional methods often suffers.

[0003] In practical applications, especially in environments with severe noise or interference, existing technologies often suffer significant performance degradation due to their inability to effectively handle tag errors or contamination. Furthermore, existing solutions have significant shortcomings in utilizing information from multi-channel signals, dynamically adjusting for mislabeling, and adapting to complex contaminated environments, which limits their effectiveness in high-noise environments.

[0004] Existing Implementation Schemes: In recent years, the introduction of deep learning technology has brought new possibilities to electromagnetic signal classification. Deep neural networks, especially convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have been widely studied and applied due to their powerful automatic feature extraction capabilities. These methods can achieve more accurate classification in complex signal environments by learning latent patterns in large amounts of data. However, despite the significant progress made by deep learning models in signal classification tasks, they still face challenges when dealing with labeled data, especially when the training data contains mislabeled or noisy labels, where the model's performance and robustness often degrade significantly. This indicates the need for further technological innovation to address these emerging problems.

[0005] Existing electromagnetic signal classification techniques face several significant drawbacks, including low robustness of the model to noise and label contamination, low classification efficiency, and insufficient ability to process multi-channel data.

[0006] First, existing methods are too sensitive to mislabeled or noisy data, which often leads to a significant performance drop in practical applications, especially in environments with inconsistent data quality.

[0007] Secondly, traditional signal processing algorithms often require complex feature engineering and long model training times, which not only consume a lot of computing resources but also prolong the development cycle.

[0008] Furthermore, many models fail to effectively integrate information from multi-channel signal data, resulting in the underutilization of important features and affecting the accuracy and efficiency of classification. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a deep learning method and system for electromagnetic signal classification and recognition, which addresses the shortcomings of the prior art and solves the technical problems of reduced recognition accuracy and insufficient model robustness caused by label contamination.

[0010] The present invention adopts the following technical solution:

[0011] A deep learning method for electromagnetic signal classification and recognition includes the following steps:

[0012] Construct electromagnetic signal datasets, each containing multi-dimensional electromagnetic signal features and corresponding label information;

[0013] Select and configure the corresponding deep learning model;

[0014] The obtained electromagnetic signal dataset is labeled with contamination, including contamination at the Y-level and M-level. Y-level contamination involves randomly selecting a specific proportion of data points and modifying their labels, while M-level contamination involves systematically modifying the labels of all or part of the dataset by applying a predefined contamination matrix.

[0015] The deep learning model was trained using the obtained electromagnetic signal dataset with pollution labels.

[0016] The uncontaminated electromagnetic signal dataset is input into the obtained deep learning model, which outputs the identified electromagnetic signals.

[0017] Preferably, the magnetic signal dataset includes the RML2016.10a dataset, the WiFi signal dataset, and the Bluetooth individual dataset.

[0018] Preferably, the deep learning model includes MCLDNN and CNN5. MCLDNN is used for multi-channel data input processing of the RML dataset, and CNN5 is used for processing WiFi signal datasets and Bluetooth individual datasets.

[0019] Preferably, the Y-level pollution specifically refers to:

[0020] Implement a label contamination function `add_symmetric_noise`, which systematically introduces symmetric noise into the label data; randomly relabel the sample labels in each category according to a specified noise ratio `noise_rate`;

[0021] Asymmetric noise is introduced through the function `add_asymmetric_noise`, and the class label is modified according to the predefined mislabeled pair (source, target);

[0022] By using `flip_labels` to flip labels, a certain proportion of data points are randomly selected and their labels are changed to any other label besides the current label, thereby increasing the diversity of the data and testing the model's robustness to noise.

[0023] Preferably, the label flipping using `flip_labels` is specifically as follows:

[0024]

[0025] in, for, for, This represents the number of signal categories.

[0026] Preferably, M-matrix level contamination specifically refers to:

[0027] The contamination matrix is ​​constructed using the `construct_M_matrix_simple` method. ;

[0028] Implement the `construct_M_matrix_block` method, targeting the total number of categories. When it can be divided equally, construct a block-diagonal contamination matrix;

[0029] Construct a contamination matrix with randomness using the `construct_M_matrix_random` method;

[0030] Deploy the `LinearContaminationModel` class, based on the total number of categories. Predefined pollution matrix and the original probability of the category and the probability of the category after pollution The parameters are initialized; the pollution weight of each category is calculated, and a random selection is made in the pollution matrix according to the weight to generate pollution labels that conform to the pollution matrix settings, thereby realizing the systematic pollution of the original data labels.

[0031] Preferably, in constructing the block-diagonal contamination matrix, for even numbers... Use size All-one matrix Construct two identical blocks; for odd numbers The size of the last block was adjusted to This ensures that all categories are properly considered.

[0032] Preferably, constructing a contamination matrix with randomness specifically involves:

[0033] Generate a random number matrix After summing the rows, normalization is performed to ensure that the sum of each row is 1; The identity matrix of times and A random matrix of times Linear combinations form the final random contamination matrix. Random contamination matrix Able to use probability Randomly pollute category labels onto any other category.

[0034] Preferably, the model robustness is enhanced at a specific contamination rate; the Huber loss function is introduced to optimize model training; the Log-Cosh loss function is implemented; and the deep learning model is optimized.

[0035] Secondly, embodiments of the present invention provide a deep learning system for electromagnetic signal classification and recognition, characterized in that it includes:

[0036] The data module constructs electromagnetic signal datasets, with each dataset containing multi-dimensional electromagnetic signal features and corresponding label information;

[0037] Build the module, select and configure the corresponding deep learning model;

[0038] The contamination module performs label contamination on the electromagnetic signal dataset, including Y-level individual contamination and M-matrix level contamination. Y-level individual contamination involves randomly selecting a specific proportion of data points and modifying their labels, while M-matrix level contamination involves systematically modifying the labels of all or part of the dataset by applying a predefined contamination matrix.

[0039] The training module uses a deep learning model trained on a dataset of electromagnetic signals with pollution labels.

[0040] The recognition module inputs the uncontaminated electromagnetic signal dataset into the obtained deep learning model and outputs the recognized electromagnetic signals.

[0041] Thirdly, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the deep learning method for electromagnetic signal classification and recognition described above.

[0042] Fourthly, embodiments of the present invention provide a computer-readable storage medium including a computer program, which, when executed by a processor, implements the steps of the deep learning method for electromagnetic signal classification and recognition described above.

[0043] Compared with the prior art, the present invention has at least the following beneficial effects:

[0044] A deep learning method for electromagnetic signal classification and recognition introduces individual-level and matrix-level label contamination simulation methods. Individual-level contamination simulates random errors that may occur during data labeling by randomly modifying the labels of individual samples. Matrix-level contamination systematically modifies labels through a predefined contamination matrix, suitable for simulating errors caused by systematic biases in equipment or algorithmic recognition tendencies. This multi-level contamination simulation method effectively improves the robustness and accuracy of the model in complex real-world contamination environments. Secondly, a dynamic learning strategy and adaptive loss function are employed, which automatically adjust according to the degree of data contamination to optimize the model training process and improve model performance. By dynamically adjusting learning parameters (such as the learning rate) and selecting appropriate loss functions (such as Huber loss or Log-Cosh loss), this invention can effectively cope with different levels of data contamination, ensuring that the model achieves optimal classification results in various environments.

[0045] Furthermore, RML2016.10a is one of the standard datasets for radio spectrum modulation identification, covering 11 different modulation schemes, such as AM, FM, QAM, and PSK. It contains radio signal samples collected under different signal-to-noise ratio (SNR) conditions, making it particularly suitable for signal modulation classification and identification tasks. The WiFi signal dataset (NTU-HAR dataset) is specifically designed for human activity recognition (HAR) research, utilizing channel state information (CSI) for wireless sensing. This dataset contains 3 × 114 × 500 CSI features, covering six different types of human activities (such as boxing, falling, walking, etc.). These data represent human movement characteristics through reflected signals transmitted via wireless channels. The Bluetooth individual dataset (visualized below) is primarily a dataset collected from Bluetooth signals. This dataset records signal samples from different Bluetooth devices at a sampling rate of 250 Msps (megasamples per second). Each sample contains signal features from 150 Bluetooth devices, suitable for tasks such as device identification and authentication in Bluetooth communication.

[0046] Furthermore, MCLDNN and CNN5 were chosen to test the effectiveness of the three label contamination strategies on different signal datasets. MCLDNN was used to process the complex RML2016.10a dataset because it incorporates spatiotemporal feature extraction capabilities, while CNN5 was used for the simpler WiFi signal dataset and Bluetooth individual dataset due to its lightweight structure and ability to handle local signal features. Using these two models allows for a more comprehensive evaluation of the impact of the label contamination strategies under different signal environments, providing a basis for further optimization of label processing in signal processing.

[0047] Furthermore, the purpose of the Y-level contamination is to simulate individual labeling errors caused by isolated labeling mistakes, acquisition noise, or equipment interference in real-world scenarios by randomly selecting certain samples in the dataset and modifying their labels. This approach helps the model improve its robustness to small-scale or localized noise when facing imperfect data in the real world. By introducing individual-level label noise, the model can better learn to be immune to noise, ensuring high classification accuracy even under imperfect labeling conditions and adapting to diverse signal data characteristics.

[0048] Furthermore, the purpose of M-matrix-level contamination is to simulate widespread label noise caused by systematic errors, global interference, or data transmission problems by systematically and massively modifying the labels across the entire dataset. This method helps the model to have a stronger ability to adapt to more complex global noise. By introducing matrix-level contamination into multiple categories or regions in the dataset, the model can learn to effectively identify signal features in the presence of widespread noise interference, avoiding a significant drop in model performance due to widespread noise effects, thereby improving the model's generalization ability and robustness.

[0049] Furthermore, by performing robustness enhancements at specific contamination rates, the label contamination problem that may occur in real-world scenarios can be effectively simulated. The Huber loss function combines the advantages of mean squared error and absolute error, providing more stable gradients when dealing with outliers, thereby reducing the impact of noise interference on the model. The Log-Cosh loss function further improves the model's performance when facing large-scale errors, maintaining sufficient sensitivity while avoiding gradient explosion caused by outliers. Through these robustness enhancements, even at high contamination rates, the model can still maintain high classification accuracy, ensuring more robust handling of contaminated data in practical applications. This approach provides the model with stronger fault tolerance and generalization ability in complex real-world data environments.

[0050] It is understandable that the beneficial effects of the second aspect mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0051] In summary, this invention, by introducing label contamination mechanisms at the Y-level (individual) and M-level (matrix) levels, can effectively simulate label inaccuracies that may occur in real-world environments. By selecting appropriate deep learning models and optimization strategies, this method can significantly improve the robustness and generalization ability of electromagnetic signal classification. When dealing with complex, multi-dimensional electromagnetic signal features, the model exhibits strong adaptability, ensuring high classification accuracy under different contamination conditions, thereby enhancing the stability and practicality of the system.

[0052] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the following description of the relative embodiments will be briefly introduced. 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.

[0054] Figure 1 This is a schematic diagram illustrating the process of generating contamination tags for the electromagnetic signal classification model based on a multi-level tag contamination strategy according to the present invention.

[0055] Figure 2 A schematic diagram of the electromagnetic signal processing and model robustness optimization method based on deep learning provided in an embodiment of the present invention;

[0056] Figure 3 A schematic diagram of an electromagnetic signal processing flow based on a multi-channel deep neural network (MCLDNN) provided for an embodiment of the present invention;

[0057] Figure 4 This is a simple M-matrix multigradient contamination result;

[0058] Figure 5 The result is a multi-gradient contamination of a random M-matrix;

[0059] Figure 6 A comparison of the fitting results of the two methods;

[0060] Figure 7 For comparison of results from various contamination methods on the WIFI dataset;

[0061] Figure 8 To compare the results of various contamination methods on the Bluetooth brand individual dataset;

[0062] Figure 9 A diagram illustrating the introduction of three label contamination strategies on three datasets;

[0063] Figure 10 A schematic diagram of a computer device provided in an embodiment of the present invention;

[0064] Figure 11 This is a block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0067] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0068] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0069] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0070] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0071] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0072] This invention provides a deep learning method for electromagnetic signal classification and recognition. By simulating label contamination at both the individual and matrix levels, it effectively reproduces real-world contamination scenarios. Individual-level contamination simulates sporadic data labeling errors by randomly modifying the labels of individual samples, while matrix-level contamination simulates systematic errors caused by device system bias or algorithmic bias. Furthermore, this invention dynamically adjusts the loss function and training strategy to adapt to different levels of contamination, thereby optimizing the model's performance under various environments. Through these innovative measures, this invention not only improves the accuracy of electromagnetic signal classification but also significantly enhances the model's robustness under complex contamination conditions.

[0073] Please see Figure 1 and Figure 2 This invention discloses a deep learning method for electromagnetic signal classification and recognition, comprising the following steps:

[0074] S1. Provides various electromagnetic signal datasets, including but not limited to the RML2016.10a dataset, WiFi signal dataset, and Bluetooth individual dataset, where each dataset contains multi-dimensional electromagnetic signal features and corresponding label information;

[0075] S2. Select and configure a suitable deep learning model for electromagnetic signal processing. The model includes, but is not limited to, MCLDNN and CNN5. MCLDNN is used for multi-channel data input processing of RML datasets, and CNN5 is used for processing WiFi signal datasets and Bluetooth individual datasets.

[0076] Please see Figure 3MCLDNN (Multi-Channel Learning Deep Neural Network) is a deep neural network that learns across multiple channels, primarily used for processing and classifying complex signals. MCLDNN can simultaneously receive data input from multiple channels and extract features from different scales through its hierarchical deep structure. Its design goal is to effectively capture cross-channel correlation and discriminative information in signal processing tasks, making it particularly suitable for processing data with multi-dimensional characteristics, such as electromagnetic signals.

[0077] S3. Label contamination of the dataset, including individual-level contamination and matrix-level contamination. Individual-level contamination involves randomly selecting a specific proportion of data points and modifying their labels. Matrix-level contamination involves systematically modifying the labels of all or part of the dataset by applying a predefined contamination matrix.

[0078] Individual-level pollution specifically refers to:

[0079] S301. Implement a label contamination function `add_symmetric_noise`, which systematically introduces symmetric noise into the label data; simulate mislabeling by randomly relabeling sample labels in each category according to a specified noise ratio `noise_rate`, excluding the original category; this process is represented as conditional probability.

[0080] The conditional probability is specifically:

[0081]

[0082] in, and This represents the total number of categories.

[0083] S302. Asymmetric noise is introduced through the function `add_asymmetric_noise`, which specifically modifies the class label based on a predefined mislabeled pair (source, target).

[0084] This operation reflects prior knowledge based on false alarms, such as QAM16 being mislabeled as QAM64, reflecting potential biases in the label generation process. For each pair (s, t) in `noise_pairs`, the probability of mislabeling is defined as... This allows for precise control over the dynamics of label contamination during model training.

[0085] S303. Use `flip_labels` to flip the labels. This step randomly selects a certain proportion of data points and changes their labels to any other labels besides the current labels, thereby increasing the diversity of the data and testing the model's robustness to noise.

[0086] This operation is defined as:

[0087]

[0088] when Further simulations were conducted to detect potential random disturbances in the data.

[0089] The M-matrix level specifically refers to:

[0090] S304. Construct a simple M matrix;

[0091] The contamination matrix is ​​constructed using the `construct_M_matrix_simple(K, epsilon)` method. This matrix, through linear combination, A multiple of the identity matrix and A matrix of all ones multiplied by 1 Add them together to create a pollution rate. A symmetric contamination matrix is ​​used to ensure that the contamination probability of each category is evenly distributed across all other categories.

[0092] S305, Construct the block M matrix;

[0093] Implement the `construct_M_matrix_block(K, epsilon)` method, targeting the total number of categories. When the data can be divided equally, a block-based diagonal contamination matrix is ​​constructed; this matrix employs different construction strategies based on the parity of the number of categories:

[0094] For even numbers Use size All-one matrix Construct two identical blocks;

[0095] For odd numbers The size of the last block was adjusted to This ensures that all categories are properly considered.

[0096] This method enhances the pollution correlation within categories, reflecting the possible inherent relationships between categories in practical applications.

[0097] S306. Construct a random M matrix;

[0098] The `construct_M_matrix_random(K, epsilon, random_state)` method constructs a pollution matrix with randomness, as follows:

[0099] First generate one random number matrix After summing the rows, normalization is performed to ensure that the sum of the elements in each row is 1;

[0100] Then, The identity matrix of times and A random matrix of times Linear combinations form the final random contamination matrix. This matrix can be used with probability Randomly pollute category labels onto any other category.

[0101] S307. Apply a linear pollution model.

[0102] Deployment: The `LinearContaminationModel` class, which calculates the total number of categories. Predefined pollution matrix and the original probability of the category and the probability of the category after pollution The parameters are initialized; the model calculates the contamination weight of each category and randomly selects from the contamination matrix according to the weight to generate contamination labels that conform to the contamination matrix settings, thereby achieving systematic contamination of the original data labels.

[0103] S4. Train the deep learning model using a dataset with contamination labels. The training process includes weight initialization, forward propagation, loss calculation, backpropagation, and parameter update.

[0104] S401. Initialize the deep learning model and select a network structure suitable for electromagnetic signal classification, such as MCLDNN or CNN5.

[0105] Configuration includes setting an appropriate learning rate, batch size, and number of training iterations.

[0106] The model is trained normally using an uncontaminated electromagnetic signal dataset to ensure that it achieves optimal performance under baseline conditions.

[0107] This step provides a performance benchmark for subsequent contamination experiments.

[0108] S402, Automated gradient pollution implementation.

[0109] Design and implement an automated script that can automatically adjust the contamination rate according to a preset gradient (from 5% to 90%, increasing by 5% each time) and progressively apply it to the labeled data. The script supports two contamination levels (individual level and M-matrix level) and three specific contamination methods for each level.

[0110] S5. Evaluate the performance of the model on an uncontaminated test dataset, including but not limited to calculating the model's accuracy, recall, precision, and F1 score, and comparing the changes in model performance before and after contamination.

[0111] The script described above is used to generate datasets with different contamination rates. For each contamination level and method, the model is retrained, and the performance changes on the contaminated data are recorded and analyzed. Key performance indicators, such as accuracy, recall, and F1 score, are collected for each experiment. The impact of different contamination rates, levels, and methods on model performance is analyzed to quantitatively assess the specific impact of contamination on recognition accuracy. Based on the collected data, a graph showing the relationship between contamination rate and model performance is plotted. These graphs visually demonstrate the impact of different contamination strategies on model performance, revealing the potential threats to model performance posed by data quality issues that may be encountered in practical applications.

[0112] S6. Conduct a systematic evaluation of different pollution methods at multiple levels, and carry out pollution experiments with a 5% gradient for a total of 18 times; through repeated experiments, accurately depict the quantitative relationship between the pollution ratio and the accuracy of electromagnetic signal classification.

[0113] S7. At a specific contamination rate, such as 10%, perform robustness enhancement on the model; replace the traditional loss function with a loss function such as Huber to enhance the model's tolerance to abnormal labels; at the same time, statistically analyze the cases of frequent misclassification as prior knowledge and adjust the network to focus on more discriminative features, thereby improving the overall robustness and classification accuracy of the model.

[0114] S701. Based on the experimental results, evaluate and adjust the model parameters or introduce new algorithm strategies, such as introducing new loss functions or data augmentation techniques, to improve the robustness and reliability of the model in highly polluted environments.

[0115] S702. Introduce the Huber loss function to optimize model training;

[0116] Huber loss provides a scheme for a smooth transition between squared loss and absolute loss, through a threshold. Controlling the sensitivity of the loss function. For small errors, the function uses squared loss; for large errors, it switches to linear loss, thereby reducing excessive penalty for outlier labels. The specific expression is:

[0117] in, It is the actual value. These are predicted values, and this method is particularly suitable for handling contaminated data, reducing the model's sensitivity to outliers.

[0118] S703, Implement the Log-Cosh loss function;

[0119] This loss function provides different penalty mechanisms for small and large errors. The Log-Cosh loss function exhibits continuous and asymptotic characteristics and is less sensitive to large errors, thereby improving the robustness of the model. The expression is as follows:

[0120]

[0121] in, It is the actual value. It is a predicted value. It is a hyperbolic cosine function, which is an even function.

[0122] This loss function combines the characteristics of mean squared error and the advantages of mean absolute error, effectively mitigating the impact of large errors and making it suitable for use in contaminated label environments. The function's shape is similar to the squared error loss, but the penalty for large errors increases more slowly, thus avoiding excessive penalties when dealing with extreme values ​​or outliers.

[0123] S704. Based on the introduction of the above loss function, conduct a comprehensive evaluation of the model performance.

[0124] Adjust hyperparameters used in model training, such as learning rate and batch size, to adapt to the new loss function. Monitor the model's performance at different levels of contamination and further optimize the algorithm accordingly to ensure that the model maintains high recognition accuracy and robustness when faced with contaminated data.

[0125] Those skilled in the art will understand that various aspects of the present invention can be implemented as systems, methods, or program products. Therefore, various aspects of the present invention can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "platform."

[0126] In another embodiment of the present invention, a deep learning system for electromagnetic signal classification and recognition is provided. This system can be used to implement the above-mentioned deep learning method for electromagnetic signal classification and recognition. Specifically, the deep learning system for electromagnetic signal classification and recognition includes a data module, a construction module, a contamination module, a training module, and a recognition module.

[0127] The data module constructs electromagnetic signal datasets, each containing multi-dimensional electromagnetic signal features and corresponding label information.

[0128] Build the module, select and configure the corresponding deep learning model;

[0129] The contamination module performs label contamination on the electromagnetic signal dataset, including Y-level individual contamination and M-matrix level contamination. Y-level individual contamination involves randomly selecting a specific proportion of data points and modifying their labels, while M-matrix level contamination involves systematically modifying the labels of all or part of the dataset by applying a predefined contamination matrix.

[0130] The training module uses a deep learning model trained on a dataset of electromagnetic signals with pollution labels.

[0131] The recognition module inputs the uncontaminated electromagnetic signal dataset into the obtained deep learning model and outputs the recognized electromagnetic signals.

[0132] In another embodiment of the present invention, a terminal device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of deep learning methods for electromagnetic signal classification and recognition, including:

[0133] Construct an electromagnetic signal dataset, with each dataset containing multi-dimensional electromagnetic signal features and corresponding label information; select and configure the corresponding deep learning model; label contamination is applied to the obtained electromagnetic signal dataset, including Y-level individual contamination and M-matrix level contamination. Y-level contamination involves randomly selecting a specific proportion of data points and modifying their labels, while M-matrix level contamination involves systematically modifying the labels of all or part of the dataset by applying a predefined contamination matrix; train the deep learning model using the obtained electromagnetic signal dataset with contaminated labels; input the uncontaminated electromagnetic signal dataset into the obtained deep learning model, and output the identified electromagnetic signals.

[0134] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device; it can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that more specific examples (a non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0135] Computer-readable storage media also include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium can also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0136] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0137] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the deep learning method for electromagnetic signal classification and recognition in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps:

[0138] Construct an electromagnetic signal dataset, with each dataset containing multi-dimensional electromagnetic signal features and corresponding label information; select and configure the corresponding deep learning model; label contamination is applied to the obtained electromagnetic signal dataset, including Y-level individual contamination and M-matrix level contamination. Y-level contamination involves randomly selecting a specific proportion of data points and modifying their labels, while M-matrix level contamination involves systematically modifying the labels of all or part of the dataset by applying a predefined contamination matrix; train the deep learning model using the obtained electromagnetic signal dataset with contaminated labels; input the uncontaminated electromagnetic signal dataset into the obtained deep learning model, and output the identified electromagnetic signals.

[0139] Please see Figure 10 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the deep learning method for electromagnetic signal classification and recognition described in this embodiment. To avoid repetition, these details are not elaborated here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the deep learning system for electromagnetic signal classification and recognition described in this embodiment. To avoid repetition, these details are not elaborated here.

[0140] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 10 This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.

[0141] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, CPUs, graphics processing units (GPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, quantum computing-based data processing logic units, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0142] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or RAM of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.

[0143] Furthermore, the memory 62 may include both internal storage units of the computer device 60 and external storage devices. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0144] Any references to memory, databases, or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0145] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0146] Please see Figure 11 The terminal device 600 is an electronic device, which takes the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including storage unit 620 and processing unit 610), a display unit 640, etc.

[0147] The storage unit stores program code, which can be executed by the processing unit 610 to perform the steps described in the method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 610 can perform actions such as... Figure 1 The steps are shown in the figure.

[0148] Storage unit 620 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 6201 and / or cache memory 6202, and may further include a read-only memory (ROM) 6203.

[0149] Storage unit 620 may also include a program / utility 6204 having a set (at least one) program module 6205, such program module 6205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0150] Bus 630 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0151] Electronic device 600 can also communicate with one or more external devices 700 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 650. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 660. Network adapter 660 can communicate with other modules of electronic device 600 via bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0152] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0153] Please see Figure 4 This figure shows that, in 18 repeated experiments, the contamination rate gradually increased in 5% increments, ranging from 5% to 90%. The horizontal axis represents the signal-to-noise ratio (SNR), ranging from -20 to 18, and the vertical axis represents the model's accuracy. These experiments simulated the impact of different label contamination ratios on model performance using a simple M-matrix method.

[0154] Please see Figure 5 This figure shows that, in 18 repeated experiments, the contamination rate gradually increased in 5% increments, ranging from 5% to 90%. The horizontal axis represents the signal-to-noise ratio (SNR), ranging from -20 to 18, and the vertical axis represents the model's accuracy. These experiments simulated the impact of different label contamination ratios on model performance using a stochastic M-matrix method.

[0155] Please see Figure 6 This figure illustrates the relationship between label contamination rate and model accuracy. The experiment was conducted 18 times, with the contamination rate increasing in increments of 5%, covering a range from 5% to 90%. The figure includes two sets of data: experimental results using the simple M-matrix method and the randomized M-matrix method. The dataset used is the RML2016.10a dataset.

[0156] Please see Figure 7 This figure illustrates the relationship between label contamination rate and model accuracy. The experiment was conducted 18 times, with the contamination rate increasing in increments of 5%, covering a range from 5% to 90%. The figure includes two sets of data, corresponding to the experimental results under the simple M-matrix method and the randomized M-matrix method, respectively. The dataset used is a WiFi signal dataset.

[0157] Please see Figure 8 This figure illustrates the relationship between tag contamination rate and model accuracy. The experiment was conducted 18 times, with the contamination rate increasing in increments of 5%, covering a range from 5% to 90%. The figure includes two sets of data, corresponding to experimental results under simple M-matrix, random M-matrix, and symmetric Y-individual methods, as well as asymmetric Y-individual methods and random flipping methods. The dataset used is the Bluetooth individual dataset.

[0158] Please see Figure 9By introducing three label contamination strategies on these three datasets, the model achieves broader signal coverage, exhibiting stronger robustness and generalization ability, particularly when dealing with diverse signal patterns and complex signal environments. Each dataset corresponds to a different communication scenario, including long-range radio modulation identification, short-range WiFi environment awareness, and individual signal characteristics of Bluetooth devices. Based on these datasets, the model can not only handle common wireless communication signals but also accurately identify and classify signals in more complex and diverse environments through different model architectures. This integrated approach using multiple signal sources and different label contamination strategies enhances the model's adaptability and anti-interference capabilities when dealing with complex signal problems.

[0159] In summary, this invention provides a deep learning method and system for electromagnetic signal classification and recognition, aiming to improve the robustness and efficiency of electromagnetic signal classification. By simulating different levels of label contamination, this invention can test and enhance the model's performance under various contamination conditions, thereby improving its reliability in practical applications. Furthermore, this invention employs advanced deep learning technology, optimizing the training process by dynamically adjusting the learning strategy and adaptive loss function, reducing the demand for computing resources and shortening the model training time. This not only enhances the model's ability to process multi-channel data but also significantly improves the overall classification efficiency.

[0160] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0161] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0162] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0163] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0164] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0165] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0166] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0167] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0168] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0169] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0170] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A deep learning method for electromagnetic signal classification, characterized in that, Comprising the following steps: S1, constructing an electromagnetic signal dataset, each dataset containing multi-dimensional electromagnetic signal features and corresponding label information; S2, selecting and configuring a corresponding deep learning model; S3, performing label pollution on the electromagnetic signal dataset obtained in step S1, including Y individual level and M matrix level pollution, Y individual level pollution by randomly selecting a specific proportion of data points and modifying their labels, M matrix level pollution by applying a predefined pollution matrix to systematically modify the labels of all or part of the dataset, Y individual level pollution specifically: Implement a label pollution function `add_symmetric_noise`, which systematically introduces symmetric noise into the label data; by randomly re-labeling the sample labels in each class according to the specified noise rate `noise_rate`; Introduce asymmetric noise through the function `add_asymmetric_noise`, modify the class labels according to the pre-defined mislabeling (source, target); Use `flip_labels` to flip labels, randomly select a certain proportion of data points and change their labels to any other label except the current label, to increase the diversity of the data and test the robustness of the model to noise; the label flipping using `flip_labels` is specifically: wherein is, is, is the number of signal classes; M matrix level pollution specifically: Construct the pollution matrix using the `construct_M_matrix_simple` method ; Implement the `construct_M_matrix_block` method for the total number of categories When it can be evenly divided, construct a block-diagonal contamination matrix; Construct a pollution matrix with randomness through the `construct_M_matrix_random` method; The `LinearContaminationModel` class is deployed to initialize the total number of categories , a predefined contamination matrix , and the original probability of the category and the contaminated category probability ; by calculating the contamination weight of each category and randomly selecting in the contamination matrix according to the weight, the original data label is systematically contaminated to generate the contamination label set by the contamination matrix; in the block diagonal contamination matrix, for even , two identical blocks are constructed using a full one matrix of size ; for odd , the last block size is adjusted to , ensuring that all categories are properly considered, and a random contamination matrix is constructed, which is specific to: Generate a random number matrix After summing the rows, normalization is performed to ensure that the sum of elements in each row is 1; The identity matrix multiplied by 1 A random matrix of times Linear combinations form the final random contamination matrix. Random contamination matrix Able to use probability Randomly contaminate category labels to any other category; S4, training the deep learning model obtained in step S2 using the electromagnetic signal dataset with polluted labels obtained in step S3; S5, inputting the un-polluted electromagnetic signal dataset into the deep learning model obtained in step S4, and outputting the recognized electromagnetic signal.

2. The deep learning method for electromagnetic signal classification according to claim 1, wherein, The magnetic signal dataset includes the RML2016.10a dataset, the WiFi signal dataset, and the Bluetooth individual dataset.

3. The deep learning method for electromagnetic signal classification according to claim 1, wherein, The deep learning model includes MCLDNN and CNN5, MCLDNN is used for multi-channel data input processing RML dataset, CNN5 is used for processing WiFi signal dataset and Bluetooth individual dataset.

4. The deep learning method for electromagnetic signal classification according to claim 1, wherein, Robustness enhancement processing of the model at a specific pollution rate; introduce Huber loss function to optimize model training; implement Log-Cosh loss function; optimize the deep learning model.

5. A deep learning system for electromagnetic signal classification, characterized in that, Comprising: Data module, constructing an electromagnetic signal dataset, each dataset containing multi-dimensional electromagnetic signal features and corresponding label information; Construction module, selecting and configuring a corresponding deep learning model; Pollution module, performing label pollution on the electromagnetic signal dataset, including Y individual level and M matrix level pollution, Y individual level pollution by randomly selecting a specific proportion of data points and modifying their labels, M matrix level pollution by applying a predefined pollution matrix to systematically modify the labels of all or part of the dataset, Y individual level pollution specifically: Implement a label pollution function `add_symmetric_noise` that systematically introduces symmetric noise into the label data; by randomly relabeling the labels of samples in each class according to the specified noise rate `noise_rate`; Introduce asymmetric noise through the function `add_asymmetric_noise`, modify the class labels according to the pre-defined mislabeling (source, target); Use `flip_labels` to flip labels, randomly select a certain proportion of data points and change their labels to any other label except the current label, in order to increase the diversity of data and test the robustness of the model to noise; Use `flip_labels` to flip labels specifically: wherein is, is, is the number of signal classes; M matrix level pollution is specifically: Construct the pollution matrix using the `construct_M_matrix_simple` method ; Implement the `construct_M_matrix_block` method for the total number of categories When it can be evenly divided, construct a block-diagonal contamination matrix; Construct a pollution matrix with randomness through the `construct_M_matrix_random` method; The `LinearContaminationModel` class is deployed to initialize the total number of categories , a predefined contamination matrix , and the original probability of categories and the contaminated category probability as parameters; by calculating the contamination weight of each category and randomly selecting in the contamination matrix according to the weight, the contaminated label set by the contamination matrix is generated to realize the systematic contamination of the original data label, and a block diagonal contamination matrix is constructed. For even , two identical blocks are constructed using a full one matrix of size ; for odd , the last block size is adjusted to , ensuring that all categories are properly considered, and a contamination matrix with randomness is constructed, which is as follows: Generate a matrix of random numbers , normalize by row sum to ensure each row sums to 1; linearly combine times the identity matrix with times the random matrix to form the final random corruption matrix , the random corruption matrix can randomly corrupt class labels to any other class with probability ; Training module, use the deep learning model trained with the electromagnetic signal dataset with polluted labels; Recognition module, input the deep learning model obtained by the electromagnetic signal dataset without pollution, and output the recognized electromagnetic signal.

Images