Electromagnetic signal processing method and device, computer equipment and storage medium
By filtering and feature extraction of electromagnetic signals, combined with self-attention processing and aggregation technology, the problem of poor feature extraction accuracy in electromagnetic signal processing is solved, and more efficient and accurate signal recognition is achieved.
Patent Information
- Application Number
- CN202510016009.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-06
- Publication Date
- 2025-05-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has poor feature extraction accuracy in electromagnetic signal processing, making it difficult to accurately identify the modulation type of signal in complex electromagnetic environments.
By obtaining the target electromagnetic signal for filtering, the signal characteristics of the spectrum map are extracted and mapped to the image space to generate the target image. The image is then divided into sub-feature maps, self-attention processing and aggregation are performed to obtain aggregated features for identifying electromagnetic signals.
It improves the accuracy of signal feature extraction, enhances the ability to identify electromagnetic signals, reduces information redundancy, and improves processing speed and efficiency.
Smart Images
Figure CN119939213A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of signal processing, and in particular relates to a method, device, computer equipment and storage medium for processing electromagnetic signals. Background Art
[0002] Electromagnetic signals are currently the most important type of signal in the field of communications, and their analysis and processing are extremely important. Deep learning, as a new research direction in machine learning, builds an artificial neural network (ANN) and uses multi-level analysis and calculation to extract more features. It is widely used in fields such as face recognition, pedestrian detection, and image analysis. It is also used in the analysis and processing of electromagnetic signals.
[0003] First of all, electromagnetic signals have high requirements for the linearity of the transmission channel. The harmonic signal generated by the nonlinearity of the device in the transmission channel will be coupled to the receiving channel. Usually, this signal is much larger than the harmonic radiation generated by the target. Since the target re-radiated harmonic signal is weak, the design of the receiver requires high sensitivity and large processing gain. In addition, in some specific scenarios, such as damage to the shielding facilities, there will be complex electromagnetic interference in the scanning shielding facilities, which will cause certain interference to the image generated by the electromagnetic signal, causing image artifacts.
[0004] In the prior art, in order to solve the above problems, electromagnetic signals are analyzed and processed based on deep learning, including denoising, identification and classification of signals through neural networks. For example, automatic modulation recognition (AMR) is a major technical direction in electromagnetic signal analysis. It is a technology that can distinguish the modulation method used by the modulated signal and estimate the modulation-related parameters without relevant prior information.
[0005] However, in the existing technology, AMR's processing of electromagnetic signals is inefficient. For example, processing through machine learning algorithms such as support vector machines or K nearest neighbors requires large amounts of computation and high complexity, making parameter adjustment difficult. Secondly, it has poor accuracy. Support vector machines are sensitive to parameter selection. When the signal feature dimension is too high, classification performance may be reduced. K nearest neighbors are sensitive to noise and outliers, which may lead to classification errors. The unique environmental complexity of the electromagnetic environment makes it difficult for existing AMR-related technologies to accurately identify the modulation type of the signal, so it is very easy to encounter the problem of insufficient feature extraction.
[0006] Therefore, the existing technology for processing electromagnetic signals has the problem of poor feature extraction accuracy. Summary of the invention
[0007] In order to solve the problem of poor accuracy in feature extraction in the prior art for electromagnetic signal processing, the present invention provides an electromagnetic signal processing method, apparatus, computer equipment and storage medium.
[0008] In order to achieve the above object, the present invention provides the following technical solutions:
[0009] First, a method for processing an electromagnetic signal is provided, the method comprising:
[0010] Acquire a target electromagnetic signal, filter the target electromagnetic signal, and obtain a filtered spectrum diagram;
[0011] Extracting signal features of the spectrum graph in frequency domain characteristics and structure, and mapping the signal features to image space to generate a target image;
[0012] Dividing the target image into a plurality of sub-feature maps, performing interval selection on the sub-feature maps to obtain a sub-feature map set;
[0013] Self-attention processing is performed on the sub-feature graphs in the sub-feature graph set to determine the correlation between different sub-feature graphs, and then the different sub-feature graphs are aggregated according to the correlation to obtain aggregated features for identifying target electromagnetic signals.
[0014] Optionally, filtering the target electromagnetic signal to obtain a filtered spectrum diagram includes:
[0015] The convolutional neural network CNN and the recurrent neural network RNN are combined to obtain the convolutional recurrent neural network CRNN; the CRNN is trained using pre-acquired training samples to obtain an electromagnetic filtering model;
[0016] The target electromagnetic signal is processed by Fourier transform, the target electromagnetic signal based on the time domain is converted into a target electromagnetic signal based on the frequency domain, and a frequency spectrum of the target electromagnetic signal is obtained;
[0017] The spectrum graph is input into the electromagnetic filtering model, the signal features are extracted by the convolutional layer CNN in the CRNN, and then these features are processed by the recurrent layer RNN to predict the filtered spectrum graph.
[0018] Optionally, extracting signal features of the spectrum graph in frequency domain characteristics and structure, and mapping the signal features to an image space to generate a target image includes:
[0019] Extracting key signal features of the spectrum in frequency domain characteristics and structure through principal component analysis;
[0020] Converting the key signal features into feature vectors, and performing dimensionality reduction processing on the feature vectors to map the high-dimensional feature vectors into two-dimensional data;
[0021] A target image is generated according to the two-dimensional data.
[0022] Optionally, the target image is divided into a plurality of sub-feature maps, and the sub-feature maps are selected at intervals to obtain a sub-feature map set including:
[0023] Preprocessing the target image to obtain a preprocessed image of a standard size;
[0024] Cutting the target image based on a preset grid size to obtain a plurality of sub-feature maps;
[0025] A sub-feature atlas set is obtained by selecting from multiple sub-feature atlases according to a preset interval, and the preset interval is greater than or equal to a sub-feature atlas set.
[0026] Optionally, performing self-attention processing on the sub-feature graphs in the sub-feature graph set to determine the correlation between different sub-feature graphs includes:
[0027] For each sub-feature graph in the sub-feature graph set, calculate its query feature graph, key feature graph and value feature graph, and determine the correlation score between the query feature graph and the key feature graph;
[0028] Applying a softmax function to convert the correlation scores into a probability distribution to obtain an autocorrelation matrix;
[0029] Generate attention weights based on the autocorrelation matrix, and use the attention weights to perform weighted summation on the value feature map to obtain the attention output of each sub-feature map;
[0030] The attention output is subjected to maximum pooling to determine the correlation between different sub-feature maps.
[0031] Optionally, aggregating different sub-feature graphs according to the correlation to obtain aggregate features for identifying the target electromagnetic signal includes:
[0032] Initialize an empty aggregate feature vector according to the number and dimension of the sub-feature maps;
[0033] Traverse the sub-feature graphs one by one, and for each sub-feature graph, aggregate it according to its correlation with other sub-feature graphs;
[0034] The aggregated sub-feature map features are integrated into an aggregated feature vector, and the aggregated feature vector is normalized to obtain an aggregated feature of the target image, and the aggregated feature is used to perform signal recognition on the target electromagnetic signal.
[0035] Secondly, a device for processing electromagnetic signals is provided, the device comprising:
[0036] An acquisition module, used for acquiring a target electromagnetic signal, filtering the target electromagnetic signal, and obtaining a filtered spectrum diagram;
[0037] A generation module, used for extracting the signal features of the spectrum graph in frequency domain characteristics and structure, and mapping the signal features to the image space to generate a target image;
[0038] A selection module, used for dividing the target image into a plurality of sub-feature maps, performing interval selection on the sub-feature maps, and obtaining a sub-feature map set;
[0039] A processing module is used to perform self-attention processing on the sub-feature graphs in the sub-feature graph set, determine the correlation between different sub-feature graphs, and then aggregate the different sub-feature graphs according to the correlation to obtain aggregated features for identifying target electromagnetic signals.
[0040] In addition, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned electromagnetic signal processing method is implemented.
[0041] Finally, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned electromagnetic signal processing method when executing the program.
[0042] The electromagnetic signal processing method provided by the present invention has the following beneficial effects:
[0043] First, the target electromagnetic signal is filtered through a neural network to preliminarily eliminate the noise components in the target electromagnetic signal, enhance the features related to the target signal in the spectrum diagram, and make the target signal more prominent. Secondly, the target image is segmented, which helps to distinguish different target areas for subsequent feature extraction and image analysis, and the segmented areas have consistent features and attributes, which is conducive to improving the stability and accuracy of subsequent feature extraction and classification processes; in addition, the sub-feature atlas contains multiple types of features that can reflect different aspects of the target image, and by selecting the sub-feature atlas, the amount of calculation in subsequent processing can be reduced, which helps to improve the processing speed and efficiency of the entire system; finally, self-attention processing is performed to obtain aggregated features, which is conducive to capturing global information of the target image, deeply mining feature correlations, reducing information redundancy, and helping to more accurately understand the content of the target image, thereby improving the accuracy of signal feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0045] Figure 1 The present invention is a flowchart of a method for processing an electromagnetic signal according to an exemplary embodiment of the present invention.
[0046] Figure 2 A structural diagram of a convolutional recurrent neural network CRNN provided according to an exemplary embodiment of the present invention.
[0047] Figure 3 A spectrum diagram of a target electromagnetic signal provided according to an exemplary embodiment of the present invention.
[0048] Figure 4 A block diagram of an electromagnetic signal processing device provided according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION
[0049] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.
[0050] The technical solutions provided by various embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.
[0051] First, the present invention provides a method for processing electromagnetic signals, specifically, Figure 1 As shown, the following steps are included:
[0052] S101, acquiring a target electromagnetic signal, filtering the target electromagnetic signal, and obtaining a filtered spectrum diagram.
[0053] In one embodiment, a convolutional neural network CNN and a recurrent neural network RNN may be combined to obtain a convolutional recurrent neural network CRNN; the CRNN is trained using pre-acquired training samples to obtain an electromagnetic filtering model.
[0054] For example, first, a convolutional recurrent neural network CRNN is constructed. Determine the network architecture of CRNN, including the number, type, convolution kernel size, step size, padding method and other parameters of the convolutional layer (CNN); determine the type of recurrent layer (RNN) (such as LSTM, GRU, etc.), as well as the number of RNN layers, the number of hidden units and other parameters; design the output layer. According to the requirements of the filtering task, the output layer can be a transcription layer for constructing a filtered spectrum graph. Finally, initialize the parameters of CRNN to complete the construction of the convolutional recurrent neural network CRNN. Secondly, obtain a training sample, which may include an original electromagnetic signal and a corresponding filtered real electromagnetic signal; input the original electromagnetic signal into the CRNN to obtain an output predicted electromagnetic signal, and train the CRNN with the goal of minimizing the deviation between the predicted electromagnetic signal and the true electromagnetic signal to obtain an electromagnetic filtering model.
[0055] In another embodiment, the target electromagnetic signal is processed by Fourier transform, and the target electromagnetic signal based on the time domain is converted into a target electromagnetic signal based on the frequency domain to obtain a spectrum diagram of the target electromagnetic signal; the spectrum diagram is input into the electromagnetic filtering model, and the signal features are extracted by the convolutional layer CNN in the CRNN, and then these features are processed by the recurrent layer RNN to predict the filtered spectrum diagram.
[0056] For example, first you need to sample the target electromagnetic signal to obtain a series of signal values at discrete time points, and use the Fourier transform algorithm to process the sampled discrete signal. In this way, the Fourier transform converts the time domain signal into a frequency domain signal to obtain a series of frequency domain components represented by complex numbers. The modulus values of these complex numbers represent the signal amplitude at the corresponding frequency, and the phase represents the signal phase at the corresponding frequency. Then, in order to obtain a spectrum diagram, you need to calculate the amplitude corresponding to each frequency component. For example, you can use the abs function to calculate the modulus value of the complex number, and use the plot function to draw the spectrum diagram. Finally, set the frequency axis and draw the spectrum diagram, which can be shown as follows. Figure 3 shown.
[0057] S102: extracting signal features from the spectrum graph, and mapping the signal features to an image space to generate a target image.
[0058] In one embodiment, key signal features of the frequency domain characteristics and structure of the spectrum graph may be extracted through principal component analysis.
[0059] Specifically, each spectrogram can be regarded as a two-dimensional matrix, in which each element represents the signal strength at a specific frequency. The spectrogram data is organized into a matrix form, in which each row represents a spectrogram and each column represents the signal strength at a frequency point. Since the signal strength at different frequency points may vary greatly, the data needs to be standardized to ensure that each feature (i.e., the signal strength at the frequency point) contributes equally to the principal component analysis PCA. Among them, standardization usually includes calculating the mean and standard deviation of each feature, and then subtracting the mean from each feature value and dividing it by the standard deviation.
[0060] Calculate the covariance matrix of the standardized data. The covariance matrix reflects the correlation between different features. In PCA, the covariance matrix is used to calculate the eigenvectors (i.e., principal components). Perform eigendecomposition on the covariance matrix to obtain eigenvectors and eigenvalues. The eigenvector represents the direction of the principal component, and the eigenvalue represents the importance of the corresponding principal component (i.e., variance contribution). Sort the eigenvalues from large to small, and select the eigenvectors corresponding to the first few largest eigenvalues of a preset number as the principal components. Finally, project the original data into the space formed by the selected principal components to obtain the reduced-dimensional data. These reduced-dimensional data are the key signal features.
[0061] In another embodiment, the key signal feature is converted into a feature vector, and the feature vector is subjected to dimensionality reduction processing, and the high-dimensional feature vector is mapped into two-dimensional data; and a target image is generated based on the two-dimensional data.
[0062] For example, the key signal features of each spectrogram are organized into a feature vector, and then PCA or other dimensionality reduction methods, such as linear discriminant analysis (LDA) or t-distributed random neighbor embedding (t-SNE) are used to reduce its dimension. In this step, the high-dimensional feature vector needs to be mapped to two-dimensional data for visualization on a two-dimensional plane. The reduced two-dimensional data is mapped to a two-dimensional plane, and each data point corresponds to a key signal feature of the spectrogram. Then, a drawing software is used to generate a target image based on the two-dimensional data.
[0063] Through the above steps, principal component analysis can be used to extract the key signal features of the spectrum graph, and these features can be converted into feature vectors for dimensionality reduction. Finally, the target image is generated to visually display the frequency domain characteristics and structural information of the spectrum graph, which helps to better understand and analyze the frequency components of electromagnetic signals and their relationships.
[0064] S103, dividing the target image into a plurality of sub-feature maps, performing interval selection on the sub-feature maps to obtain a sub-feature map set.
[0065] Specifically, the target image is preprocessed to obtain a preprocessed image of a standard size; the target image is cut based on a preset grid size to obtain multiple sub-feature maps; and a sub-feature map set is obtained by selecting from the multiple sub-feature maps according to a preset interval, and the preset interval is greater than or equal to a sub-feature map set.
[0066] Among them, the preprocessing of the target image can be performed by scaling the target image through a scaling function according to actual needs to ensure the consistency and comparability of the processing.
[0067] For example, the original size of the target image is 4032x3024 pixels. For unified processing and analysis, the target image can be adjusted to a standard size of 1024x768 pixels. In order to extract the detailed features in the target image, the image can be cut using a 64x64 pixel grid. Then, according to the size of the target image (1024x768 pixels) and the grid size (64x64 pixels), the number of sub-feature maps that can be cut is calculated. Here, 16x12=192 sub-feature maps can be obtained. Then, the target image is cut using the image cutting function, so that 192 sub-feature maps are obtained, and the size of each sub-feature map is 64x64 pixels. Then, the sub-feature maps are selected according to the preset interval. Here, the sub-feature maps include 16 rows and 12 columns. The preset interval can be a sub-feature map, so that 8 sub-feature maps can be selected per row and 6 sub-feature maps can be selected per column.
[0068] S104, performing self-attention processing on the sub-feature graphs in the sub-feature graph set to determine the correlation between different sub-feature graphs, and then aggregating the different sub-feature graphs according to the correlation to obtain aggregated features.
[0069] Among them, to perform self-attention processing on the sub-feature graph, it is necessary to first calculate the query feature graph, key feature graph and value feature graph of each sub-feature graph in the sub-feature graph set, and determine the correlation score between the query feature graph and the key feature graph; then apply the softmax function to convert the correlation score into a probability distribution to obtain an autocorrelation matrix; then generate attention weights based on the autocorrelation matrix, and use the attention weights to perform weighted summation on the value feature graph to obtain the attention output of each sub-feature graph; finally, perform maximum pooling processing on the attention output to determine the correlation between different sub-feature graphs.
[0070] Specifically, for each sub-feature graph, a linear transformation is used to generate a query feature graph (Q), a key feature graph (K), and a value feature graph (V). The purpose of these linear transformations is to map the original feature space to a new space where Q, K, and V can better represent the relationship between features. For each sub-feature graph X, its Q, K, and V are: Q = W Q *X, K = W K*X and V = W V *X, where W Q , W K and W V is a learnable weight matrix. The dot product of the query feature map (Q) and the key feature map (K) is used to calculate the correlation score between them. For example, this can be achieved by matrix multiplication, that is, Q and K T and then apply a scaling factor (usually where d k is the dimension of the key feature map) to prevent the gradient vanishing problem caused by excessive dot product results. The correlation score matrix S can be expressed as:
[0071] The relevance score matrix S is normalized by the Softmax function to obtain the probability distribution between each query and all key features.
[0072] For example, for each sub-feature map F i , normalize the correlation scores with all sub-feature maps through the Softmax function to obtain a probability distribution (i.e., a row of the autocorrelation matrix). This probability distribution represents F i The relative importance or relevance of all sub-feature graphs. This allows each query to have a higher score for its most relevant key feature graph and a lower score for less relevant key feature graphs. The normalized relevance score matrix is It can be expressed as:
[0073] Each row of the autocorrelation matrix represents the correlation score between a sub-feature map and all other sub-feature maps (probability distribution after Softmax normalization). These scores are attention weights, which are used to guide how to weight feature maps.
[0074] For each sub-feature map F i , we can use its corresponding attention weight pair feature map {V1,V2,...,V N}Perform weighted summation.
[0075] Specifically, using the normalized correlation score matrix To weight the corresponding value feature map (V) to obtain the weighted output, this can be achieved by matrix multiplication, that is, The product of and V.
[0076] For each sub-feature map F i , the result of weighted summation is its attention output A i This output combines the information of all sub-feature maps and is weighted according to the correlation between them.
[0077] Finally, we can also output A for attention i Perform post-processing, such as maximum pooling, to retain the most significant information in the local features and determine the correlation between different sub-feature maps.
[0078] In addition, different sub-feature graphs are aggregated according to their correlation. First, an empty aggregate feature vector needs to be initialized according to the number and dimension of the sub-feature graphs. Then, the sub-feature graphs are traversed one by one, and for each sub-feature graph, it is aggregated according to its correlation with other sub-feature graphs. Finally, the aggregated sub-feature graph features are integrated into the aggregate feature vector, and the aggregate feature vector is normalized to obtain the aggregate feature of the target image, which can be used to identify the target electromagnetic signal.
[0079] For example, the number of sub-feature graphs is N, and the dimension of each sub-feature graph is D (that is, each sub-feature graph can be regarded as a D-dimensional vector). According to the number of sub-feature graphs N and the dimension D, initialize an all-zero aggregate feature vector F agg , whose dimension is D. This vector will be used to store the aggregated features of all sub-feature graphs. Use an iterator to traverse all sub-feature graphs one by one. For the sub-feature graph F currently traversed i (i from 1 to N), calculate its correlation with all other sub-feature maps (including itself). Correlation can be calculated in many ways, such as dot product, cosine similarity, Euclidean distance, etc. Here, some correlation measure R(F i ,F j ) to represent F i and F j For each sub-feature map F i , according to their correlation with other sub-feature graphs, their features are aggregated. The aggregation method can be weighted average, summation, maximum value, etc. Here, weighted average is used for aggregation, that is, the contribution of each sub-feature graph is determined according to its correlation weight with other sub-feature graphs. Specifically, for F i , we can calculate a weighted sum For each sub-feature map F i , and the aggregated feature S i Integrate into the aggregate feature vector F agg Then, for F agg Normalize to ensure that F agg Each element of is in a reasonable range (such as [0,1] or [-1,1]), and finally the aggregated features of the target image are obtained.
[0080] Finally, the obtained aggregated features are identified to obtain the signal content of the target electromagnetic signal.
[0081] For example, the aggregated features can be input into a trained recognition model, and the model will parse and recognize the aggregated features based on the knowledge and rules it has learned, and the model will output a recognition result indicating the type, attributes or characteristics of the target electromagnetic signal. The recognition model can be a neural network model based on machine learning and its variants.
[0082] By adopting the above method, the target electromagnetic signal is first filtered through a neural network to preliminarily eliminate the noise components in the target electromagnetic signal, enhance the features related to the target signal in the spectrum diagram, and make the target signal more prominent. Secondly, the target image is segmented, which is helpful to distinguish different target areas for subsequent feature extraction and image analysis, and the segmented areas have consistent features and attributes, which is conducive to improving the stability and accuracy in the subsequent feature extraction and classification process; in addition, the sub-feature atlas contains multiple types of features, which can reflect different aspects of the target image, and by selecting the sub-feature atlas, the amount of calculation in subsequent processing can be reduced, which is helpful to improve the processing speed and efficiency of the entire system; finally, self-attention processing is performed to obtain aggregated features, which is conducive to global information capture of the target image, in-depth feature correlation mining, and reducing information redundancy, which is helpful to more accurately understand the content of the target image and improve the accuracy of signal feature extraction.
[0083] Secondly, the present invention also provides an electromagnetic signal processing device, such as Figure 4 As shown, including:
[0084] An acquisition module 401 is used to acquire a target electromagnetic signal, filter the target electromagnetic signal, and obtain a filtered spectrum diagram;
[0085] A generating module 402 is used to extract the signal features of the frequency domain characteristics and structure of the spectrum graph, and map the signal features to the image space to generate a target image;
[0086] A selection module 403 is used to divide the target image into a plurality of sub-feature graphs, and perform interval selection on the sub-feature graphs to obtain a sub-feature graph set;
[0087] The processing module 404 is used to perform self-attention processing on the sub-feature graphs in the sub-feature graph set, determine the correlation between different sub-feature graphs, and then aggregate the different sub-feature graphs according to the correlation to obtain aggregated features for identifying the target electromagnetic signal.
[0088] By using the above device, the target electromagnetic signal is first filtered through a neural network to preliminarily eliminate the noise components in the target electromagnetic signal, enhance the features related to the target signal in the spectrum diagram, and make the target signal more prominent. Secondly, the target image is segmented, which helps to distinguish different target areas for subsequent feature extraction and image analysis, and the segmented areas have consistent features and attributes, which is conducive to improving the stability and accuracy in the subsequent feature extraction and classification process; in addition, the sub-feature atlas contains multiple types of features that can reflect different aspects of the target image, and by selecting the sub-feature atlas, the amount of calculation in subsequent processing can be reduced, which helps to improve the processing speed and efficiency of the entire system; finally, self-attention processing is performed to obtain aggregated features, which is conducive to capturing global information of the target image, deeply mining feature correlations, reducing information redundancy, and helping to more accurately understand the content of the target image, thereby improving the accuracy of signal feature extraction.
[0089] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 The steps of a method for processing an electromagnetic signal are provided.
[0090] The present invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 The steps of a method for processing an electromagnetic signal are provided.
[0091] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0092] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0093] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0094] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0095] It should be noted that the above specific implementation method can enable those skilled in the art to more fully understand the invention, but does not limit the invention in any way. Therefore, although the invention has been described in detail in this specification, those skilled in the art should understand that the invention can still be modified or replaced by equivalents; and all technical solutions and improvements that do not deviate from the spirit and scope of the invention are included in the protection scope of the patent for the invention. Any figure mark in the claims should not be regarded as limiting the claims involved.
Claims
1. A method for processing electromagnetic signals, characterized in that: The method comprises: Acquire a target electromagnetic signal, filter the target electromagnetic signal, and obtain a filtered spectrum diagram; Extracting signal features of the spectrum graph in frequency domain characteristics and structure, and mapping the signal features to image space to generate a target image; Dividing the target image into a plurality of sub-feature maps, performing interval selection on the sub-feature maps to obtain a sub-feature map set; Self-attention processing is performed on the sub-feature graphs in the sub-feature graph set to determine the correlation between different sub-feature graphs, and then the different sub-feature graphs are aggregated according to the correlation to obtain aggregated features for identifying target electromagnetic signals.
2. The method for processing electromagnetic signals according to claim 1, characterized in that: The target electromagnetic signal is filtered to obtain a filtered spectrum diagram including: The convolutional neural network CNN and the recurrent neural network RNN are combined to obtain the convolutional recurrent neural network CRNN; the CRNN is trained using pre-acquired training samples to obtain an electromagnetic filtering model; The target electromagnetic signal is processed by Fourier transform, the target electromagnetic signal based on the time domain is converted into a target electromagnetic signal based on the frequency domain, and a frequency spectrum of the target electromagnetic signal is obtained; The spectrum graph is input into the electromagnetic filtering model, the signal features are extracted through the convolutional layer CNN in the CRNN, and then these features are processed through the recurrent layer RNN to predict the filtered spectrum graph.
3. The method for processing electromagnetic signals according to claim 1, characterized in that: Extracting the signal features of the spectrum graph in frequency domain characteristics and structure, and mapping the signal features to the image space, generating the target image includes: Extracting key signal features of the spectrum in frequency domain characteristics and structure through principal component analysis; Converting the key signal features into feature vectors, and performing dimensionality reduction processing on the feature vectors to map the high-dimensional feature vectors into two-dimensional data; A target image is generated according to the two-dimensional data.
4. The method for processing electromagnetic signals according to claim 1, characterized in that: The target image is divided into a plurality of sub-feature maps, and the sub-feature maps are selected at intervals to obtain a sub-feature map set including: Preprocessing the target image to obtain a preprocessed image of a standard size; Cutting the target image based on a preset grid size to obtain a plurality of sub-feature maps; A sub-feature atlas set is obtained by selecting from multiple sub-feature atlases according to a preset interval, and the preset interval is greater than or equal to a sub-feature atlas set.
5. The method for processing electromagnetic signals according to claim 1, characterized in that: Performing self-attention processing on the sub-feature graphs in the sub-feature graph set to determine the correlation between different sub-feature graphs includes: For each sub-feature graph in the sub-feature graph set, calculate its query feature graph, key feature graph and value feature graph, and determine the correlation score between the query feature graph and the key feature graph; Applying a softmax function to convert the correlation scores into a probability distribution to obtain an autocorrelation matrix; Generate attention weights according to the autocorrelation matrix, and use the attention weights to perform weighted summation on the value feature map to obtain the attention output of each sub-feature map; The attention output is subjected to maximum pooling to determine the correlation between different sub-feature maps.
6. The method for processing electromagnetic signals according to claim 1, characterized in that: Aggregating different sub-feature graphs according to the correlation to obtain aggregate features for identifying target electromagnetic signals includes: Initialize an empty aggregate feature vector according to the number and dimension of the sub-feature maps; Traverse the sub-feature graphs one by one, and for each sub-feature graph, aggregate it according to its correlation with other sub-feature graphs; The aggregated sub-feature map features are integrated into an aggregated feature vector, and the aggregated feature vector is normalized to obtain an aggregated feature of the target image, and the aggregated feature is used to perform signal recognition on the target electromagnetic signal.
7. An electromagnetic signal processing device, characterized in that: The device comprises: An acquisition module, used for acquiring a target electromagnetic signal, filtering the target electromagnetic signal, and obtaining a filtered spectrum diagram; A generation module, used for extracting the signal features of the spectrum graph in frequency domain characteristics and structure, and mapping the signal features to the image space to generate a target image; A selection module, used for dividing the target image into a plurality of sub-feature maps, performing interval selection on the sub-feature maps, and obtaining a sub-feature map set; A processing module is used to perform self-attention processing on the sub-feature graphs in the sub-feature graph set, determine the correlation between different sub-feature graphs, and then aggregate the different sub-feature graphs according to the correlation to obtain aggregated features for identifying target electromagnetic signals.
8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.
9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 6 is implemented.