Electromagnetic data identification method and device, electronic equipment and storage medium

Abstract

The invention provides an electromagnetic data recognition method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence, and the method comprises the steps: inputting a to-be-recognized electromagnetic signal to an electromagnetic data analysis model which comprises an electromagnetic signal coding network and a feature recognition network which are connected in series; the electromagnetic signal coding network comprises a plurality of convolutional coding layers and a plurality of feature enhancement layers; the complex convolutional coding layer performs feature extraction on an input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer performs feature enhancement on the input shallow complex features to generate electromagnetic features; and the feature recognition network generates an electromagnetic data recognition result based on the input electromagnetic features. According to the method, the electromagnetic data analysis model comprising the electromagnetic signal coding network and the feature recognition network is creatively utilized, automatic learning and extraction of complex features of the electromagnetic signals are achieved, deep features of the electromagnetic data can be effectively captured, and therefore the efficiency and accuracy of electromagnetic recognition are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an electromagnetic data recognition method, apparatus, electronic device, and storage medium. Background Technology

[0002] Electromagnetic identification refers to the analysis of acquired electromagnetic data to identify specific attributes of the electromagnetic data, such as the communication characteristics, modulation patterns, or technical systems of the electromagnetic data.

[0003] Currently, electromagnetic identification is typically performed using signal analysis tools combined with expert rule bases. The collected electromagnetic signals are transformed and processed by signal analysis tools, and feature parameters such as the transform domain of the electromagnetic signals are manually designed and extracted. The extracted feature parameters are then compared with the expert rule base to achieve electromagnetic identification.

[0004] However, this method relies on manually designed features and pre-established expert knowledge, making it difficult to effectively capture the deep essential characteristics of electromagnetic data, resulting in low efficiency and low accuracy in electromagnetic identification. Summary of the Invention

[0005] The electromagnetic data identification method, device, electronic device, and storage medium provided by this invention are used to solve the defects of low efficiency and low accuracy of electromagnetic identification in the prior art.

[0006] This invention provides an electromagnetic data identification method, comprising the following steps: The electromagnetic signal to be identified is input into the electromagnetic data analysis model to obtain the electromagnetic data identification result output by the electromagnetic data analysis model. The electromagnetic data parsing model includes a cascaded electromagnetic signal encoding network and a feature recognition network; the electromagnetic signal encoding network includes a complex convolutional encoding layer and a complex feature enhancement layer; the complex convolutional encoding layer extracts features from the input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer enhances the input shallow complex features to generate the electromagnetic features; and the feature recognition network generates the electromagnetic data recognition result based on the input electromagnetic features.

[0007] According to an electromagnetic data identification method provided by the present invention, before inputting the electromagnetic signal to be identified into the electromagnetic data analysis model, the method further includes: The electromagnetic signal is subjected to a Fourier transform to obtain an electromagnetic digital signal, which is a complex signal containing in-phase and quadrature components.

[0008] According to an electromagnetic data identification method provided by the present invention, the complex convolutional coding layer includes multiple cascaded complex convolutional blocks, and each complex convolutional block includes a complex convolutional layer, a complex batch normalization layer, and a complex activation function layer; The complex convolutional layer performs complex convolution operations on the in-phase and quadrature components of the complex signal to generate complex features; the complex batch normalization layer normalizes the complex features to generate normalized complex features; and the complex activation function layer performs nonlinear activation on the normalized complex features to generate the shallow complex features.

[0009] According to an electromagnetic data recognition method provided by the present invention, the complex feature enhancement layer includes a complex multi-head attention mechanism module; For each attention head of the complex multi-head attention mechanism module, the following processing is performed: a query matrix, a key matrix, and a value matrix are generated based on the shallow complex features; complex attention calculation is performed based on the query matrix, the key matrix, and the value matrix to generate a complex attention calculation result; The electromagnetic feature is generated by concatenating and linearly transforming the complex attention calculation results generated by all attention heads of the complex multi-head attention mechanism module.

[0010] According to an electromagnetic data identification method provided by the present invention, the electromagnetic signal encoding network further includes a multi-task selection layer, the multi-task selection layer comprising: Multiple expert modules, each of which is used to handle a specific type of electromagnetic identification task; The routing module is used to generate a probability distribution based on the electromagnetic features, and select at least one expert module to process the electromagnetic features based on the probability distribution.

[0011] According to an electromagnetic data identification method provided by the present invention, the electromagnetic data analysis model further includes a pixel reconstruction module, which processes the electromagnetic features in the following manner: The electromagnetic features are segmented according to a preset window size to obtain multiple feature subsequences; Multiple temporally consecutive feature elements within each feature subsequence are combined to generate local aggregated features; Randomly concatenate all local aggregated features; The spliced ​​global aggregated features are subjected to a nonlinear transformation to generate the recombined electromagnetic features.

[0012] According to an electromagnetic data identification method provided by the present invention, the electromagnetic data parsing model further includes a feature alignment mapping layer; The feature alignment mapping layer generates aligned electromagnetic features that match the feature space dimension of the feature recognition network based on the recombined electromagnetic features. The feature recognition network generates the electromagnetic data recognition result based on the input aligned electromagnetic features and text token sequence; the text token sequence is obtained by word segmentation of the user's input text query command. The electromagnetic data identification result includes a text response to the text query command, and the text response includes attribute identification information of electromagnetic digital signals and / or knowledge question and answer information.

[0013] According to an electromagnetic data identification method provided by the present invention, the attribute identification information of the electromagnetic digital signal includes any one of the following: Communication entity identification information, used to identify the individual emitting electromagnetic signals; Modulation pattern identification information is used to identify the modulation method of electromagnetic signals; Technical system identification information is used to identify the communication technology system used in electromagnetic signals.

[0014] According to the electromagnetic data identification method provided by the present invention, the electromagnetic data analysis model is trained based on the following method: Collect an electromagnetic training sample set, which includes multiple sets of training samples. Each set of training samples includes electromagnetic digital signal samples, attribute labels of the electromagnetic digital signal samples, and constructed electromagnetic text question-and-answer pairs. The electromagnetic text question-and-answer pairs include text questions and standard text answers. The electromagnetic signal encoding network is trained based on the electromagnetic digital signal samples and the attribute labels until the electromagnetic attribute recognition accuracy reaches a first preset threshold. Freeze the parameters of the feature recognition network and iteratively perform fine-tuning of the electromagnetic signal encoding network and the feature alignment mapping layer until the second preset cutoff condition is met; The fine-tuning process of the electromagnetic signal encoding network and the feature alignment mapping layer includes: Select a batch of electromagnetic text question-and-answer pairs from the electromagnetic training sample set; The electromagnetic digital signal and the text question are input into the electromagnetic data analysis model to obtain the predicted text response; Calculate the loss between the predicted text response and the standard text response; The parameters of the electromagnetic signal encoding network and the feature alignment mapping layer are updated based on the loss.

[0015] The present invention also provides an electromagnetic data identification device, comprising the following modules: An electromagnetic signal input unit is used to input the electromagnetic signal to be identified into the electromagnetic data analysis model; The identification result output unit is used to obtain the electromagnetic data identification result output by the electromagnetic data analysis model; The electromagnetic data parsing model includes a cascaded electromagnetic signal encoding network and a feature recognition network; the electromagnetic signal encoding network includes a complex convolutional encoding layer and a complex feature enhancement layer; the complex convolutional encoding layer extracts features from the input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer enhances the input shallow complex features to generate the electromagnetic features; and the feature recognition network generates the electromagnetic data recognition result based on the input electromagnetic features.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the electromagnetic data identification method as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the electromagnetic data identification method as described above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the electromagnetic data identification method as described above.

[0019] The electromagnetic data identification method, device, electronic device, and storage medium provided by this invention innovatively utilize an electromagnetic data analysis model that includes an electromagnetic signal encoding network and a feature recognition network to achieve automatic learning and extraction of complex features of electromagnetic signals. This effectively captures the deep features of electromagnetic data, thereby improving the efficiency and accuracy of electromagnetic identification. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts of the electromagnetic data identification method provided by the present invention.

[0022] Figure 2 This is a schematic diagram of the electromagnetic signal encoding network provided by the present invention.

[0023] Figure 3 This is a flowchart illustrating the complex multi-head attention mechanism module provided by the present invention.

[0024] Figure 4 This is a flowchart illustrating the multi-task selection layer provided by the present invention.

[0025] Figure 5 This is a schematic diagram of the process by which the pixel recombination module provided by the present invention processes electromagnetic features.

[0026] Figure 6 This is a schematic diagram of the electromagnetic data analysis model provided by the present invention.

[0027] Figure 7 This is a schematic diagram of the training process of the electromagnetic data analysis model provided by the present invention.

[0028] Figure 8 This is a schematic diagram of the question-and-answer interface of the electromagnetic data analysis model provided by the present invention.

[0029] Figure 9 This is a schematic diagram of the electromagnetic data identification device provided by the present invention.

[0030] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this 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 this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0032] It should be noted that, in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0033] The terms "first," "second," etc., used in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0034] The following is combined Figures 1-10 This invention describes the electromagnetic data identification method, apparatus, electronic device, and storage medium provided by the present invention.

[0035] Figure 1 This is one of the flowcharts illustrating the electromagnetic data identification method provided by the present invention, such as... Figure 1 As shown, the execution subject of the electromagnetic data identification method provided by the present invention can be a server, a cloud computing platform, or a computer capable of executing the method of the present invention, etc. Unless otherwise specified, the following embodiments will be described using a server as an example.

[0036] As an optional embodiment, this electromagnetic data identification method mainly includes, but is not limited to, the following steps: Step 110: Input the electromagnetic signal to be identified into the electromagnetic data analysis model.

[0037] The electromagnetic data analysis model employed mainly includes a cascaded electromagnetic signal encoding network and a feature recognition network. The electromagnetic signal encoding network primarily consists of a complex convolutional coding layer and a complex feature enhancement layer. The complex convolutional coding layer extracts features from the input electromagnetic signal, generating shallow complex features. The complex feature enhancement layer enhances the input shallow complex features, generating electromagnetic features. The feature recognition network generates electromagnetic data recognition results based on the input electromagnetic features.

[0038] The electromagnetic signals to be identified can be collected using specialized acquisition equipment such as electronic reconnaissance equipment or software-defined radio (SDR) equipment. The collected electromagnetic signals are usually sampling point data containing in-phase and quadrature components, and are presented in the form of a one-dimensional time series.

[0039] The electromagnetic data analysis model can be a large electromagnetic-text multimodal model. The electromagnetic signal to be identified can be input into the electromagnetic data analysis model through various methods such as file upload and data stream interface.

[0040] This electromagnetic data analysis model comprises a cascaded electromagnetic signal encoding network and a feature recognition network. The electromagnetic signal encoding network is specifically designed to process the characteristics of electromagnetic signals. This network includes a complex convolutional coding layer, which receives the input electromagnetic signal and utilizes its complex processing capabilities to perform preliminary feature encoding and multi-scale feature extraction to generate shallow complex features. For example, the complex convolutional coding layer can process the original electromagnetic signal... x =( x 1 , x 2 , ..., x n The signal is transformed and downsampled to obtain a shallow complex feature that can preliminarily characterize the signal. z =( z 1 , z 2 , ..., z m ).

[0041] Concatenated after the complex convolutional coding layer is a complex feature enhancement layer. Its function is to perform deeper information aggregation and contextual semantic extraction on the shallow complex features output by the complex convolutional coding layer, generating electromagnetic features richer in information. For example, the complex feature enhancement layer can further process the shallow complex features, transforming them into a high-level semantic complex feature sequence. c =( c 1 , c 2 , ..., c k ).

[0042] Electromagnetic features generated by an electromagnetic signal encoding network are input into a feature recognition network. This feature recognition network can be a pre-trained large language model used to receive the electromagnetic features and generate the final electromagnetic data recognition result based on these features. This electromagnetic data recognition result can be a classification judgment of a certain attribute of the electromagnetic signal, or a natural language text response generated in conjunction with a user's question, thereby achieving intelligent analysis and interactive question-and-answering of electromagnetic data.

[0043] Step 120: Obtain the electromagnetic data identification results output by the electromagnetic data analysis model.

[0044] Obtaining electromagnetic data identification results can be achieved through an interactive user interface. For example, an electromagnetic data analysis model question-and-answer system can be built. Users can submit electromagnetic data files to be analyzed and input text query commands through the system's graphical user interface (GUI). The electromagnetic data identification results generated by a feature recognition network (such as a large language model) will then be displayed directly to the user in the form of natural language text within a dialog window. This output text can include not only direct analytical conclusions but also explanations of the analysis process and elaborations on relevant electromagnetic knowledge, thus providing a rich interactive experience.

[0045] As an alternative embodiment, the electromagnetic data identification result is obtained through an Application Programming Interface (API). In this way, other programs or systems can invoke the electromagnetic data parsing model of the present invention, which then returns the generated text response in the form of structured data (e.g., JSON format) to facilitate subsequent automated processing and system integration.

[0046] The electromagnetic data identification method provided by this invention innovatively utilizes an electromagnetic data analysis model that includes an electromagnetic signal encoding network and a feature recognition network to achieve automatic learning and extraction of complex features of electromagnetic signals. This effectively captures the deep features of electromagnetic data, thereby improving the efficiency and accuracy of electromagnetic identification.

[0047] In another embodiment of the present invention, before inputting the electromagnetic signal to be identified into the electromagnetic data analysis model, the method further includes: performing a Fourier transform on the electromagnetic signal to obtain an electromagnetic digital signal, wherein the electromagnetic digital signal is a complex signal containing in-phase components and quadrature components.

[0048] Fourier transform of an electromagnetic signal is a signal preprocessing step used to convert the received raw electromagnetic signal from the time domain to the frequency domain and demodulate it to baseband or intermediate frequency to obtain its in-phase and quadrature components.

[0049] Electromagnetic digital signals refer to one-dimensional complex time series obtained by performing Fourier transform and digital sampling on electromagnetic signals, which completely preserves the amplitude and phase information of the original electromagnetic signals. For example, electromagnetic digital signals are electromagnetic signal sampling point data that can be directly input into electromagnetic data analysis models, where each sampling point is a complex number composed of a real part (in-phase component) and an imaginary part (quadrature component).

[0050] The electromagnetic data identification method provided by this invention obtains a complex electromagnetic digital signal containing in-phase and quadrature components by performing a Fourier transform on the electromagnetic signal before processing by the electromagnetic data analysis model. This method can completely preserve the key amplitude and phase information in the original electromagnetic signal, providing a standardized and complete input for subsequent complex network processing. This lays the foundation for the electromagnetic data analysis model to learn and extract electromagnetic features more accurately.

[0051] Figure 2 This is a schematic diagram of the electromagnetic signal encoding network provided by the present invention, as shown below. Figure 2 As shown, as another optional embodiment provided by the present invention, the complex convolutional coding layer includes multiple cascaded complex convolutional blocks, each complex convolutional block including a complex convolutional layer, a complex batch normalization layer and a complex activation function layer.

[0052] Specifically, the complex convolutional layer is responsible for performing complex convolution operations on the in-phase and quadrature components of the input complex signal to generate complex features. Complex convolution operation refers to extending the traditional convolution operation to the complex domain, where both the input features and the convolution kernel weights are complex numbers, thereby better capturing and preserving the intrinsic phase relationship between the in-phase and quadrature components of the electromagnetic signal.

[0053] For example, when the input to a complex convolutional layer is an electromagnetic digital signal X = a + ib The complex convolution kernel weights are W=A+iB The convolution result can be calculated in the following way:

[0054] in, Represents the complex convolution kernel weights. Represents electromagnetic digital signals, This represents the convolution operation. The real part of the weights of the complex convolution kernel is represented. The imaginary part of the weights of the complex convolution kernel is represented. Represents the real part of an electromagnetic digital signal. The imaginary part of an electromagnetic digital signal. i It is the imaginary unit.

[0055] Furthermore, the complex batch normalization layer normalizes the complex features output by the complex convolutional layer. This process aims to adjust the distribution of complex features, for example, by jointly normalizing the statistics (such as mean and variance) of the real and imaginary parts of the complex features within a batch, thereby accelerating model convergence and improving the stability of the training process.

[0056] Complex activation function layers perform nonlinear activation on the normalized complex features to increase the nonlinear expressive power of the entire electromagnetic signal coding network, enabling it to learn and fit more complex patterns in electromagnetic signals. For example, nonlinear activation can use the Complex Gaussian Error Linear Unit (complexGELU) as the activation function.

[0057] As an optional embodiment, the complex convolutional layer consists of 5 cascaded complex convolutional blocks. The kernel size, stride, and other parameters of each layer can be set to different values ​​to achieve multi-scale feature extraction and downsampling of the input electromagnetic signal layer by layer, and finally output a shallow complex feature containing rich information for use by subsequent complex feature enhancement layers.

[0058] The electromagnetic data recognition method provided by this invention combines complex convolutional layers, complex batch normalization layers, and complex activation function layers into a series-connected complex convolutional block, which can construct a stable deep complex network structure. While ensuring the stability of the training process, it enhances the model's ability to express complex electromagnetic patterns nonlinearly, thereby achieving deeper and more effective feature extraction of electromagnetic signals.

[0059] Figure 3 This is a flowchart illustrating the complex multi-head attention mechanism module provided by the present invention, as shown below. Figure 3 As shown, as another optional embodiment provided by the present invention, the complex feature enhancement layer includes a complex multi-head attention mechanism module, including but not limited to the following steps: Step 310: For each attention head of the complex multi-head attention mechanism module, perform the following processing: generate a query matrix, a key matrix, and a value matrix based on the shallow complex features; perform complex attention calculation based on the query matrix, the key matrix, and the value matrix to generate a complex attention calculation result.

[0060] Specifically, for the shallow complex features of the input Z=M+iN ,in M and N These are the real and imaginary parts of the complex feature sequence, respectively, which can be obtained through a learnable real-valued weight matrix. W Q , W K as well as W V Shallow complex features Z Perform a linear projection to obtain a complex-valued query matrix. Q=ZW Q Key matrix K=ZW K and value matrix V=ZW V Next, we perform complex attention calculations, the core idea of ​​which is to calculate... Q , K transpose K T as well as V The complex matrix product of the three components, and the complete expansion of the calculation process are shown below:

[0061] in, The real part of the complex attention calculation result is represented. The imaginary part of the result of complex attention calculation is represented. i It is the imaginary unit.

[0062] Step 320: The complex attention calculation results generated by all attention heads of the complex multi-head attention mechanism module are spliced ​​and linearly transformed to generate the electromagnetic feature.

[0063] To efficiently implement the complex attention calculation described above, this embodiment decomposes it into a combined operation of multiple parallel real-field multi-head attention (MH) modules. Each MH module MH( Q , K , V The standard scaled dot product attention calculation is performed, which involves calculating the dot product of the query matrix and the key matrix to obtain a similarity score, normalizing the result using the Softmax function to obtain attention weights, and finally performing a weighted summation of the value matrices.

[0064] Based on this, the above complex attention calculation can be achieved using the following formula:

[0065] in, Y For the complex features of the input, M and N These are their corresponding real and imaginary parts, respectively. This indicates that its query matrix, key matrix, and value matrix all come from the real part. M , This indicates that the query matrix comes from the real part. M The key matrix and value matrix come from the imaginary part. N , This indicates that the query matrix comes from the imaginary part. N The key matrix comes from the real part. M The value matrix comes from the imaginary part. N , This indicates that the query matrix and key matrix come from the imaginary part. N The value matrix comes from the real part. M , This indicates that the query matrix and key matrix are derived from the real part. M The value matrix comes from the imaginary part. N , This indicates that the query matrix and key matrix come from the imaginary part. N The value matrix comes from the real part. M , This indicates that the query matrix comes from the imaginary part. N The key matrix and value matrix come from the real part. M , This indicates that the query matrix, key matrix, and value matrix all originate from the imaginary part. N .

[0066] It should be noted that each attention head in the complex multi-head attention mechanism focuses on a different subspace of the input features. This invention concatenates the complex features output by all attention heads along the channel dimension, and then inputs the concatenated high-dimensional features into a final linear transformation layer for feature fusion and dimensionality reduction, thereby generating electromagnetic features with high-level, deep semantics as the final output of this layer. The entire complex feature enhancement layer, together with components such as residual connections (Add) and layer normalization (Norm), can form a complete complex Transformer Encoder layer, and can be stacked N times to enhance the model's feature extraction capabilities.

[0067] The electromagnetic data recognition method provided by this invention, through a complex multi-head attention mechanism, can effectively capture long-distance dependencies in the signal sequence while maintaining the complex characteristics of the electromagnetic signal, thus providing richer and more accurate feature representations for subsequent recognition tasks.

[0068] Figure 4 This is a flowchart illustrating the multi-task selection layer provided by the present invention, such as... Figure 4 As shown, as another optional embodiment provided by the present invention, the electromagnetic signal coding network further includes a multi-task selection layer, which includes: multiple expert modules, each expert module being used to process a specific type of electromagnetic identification task; and a routing module being used to generate a probability distribution based on electromagnetic features and select at least one expert module to process the electromagnetic features based on the probability distribution.

[0069] Electromagnetic identification tasks refer to the classification or analysis of specific attributes of electromagnetic signals. For example, electromagnetic identification tasks can be the identification of communication entities, modulation patterns (such as 2ASK, 8PSK, etc.), and technical systems (such as Bluetooth, WIFI, etc.) of electromagnetic signals.

[0070] The probability distribution refers to a weight or score output by the routing module for each expert module. This score represents the degree of matching of each expert module in processing the current input electromagnetic features. For example, for four expert modules, namely expert 1, expert 2, expert 3, and expert 4, the routing module may output a probability distribution such as [0.1, 0.8, 0.05, 0.05]. In this case, the routing module will select expert 2 with the highest probability value to participate in the calculation based on the probability distribution, while the other expert modules will not be activated, thereby flexibly allocating computing resources according to the input characteristics.

[0071] The electromagnetic data identification method provided by this invention introduces a multi-task selection layer composed of a routing module and multiple expert modules, and dynamically selects a specific expert module for processing based on input features. This enables unified modeling of various electromagnetic identification tasks such as communication individuals and modulation patterns within a single model framework, while significantly improving the inference efficiency and task expansion capability of the electromagnetic data parsing model.

[0072] Figure 5 This is a schematic diagram of the pixel recombination module provided by the present invention processing electromagnetic features, as shown below. Figure 5 As shown, as another optional embodiment provided by the present invention, the electromagnetic data parsing model further includes a pixel reconstruction module. The pixel reconstruction module processes the electromagnetic features in the following ways, including but not limited to the following steps: Step 510: Divide the electromagnetic features into segments according to a preset window size to obtain multiple feature subsequences.

[0073] Specifically, the continuous one-dimensional electromagnetic feature sequence received from the complex feature enhancement layer is segmented in the time dimension. For example, if the input electromagnetic feature is a time series of length 1024 and the preset window size is 64, the sequence will be uniformly and non-overlappingly segmented into 1024 / 64=16 continuous feature subsequences, each of which has a length of 64.

[0074] Step 520: Combine multiple temporally consecutive feature elements within each feature subsequence to generate local aggregated features.

[0075] Specifically, each longer feature subsequence is compressed into one or a few feature vectors that can represent its core information. This combination operation can be a pooling operation, such as performing mean pooling or max pooling on all feature elements within each feature subsequence. For example, after performing pooling operations on the aforementioned 16 subsequences of length 64, 16 corresponding local aggregated feature vectors will be obtained.

[0076] Step 530: Randomly concatenate all local aggregated features.

[0077] Specifically, this is the core step in achieving the shuffle. It breaks the inherent order of the individual feature fragments in the original time series. For example, after obtaining 16 local aggregated features in the previous step, they still logically maintain their temporal order (numbered 1 to 16). This step generates a random index sequence, such as [5, 12, 2, ..., 9], and then, based on this new random order, rearranges and concatenates these 16 local aggregated feature vectors to form a global aggregated feature that is related to the length or dimension of the original input but whose order has been shuffled.

[0078] Step 540: Perform a nonlinear transformation on the spliced ​​global aggregated features to generate the recombined electromagnetic features.

[0079] Specifically, the globally aggregated features can be input into a nonlinear transformation module, which is typically a multilayer perceptron (MLP). Feature mapping and integration are performed through fully connected layers and activation functions in the MLP, ultimately outputting recombined electromagnetic features with stronger position invariance and generalization capabilities.

[0080] The electromagnetic data recognition method provided by this invention, by employing a pixel recombination module to randomly splice and recombine local fragments of electromagnetic features, can break the fixed positional dependency of features in the time series, allowing the electromagnetic data analysis model to focus more on the intrinsic semantics of the features themselves, thereby significantly enhancing the generalization ability and recognition robustness of the electromagnetic data analysis model.

[0081] Figure 6 This is a schematic diagram of the electromagnetic data analysis model provided by the present invention, as shown below. Figure 6 As shown, as another optional embodiment provided by the present invention, the electromagnetic data parsing model further includes a feature alignment mapping layer; the feature alignment mapping layer generates aligned electromagnetic features that match the feature space dimension of the feature recognition network based on the recombined electromagnetic features; the feature recognition network generates electromagnetic data recognition results based on the input aligned electromagnetic features and text token sequence; the text token sequence is obtained by segmenting the text query command input by the user; the electromagnetic data recognition results include a text response to the text query command, and the text response includes attribute recognition information of electromagnetic digital signals and / or knowledge question answering information.

[0082] Specifically, the feature alignment mapping layer consists of a linear transformation layer and a GELU activation function. It projects and transforms the feature vector representing electromagnetic information output by the electromagnetic signal encoding network into a feature space of the same dimension as the feature recognition network (i.e., the large language model). The tokenizer can convert the user's input text query command into a text token sequence that the large language model can understand. The large language model takes the aligned electromagnetic features and the text token sequence as input, performs cross-modal semantic understanding and reasoning, and generates a natural language text as a text response to the text query command.

[0083] For example, a user inputs an electromagnetic signal and simultaneously enters the question, "What is the modulation style of this electromagnetic signal?" in the text box. The feature alignment mapping layer converts the electromagnetic features extracted from the signal into aligned electromagnetic features, and the token segmenter converts the question into a sequence of text tokens. After receiving these two inputs, the large language model outputs attribute recognition information, such as the text response: "Based on electromagnetic feature extraction and pattern recognition, it is determined that this signal uses 16-QAM modulation." If the user continues with a knowledge-based question, such as, "Explain what 16QAM modulation is," the large language model can further generate detailed explanatory text, thus achieving intelligent question-and-answer functionality for electromagnetic knowledge.

[0084] The electromagnetic data recognition method provided by this invention introduces a feature alignment mapping layer to perform cross-modal semantic alignment between electromagnetic features and text, and uses a large language model to jointly understand and generate the fused information. This method can overcome the limitation of traditional recognition models that only output fixed category labels, and enhance the human-computer interaction capability of the electromagnetic recognition system.

[0085] In another embodiment of the present invention, the attribute identification information of the electromagnetic digital signal includes any one of the following: communication individual identification information, used to identify the individual transmitting the electromagnetic signal; modulation pattern identification information, used to identify the modulation method of the electromagnetic signal; and technical system identification information, used to identify the communication technical system adopted by the electromagnetic signal.

[0086] An individual source of electromagnetic signals refers to a physical device that generates and emits electromagnetic signals.

[0087] The modulation method of an electromagnetic signal refers to the specific technical means of loading a modulating signal onto a high-frequency carrier wave. For example, the modulation method can be 2-level amplitude-shift keying (2ASK), 8-phase shift keying (8PSK), or 16-quadrature amplitude modulation (16-QAM).

[0088] A communication technology system refers to a set of communication protocols, standards, or system specifications that electromagnetic signals follow. For example, communication technology systems can include Bluetooth, Wi-Fi, CDMA2000, and 5G New Radio (5GNR).

[0089] The electromagnetic data identification method provided by this invention, by concretizing the electromagnetic data identification results into key attribute information such as communication individual, modulation pattern and technical system, can clarify the specific application direction and practical problems solved by this invention, so that it can directly serve the core business needs in the field of electromagnetic sensing and identification.

[0090] Figure 7 This is a schematic diagram of the training process of the electromagnetic data analysis model provided by the present invention, as shown below. Figure 7 As shown, as another optional embodiment provided by the present invention, the electromagnetic data analysis model is trained based on the following method: Step 710: Collect an electromagnetic training sample set. The electromagnetic training sample set includes multiple sets of training samples. Each set of training samples includes electromagnetic digital signal samples, attribute labels of electromagnetic digital signal samples, and constructed electromagnetic text question-and-answer pairs. The electromagnetic text question-and-answer pairs include text questions and standard text answers.

[0091] For example, an electromagnetic training sample may include a specific electromagnetic digital signal sample, an attribute label corresponding to the sample (such as the modulation style label 8PSK), and an electromagnetic text question-and-answer pair built based on the label. The text question may be "What is the modulation style of this electromagnetic signal?", while the standard text answer is "This electromagnetic signal uses 8PSK modulation".

[0092] Step 720: Train the electromagnetic signal encoding network based on electromagnetic digital signal samples and attribute labels until the electromagnetic attribute recognition accuracy reaches the first preset threshold.

[0093] Specifically, the electromagnetic signal coding network receives electromagnetic digital signal samples as input and outputs prediction results of the attributes of those samples. The prediction results are compared with the true attribute labels, and the loss is calculated using the cross-entropy loss function. The parameters of the electromagnetic signal coding network are then continuously optimized using the backpropagation algorithm until a preset cutoff condition is met.

[0094] For example, for training tasks involving the identification of technical systems, the cutoff condition can be set to the electromagnetic signal encoding network achieving a recognition accuracy of 96% on the validation set. Once this threshold is reached, the encoding network can be considered sufficiently trained, and this pre-training phase can be terminated.

[0095] Step 730: Freeze the parameters of the feature recognition network and iteratively perform the fine-tuning process of the electromagnetic signal encoding network and the feature alignment mapping layer until the second preset cutoff condition is met.

[0096] The fine-tuning process for the electromagnetic signal encoding network and feature alignment mapping layer includes: selecting a batch of electromagnetic text question-answer pairs from the electromagnetic training sample set; inputting the electromagnetic digital signal and text question into the electromagnetic data parsing model to obtain the predicted text response; calculating the loss between the predicted text response and the standard text response; and updating the parameters of the electromagnetic signal encoding network and feature alignment mapping layer based on the loss.

[0097] Considering that the feature recognition network (i.e., the large language model) already has the ability to understand general knowledge and generate text, if its parameters are also updated during the fine-tuning process for the electromagnetic domain, it may destroy the general knowledge learned in the pre-training stage, resulting in a catastrophic decline in its performance in general question-answering scenarios. Therefore, this invention freezes the parameters of the feature recognition network during the fine-tuning process of the electromagnetic signal encoding network and the feature alignment mapping layer.

[0098] For example, when calculating the loss between the predicted text response and the standard text response, a cross-entropy loss function can be used to update the parameters of the electromagnetic signal encoding network and the feature alignment mapping layer. The second preset cutoff condition could be that training reaches a preset number of iterations, or that the loss on the validation set no longer decreases significantly.

[0099] It should be noted that the evaluation metrics for electromagnetic data parsing models can include recognition accuracy, Bilingual Evaluation Understudy (BLEU), Recall-Oriented Understudy for Gisting Evaluation (ROUGE), and Metric for Evaluation of Translation with Explicit Ordering (METEOR). Recognition accuracy, used for electromagnetic data classification tasks, calculates the proportion of correctly predicted samples out of the total samples, reflecting the model's basic judgment ability. BLEU measures the similarity of the generated text response to the standard text response in terms of lexical fragments, assessing the sentence coherence of the generated content. ROUGE focuses on evaluating the recall rate of the generated content to the core information in the standard text response, judging the completeness of the content. METEOR integrates word matching and semantic matching, considering synonyms, semantic similarity, and word order differences to provide a score closer to semantic consistency.

[0100] The electromagnetic data recognition method provided by this invention adopts a two-stage training strategy of pre-training and alignment fine-tuning, and freezes the parameters of the feature recognition network in the alignment fine-tuning stage. This method can efficiently align professional knowledge in the electromagnetic field to the semantic space of a large language model, while fully preserving the powerful general knowledge and text generation capabilities of the large language model. It avoids the catastrophic forgetting problem in the training process, thereby improving the training efficiency and final performance of the electromagnetic data parsing model.

[0101] Figure 8 This is a schematic diagram of the question-and-answer interface of the electromagnetic data analysis model provided by the present invention, as shown below. Figure 8 As shown, this invention provides users with a human-computer interaction platform. The left side of the interface displays a "System" response area, while the right side shows a "User" question area, clearly presenting the multi-round question-and-answer process between the user and the system. The system first responds with "Data uploaded successfully," indicating that the electromagnetic data has been successfully uploaded. Then, based on the characteristics of the input electromagnetic signal, the system identifies and explains the modulation method, such as determining that the input signal uses "16QAM modulation type," and describes in detail the characteristics of 16QAM modulation and its typical application scenarios in digital communication. Users can input text query commands through the interface, such as "What is the modulation style of this data?" or "Explain what 16QAM modulation is." The system then provides natural language responses based on cross-modal multi-task reasoning between electromagnetic data and text, covering electromagnetic signal attribute recognition and knowledge-based question answering, demonstrating the intelligent interactive capability of this invention's large language model aligned with electromagnetic signal features. At the bottom of the interface are a file selection box and a mode selection area. Users can easily upload electromagnetic data files, select "Electromagnetic data + text mode" or "Plain text mode" as the interaction mode, enter their question in the input field, and click "Send Question" to receive the question-and-answer service. This interface design realizes the complete workflow of the electromagnetic data analysis model from uploading raw electromagnetic signal data to cross-modal semantic understanding and knowledge question answering output, demonstrating the practicality and interactivity of the multimodal electromagnetic analysis technology of this invention.

[0102] Figure 9 This is a schematic diagram of the electromagnetic data identification device provided by the present invention, as shown below. Figure 9 As shown, it mainly includes, but is not limited to: The electromagnetic signal input unit 910 is used to input the electromagnetic signal to be identified into the electromagnetic data analysis model.

[0103] The recognition result output unit 920 is used to obtain the electromagnetic data recognition result output by the electromagnetic data analysis model.

[0104] The electromagnetic data parsing model includes a cascaded electromagnetic signal encoding network and a feature recognition network; the electromagnetic signal encoding network includes a complex convolutional encoding layer and a complex feature enhancement layer; the complex convolutional encoding layer extracts features from the input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer enhances the input shallow complex features to generate the electromagnetic features; and the feature recognition network generates the electromagnetic data recognition result based on the input electromagnetic features.

[0105] It should be noted that the electromagnetic data identification device provided by the present invention can execute the electromagnetic data identification method described in any of the above embodiments during specific operation, and this embodiment will not elaborate on this.

[0106] The electromagnetic data identification device provided by this invention innovatively utilizes an electromagnetic data analysis model that includes an electromagnetic signal encoding network and a feature recognition network to achieve automatic learning and extraction of complex features of electromagnetic signals. This effectively captures the deep features of electromagnetic data, thereby improving the efficiency and accuracy of electromagnetic identification.

[0107] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 10 As shown, the electronic device may include a processor 1010, a communications interface 1020, a memory 1030, and a communication bus 1040, wherein the processor 1010, the communications interface 1020, and the memory 1030 communicate with each other via the communication bus 1040. The processor 1010 can call logical instructions in the memory 1030 to execute an electromagnetic data identification method. This method includes: inputting an electromagnetic signal to be identified into an electromagnetic data analysis model, and obtaining an electromagnetic data identification result output by the electromagnetic data analysis model; the electromagnetic data analysis model includes a cascaded electromagnetic signal encoding network and a feature recognition network; the electromagnetic signal encoding network includes a complex convolutional coding layer and a complex feature enhancement layer; the complex convolutional coding layer extracts features from the input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer enhances the input shallow complex features to generate the electromagnetic features; and the feature recognition network generates the electromagnetic data identification result based on the input electromagnetic features.

[0108] Furthermore, the logical instructions in the aforementioned memory 1030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0109] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the electromagnetic data recognition method provided in the above embodiments, the method comprising: inputting an electromagnetic signal to be recognized into an electromagnetic data analysis model, and obtaining an electromagnetic data recognition result output by the electromagnetic data analysis model; the electromagnetic data analysis model comprising a cascaded electromagnetic signal encoding network and a feature recognition network; the electromagnetic signal encoding network comprising a complex convolutional encoding layer and a complex feature enhancement layer; the complex convolutional encoding layer performing feature extraction on the input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer performing feature enhancement on the input shallow complex features to generate the electromagnetic features; and the feature recognition network generating the electromagnetic data recognition result based on the input electromagnetic features.

[0110] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the electromagnetic data recognition method provided in the above embodiments. The method includes: inputting an electromagnetic signal to be recognized into an electromagnetic data analysis model to obtain an electromagnetic data recognition result output by the electromagnetic data analysis model; the electromagnetic data analysis model includes a cascaded electromagnetic signal encoding network and a feature recognition network; the electromagnetic signal encoding network includes a complex convolutional coding layer and a complex feature enhancement layer; the complex convolutional coding layer extracts features from the input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer enhances the input shallow complex features to generate the electromagnetic features; and the feature recognition network generates the electromagnetic data recognition result based on the input electromagnetic features.

[0111] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0112] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An electromagnetic data identification method, characterized in that, include: The electromagnetic signal to be identified is input into the electromagnetic data analysis model to obtain the electromagnetic data identification result output by the electromagnetic data analysis model. The electromagnetic data parsing model includes a cascaded electromagnetic signal encoding network and a feature recognition network; the electromagnetic signal encoding network includes a complex convolutional encoding layer and a complex feature enhancement layer; the complex convolutional encoding layer extracts features from the input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer enhances the input shallow complex features to generate electromagnetic features. The feature recognition network generates the electromagnetic data recognition result based on the input electromagnetic features.

2. The electromagnetic data identification method according to claim 1, characterized in that, Before inputting the electromagnetic signal to be identified into the electromagnetic data analysis model, the following steps are also included: The electromagnetic signal is subjected to a Fourier transform to obtain an electromagnetic digital signal, which is a complex signal containing in-phase and quadrature components.

3. The electromagnetic data identification method according to claim 2, characterized in that, The complex convolutional coding layer includes multiple cascaded complex convolutional blocks, each of which includes a complex convolutional layer, a complex batch normalization layer, and a complex activation function layer. The complex convolutional layer performs complex convolution operations on the in-phase and quadrature components of the complex signal to generate complex features; the complex batch normalization layer normalizes the complex features to generate normalized complex features; and the complex activation function layer performs nonlinear activation on the normalized complex features to generate the shallow complex features.

4. The electromagnetic data identification method according to claim 1, characterized in that, The complex feature enhancement layer includes a complex multi-head attention mechanism module; For each attention head of the complex multi-head attention mechanism module, the following processing is performed: a query matrix, a key matrix, and a value matrix are generated based on the shallow complex features; complex attention calculation is performed based on the query matrix, the key matrix, and the value matrix to generate a complex attention calculation result; The electromagnetic feature is generated by concatenating and linearly transforming the complex attention calculation results generated by all attention heads of the complex multi-head attention mechanism module.

5. The electromagnetic data identification method according to claim 1, characterized in that, The electromagnetic signal encoding network further includes a multi-task selection layer, which includes: Multiple expert modules, each of which is used to handle a specific type of electromagnetic identification task; A routing module is used to generate a probability distribution based on the electromagnetic features, and select at least one of the expert modules to process the electromagnetic features based on the probability distribution.

6. The electromagnetic data identification method according to claim 1, characterized in that, The electromagnetic data analysis model also includes a pixel reconstruction module, which processes the electromagnetic features in the following way: The electromagnetic features are segmented according to a preset window size to obtain multiple feature subsequences; Multiple temporally consecutive feature elements within each feature subsequence are combined to generate local aggregated features; Randomly concatenate all local aggregated features; The spliced ​​global aggregated features are subjected to a nonlinear transformation to generate the recombined electromagnetic features.

7. The electromagnetic data identification method according to claim 6, characterized in that, The electromagnetic data parsing model also includes a feature alignment mapping layer; The feature alignment mapping layer generates aligned electromagnetic features that match the feature space dimension of the feature recognition network based on the recombined electromagnetic features. The feature recognition network generates the electromagnetic data recognition result based on the input aligned electromagnetic features and text token sequence; the text token sequence is obtained by segmenting the text query command input by the user. The electromagnetic data identification result includes a text response to the text query command, and the text response includes attribute identification information of electromagnetic digital signals and / or knowledge question and answer information.

8. The electromagnetic data identification method according to claim 7, characterized in that, The attribute identification information of the electromagnetic digital signal includes any one of the following: Communication entity identification information, used to identify the individual emitting electromagnetic signals; Modulation pattern identification information is used to identify the modulation method of electromagnetic signals; Technical system identification information is used to identify the communication technology system used in electromagnetic signals.

9. The electromagnetic data identification method according to claim 7, characterized in that, The electromagnetic data analysis model was trained based on the following method: Collect an electromagnetic training sample set, which includes multiple sets of training samples. Each set of training samples includes electromagnetic digital signal samples, attribute labels of the electromagnetic digital signal samples, and electromagnetic text question-and-answer pairs. The electromagnetic text question-and-answer pairs include text questions and standard text answers. The electromagnetic signal encoding network is trained based on the electromagnetic digital signal samples and the attribute labels until the electromagnetic attribute recognition accuracy reaches a first preset threshold. Freeze the parameters of the feature recognition network and iteratively perform fine-tuning of the electromagnetic signal encoding network and the feature alignment mapping layer until the second preset cutoff condition is met; The fine-tuning process of the electromagnetic signal encoding network and the feature alignment mapping layer includes: Select a batch of electromagnetic text question-and-answer pairs from the electromagnetic training sample set; The electromagnetic digital signal and the text question are input into the electromagnetic data analysis model to obtain the predicted text response; Calculate the loss between the predicted text response and the standard text response; The parameters of the electromagnetic signal encoding network and the feature alignment mapping layer are updated based on the loss.

10. An electromagnetic data identification device, characterized in that, include: An electromagnetic signal input unit is used to input the electromagnetic signal to be identified into the electromagnetic data analysis model; The identification result output unit is used to obtain the electromagnetic data identification result output by the electromagnetic data analysis model; The electromagnetic data parsing model includes a cascaded electromagnetic signal encoding network and a feature recognition network; the electromagnetic signal encoding network includes a complex convolutional encoding layer and a complex feature enhancement layer; the complex convolutional encoding layer extracts features from the input electromagnetic signal to generate shallow complex features; the complex feature enhancement layer enhances the input shallow complex features to generate electromagnetic features. The feature recognition network generates the electromagnetic data recognition result based on the input electromagnetic features.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the electromagnetic data identification method as described in any one of claims 1 to 9.

12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the electromagnetic data identification method as described in any one of claims 1 to 9.

13. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the electromagnetic data identification method as described in any one of claims 1 to 9.

Images