Electromagnetic signal identification method based on distributed training electromagnetic identification model
By distributing the training of electromagnetic recognition models, the electromagnetic signal data is dispersed to multiple computing nodes for processing, and the model parameters are optimized using an adaptive comprehensive loss function, which solves the bottleneck problem of computing resources and achieves efficient and accurate electromagnetic signal recognition.
Patent Information
- Application Number
- CN202510742503.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-05
- Publication Date
- 2025-09-16
AI Technical Summary
Existing electromagnetic signal recognition methods face computing resource bottlenecks when processing large-scale electromagnetic signal data, with high hardware costs and low computing efficiency, making it difficult to meet real-time requirements.
A distributed training electromagnetic recognition model is adopted to disperse the electromagnetic signal sample data to multiple computing nodes for processing. The adaptive comprehensive loss function is used for multiple training optimizations. The imbalance constraint loss function and the cross entropy loss function are combined to aggregate and update parameters through the central node.
It improves training efficiency, recognition accuracy, and the generalization ability of the model, making it possible to quickly and accurately identify different types of electromagnetic signals and adapt to complex electromagnetic environments.
Smart Images
Figure CN120654113A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of radio signal processing, and in particular to an electromagnetic signal recognition method based on a distributed training electromagnetic recognition model. Background Art
[0002] With the widespread application of electromagnetic signals in many fields, accurately identifying electromagnetic signal types is crucial for ensuring information security, improving communication efficiency, and coping with complex electromagnetic environments. Traditional electromagnetic signal recognition methods often use centralized training models, centralizing all electromagnetic signal sample data on a single computer for processing and training. However, this approach faces numerous challenges when dealing with massive amounts of electromagnetic signal data. Centralized training requires powerful computing resources to process large datasets, resulting in high hardware costs and low computational efficiency, making it difficult to meet the high-real-time requirements of electromagnetic signal recognition scenarios.
[0003] Therefore, how to solve the computing resource bottleneck problem faced by existing electromagnetic signal recognition methods when processing large-scale electromagnetic signal data is a key technical problem that urgently needs to be overcome. Summary of the Invention
[0004] In view of the above analysis, an embodiment of the present invention aims to provide an electromagnetic signal recognition method based on a distributed training electromagnetic recognition model to solve the computing resource bottleneck problem faced when processing large-scale electromagnetic signal data.
[0005] The present invention discloses an electromagnetic signal recognition method based on a distributed training electromagnetic recognition model, the electromagnetic signal recognition method comprising:
[0006] Constructing an electromagnetic signal sample set and a distributed architecture; wherein the electromagnetic signal sample set includes key features and signal type labels of each historical electromagnetic signal; the distributed architecture includes multiple computing nodes and a central node, each computing node maintains an electromagnetic recognition model, and the central node is used to schedule the computing nodes;
[0007] Using a distributed architecture, the electromagnetic recognition model is trained based on the electromagnetic signal sample set to obtain a trained electromagnetic recognition model.
[0008] The key features of the electromagnetic signal to be identified are extracted and input into the trained electromagnetic recognition model, and the electromagnetic recognition model predicts and outputs the signal type of the electromagnetic signal to be identified.
[0009] On the basis of the above solution, the present invention also makes the following improvements:
[0010] Furthermore, the trained electromagnetic recognition model is obtained through the following operations:
[0011] During each round of training, each computing node uses the global training parameters sent by the central node to initialize the model parameters of the electromagnetic recognition model, and uses the adaptive comprehensive loss function to perform multiple training optimizations on the electromagnetic recognition model to obtain the local training parameters of the current round; the central node updates the global training parameters based on the local training parameters of the current round of each computing node and jumps to the next round of training; multiple rounds of training are repeated until the training end conditions are met, and a trained electromagnetic recognition model is obtained.
[0012] Furthermore, during each round of training, the computing node performs the following operations to obtain the local training parameters for the current round:
[0013] Each time a training is performed, a feature extraction vector of the electromagnetic recognition model for this training is obtained, and a softened vector for this training is obtained after selective softening;
[0014] According to the softening vector of the first training of the current round and the softening vector of this training, the imbalance constraint loss function of this training is obtained;
[0015] According to the imbalance constraint loss function and cross entropy loss function of this training, the adaptive comprehensive loss function of this training is obtained to update the model parameters of the next training;
[0016] Repeat the training for multiple times and use the model parameters updated in the last training of the current round as the local training parameters of the current round.
[0017] Furthermore, during each training process, the computing node selectively softens the feature extraction vector of this training according to the classification category set with an unbalanced sample size in the electromagnetic signal training sample set of the computing node to obtain a softened vector of this training.
[0018] Furthermore, the adaptive comprehensive loss function L is expressed as:
[0019] L=λL ce +(1-λ)L asc (1)
[0020] Among them, λ represents the weight coefficient, λ∈[0,1], L ce represents the cross entropy loss function, L asc represents the imbalance constraint loss function.
[0021] Furthermore, the imbalance constraint loss function L for the k-th training is asc (k) is expressed as:
[0022]
[0023] Where T represents the softening temperature, C imbRepresents a set of classification categories with unbalanced sample sizes in the electromagnetic signal training sample set of the computing node, Represents the jth element in the softened vector of the kth training; k ranges from 1 to K, where K represents the number of training times in each round of training.
[0024] further,
[0025]
[0026] Among them, p j (k) indicates that the feature extraction vector of the kth training is matched to C imb The jth element that needs to be softened.
[0027] Furthermore, the central node updates the global training parameters based on the local training parameters of each computing node in the current round, executing:
[0028] The central node uses a distributed aggregation strategy to aggregate the current round of local training parameters of each computing node to obtain the updated global training parameters.
[0029] Furthermore, each time a round of training is performed, the central node verifies the prediction accuracy of the electromagnetic recognition model using the updated global training parameters as model parameters on the electromagnetic signal verification sample set. If the prediction accuracy is better than the historical optimal accuracy, the corresponding global training parameters are stored as the global optimal parameters.
[0030] Furthermore, if the fluctuation of the prediction accuracy of multiple consecutive rounds of training is less than the preset deviation, or the preset number of training rounds is reached, the training end condition is met, and the electromagnetic recognition model with the global optimal parameters as the model parameters is used as the trained electromagnetic recognition model.
[0031] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:
[0032] The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model provided by the present invention has the following beneficial effects.
[0033] (1) Improve training efficiency: Through a distributed architecture, the training tasks of the electromagnetic recognition model are dispersed to multiple computing nodes, making full use of the computing resources of each node, greatly shortening the training time, and improving the training efficiency. It can quickly respond to the needs of electromagnetic signal recognition and adapt to application scenarios with high real-time requirements.
[0034] (2) Improve recognition accuracy: To address the problem of sample category imbalance, the computing node selectively softens the feature extraction vector during training, and constructs an adaptive comprehensive loss function by combining the imbalance constraint loss function and the cross entropy loss function. This effectively alleviates the impact of category imbalance on model training, enabling the model to more accurately identify different types of electromagnetic signals and improve recognition accuracy.
[0035] (3) Enhanced model generalization capability: During the distributed training process, each computing node independently trains and updates the model parameters, and the central node then updates the global parameters. This mechanism enables the model to learn the characteristics and patterns of data from different nodes, enhances the model's adaptability to different electromagnetic environments and signal characteristics, and improves the model's generalization capability, enabling it to more accurately identify new or unknown electromagnetic signals, thereby improving the overall performance and reliability of the electromagnetic signal recognition method.
[0036] In the present invention, the above-mentioned technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of the present invention will be described in the following description, and some advantages will become apparent from the description or be learned through practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. Throughout the drawings, the same reference symbols denote the same components.
[0038] Figure 1 A flowchart of an electromagnetic signal recognition method based on a distributed training electromagnetic recognition model provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0039] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.
[0040] A specific embodiment of the present invention discloses an electromagnetic signal recognition method based on a distributed training electromagnetic recognition model, the flow chart of which is as follows: Figure 1 As shown, the electromagnetic signal recognition method includes the following steps.
[0041] Step S1: Construct an electromagnetic signal sample set and a distributed architecture; wherein the electromagnetic signal sample set includes key features and signal type labels of each historical electromagnetic signal; the distributed architecture includes multiple computing nodes and a central node, each computing node maintains an electromagnetic recognition model, and the central node is used to schedule each computing node.
[0042] During the construction of the electromagnetic signal sample set, key features are extracted for each historical electromagnetic signal to obtain the corresponding key features. Preferably, depending on the electromagnetic signal, key features such as time domain features and frequency domain features can be extracted. Common time domain features include the amplitude characteristics of the electromagnetic signal, such as maximum, minimum, and average values. Furthermore, by converting the electromagnetic signal from the time domain to the frequency domain using methods such as Fourier transform, frequency distribution features of the electromagnetic signal can be extracted. For example, the signal's center frequency and bandwidth can be extracted.
[0043] In a distributed architecture, compute nodes, the core execution units, are numerous and widely distributed. Each compute node possesses a certain level of computing power and can independently run and maintain an electromagnetic identification model. They can be physical servers, virtual machines, or dedicated computing devices, with hardware resources such as processor performance, memory capacity, and storage space configured based on the application scenario and performance requirements. Compute nodes can be geographically dispersed, such as in data centers located in different locations, or logically dispersed, such as in different racks or server clusters within the same data center. The central node plays a key coordination and scheduling role in the entire architecture. It typically has high performance and stability requirements, as it handles information exchange and task allocation from multiple compute nodes. The hardware configuration of the central node also needs to be determined based on the number of compute nodes and the complexity of the tasks to ensure efficient scheduling. The central node can be a single server or a high-availability cluster consisting of multiple servers to improve system reliability and fault tolerance.
[0044] Step S2: Using a distributed architecture, the electromagnetic recognition model is subjected to distributed training based on the electromagnetic signal sample set to obtain a trained electromagnetic recognition model.
[0045] In this embodiment, the electromagnetic recognition model training process is performed independently on each computing node. Model training on different computing nodes may produce different results and performance due to factors such as data differences. Furthermore, since the models on each computing node are maintained independently, regular model updates and synchronization are required to ensure recognition performance and consistency across the distributed architecture. A central node can coordinate this process to implement model updates. Specifically, in step S2, the trained electromagnetic recognition model is obtained through the following operations.
[0046] Step S21: During each round of training, each computing node initializes the model parameters of the electromagnetic recognition model using the global training parameters sent by the central node, and uses the adaptive comprehensive loss function to perform multiple training optimizations on the electromagnetic recognition model to obtain the local training parameters of the current round.
[0047] Step S22: The central node updates the global training parameters based on the local training parameters of the current round of each computing node and jumps to the next round of training; multiple rounds of training are repeated until the training end condition is met to obtain a trained electromagnetic recognition model.
[0048] It should be noted that during the first round of training, the central node can pre-set global training parameters based on actual conditions and distribute them to each computing node. In subsequent rounds of training, the central node distributes the updated global training parameters from the previous round to each computing node, allowing each computing node to perform the next round of training accordingly. This means that the starting point of each training round is the combined results of the previous round's training for all computing nodes, allowing the training parameters of the electromagnetic intelligent model to be updated and optimized during each round of training.
[0049] Specifically, in each round of training in step S21 , the computing node performs the following operations to obtain the local training parameters of the current round.
[0050] Step S211: Each time a training is performed, a feature extraction vector of the electromagnetic recognition model for this training is obtained, and a softened vector for this training is obtained after selective softening.
[0051] Preferably, the electromagnetic recognition model used in this embodiment includes a feature extraction network component and a classification component, such as a convolutional neural network (CNN). Common convolutional neural networks include LeNet and ResNet (residual networks). During implementation, the feature extraction network component is used to obtain a feature extraction vector for each training run of the electromagnetic recognition model. The feature extraction vector for the kth training run is denoted as p(k), and its dimension equals the number of classification categories. Generally, a softmax layer is used as the classification component to obtain the probability that a training sample belongs to each classification category.
[0052] Specifically, during each training process, the computing node selectively softens the feature extraction vector of this training according to the classification category set with an unbalanced sample size in the electromagnetic signal training sample set of the computing node to obtain the softened vector of this training.
[0053]
[0054] Among them, C imb represents the classification category set with unbalanced sample size in the electromagnetic signal training sample set of the computing node; p j (k) indicates that the feature extraction vector of the kth training is matched to C imb The jth element that needs to be softened (i.e., filter out the elements that meet the set C from p(k) imb The jth element among all elements of ); represents the jth element in the softening vector of the kth training; k ranges from 1 to K, where K represents the number of training rounds; T represents the softening temperature.
[0055] For example, assuming that there are 10 categories in total and the labels of the categories are {0, 1, 2, ..., 9}, if the number of samples of categories 2 and 4 is small, that is, unbalanced, then C imb ={2,4}.
[0056] It should be noted that the dimension of the feature extraction vector of the electromagnetic recognition model is equal to the number of classification categories. During implementation, considering that the electromagnetic signal training sample sets allocated to computing nodes may have an imbalance in sample size, which can easily lead to low classification accuracy of such samples by the electromagnetic recognition model, this embodiment selectively softens the feature extraction vectors of this training to address the performance bottleneck caused by the inconsistent allocation of sample sizes across different categories during electromagnetic recognition model training.
[0057] Step S212: Obtain the imbalance constraint loss function of this training according to the softening vector of the first training of the current round and the softening vector of this training.
[0058] The imbalance constraint loss function L for the kth training asc (k) is expressed as:
[0059]
[0060] When k is 1, Therefore, the imbalance constraint loss function L for the first training is asc (1) Based on Calculated. When k is 1 to K, Therefore, the imbalance constraint loss function L for the k-th training is asc (k) Based on and It should be emphasized that the model parameters of the electromagnetic recognition model are initialized using the global training parameters issued by the central node, and the first training is performed. Therefore, is directly related to the global training parameters; and (k ranges from 1 to K) The electromagnetic recognition model is trained based on the training parameters obtained in the previous training, and is directly related to the training parameters of the previous training.
[0061] Step S213: Based on the imbalance constraint loss function and cross entropy loss function of this training, an adaptive comprehensive loss function of this training is obtained to update the model parameters of the next training.
[0062] The adaptive comprehensive loss function L is expressed as:
[0063] L=λL ce +(1-λ)L asc (1)
[0064] Among them, λ represents the weight coefficient, λ∈[0,1], L ce represents the cross entropy loss function, L asc Represents the imbalance constraint loss function. From the above, we can see that the cross entropy loss function L ce Labels are required; and the imbalance constraint loss function L asc No labels are required, and constraints are based on the training output of the electromagnetic recognition model.
[0065] Step S214: Repeat the training for multiple times, and use the model parameters updated by the last training in the current round as the local training parameters in the current round.
[0066] In step S22, the central node updates the global training parameters based on the current round local training parameters of each computing node, and executes: the central node aggregates the current round local training parameters of each computing node using a distributed aggregation strategy to obtain updated global training parameters.
[0067] Preferably, the central node can use a distributed aggregation strategy of weighted average aggregation to aggregate the current-round local training parameters of each computing node. Specifically, the central node verifies the classification accuracy of the electromagnetic recognition model using the current-round local parameters of each computing node as model parameters on the electromagnetic signal verification set, determines the weighted ratio of the current-round local training parameters of each computing node based on the classification accuracy, and then performs weighted average aggregation on the current-round local training parameters of each computing node based on the determined weighted ratio to obtain the updated global training parameters.
[0068] To achieve better training results, the central node also updates the electromagnetic signal sample set of each computing node after each training round and distributes the updated electromagnetic signal sample set and updated global training parameters to each computing node. For example, the processing capacity of each computing node can be evaluated based on its hardware configuration (CPU, memory, GPU, etc.) and historical performance data (processing time and efficiency of each computing node in previous training rounds) to calculate and update the sample size processed by each computing node, thereby updating the electromagnetic signal sample set of each computing node. During specific implementation, the corresponding evaluation criteria can be selected based on the specific circumstances of the distributed architecture, and this embodiment does not limit this.
[0069] In addition, after each round of training, the central node verifies the prediction accuracy of the electromagnetic recognition model using the updated global training parameters as model parameters on the electromagnetic signal verification sample set. If the prediction accuracy is better than the historical optimal accuracy, the corresponding global training parameters are stored as the global optimal parameters. If the fluctuation of the prediction accuracy over multiple consecutive rounds of training is less than the preset deviation, or if the preset number of training rounds is reached, the training end condition is met, and the electromagnetic recognition model using the global optimal parameters as model parameters is regarded as the trained electromagnetic recognition model.
[0070] Step S3: extract key features of the electromagnetic signal to be identified and input them into the trained electromagnetic recognition model, and the electromagnetic recognition model predicts and outputs the signal type of the electromagnetic signal to be identified.
[0071] During the specific implementation process, the key features of the electromagnetic signal to be identified are extracted in the same processing method as the historical electromagnetic signal in step S1, and the key features are input into the trained electromagnetic recognition model. After that, the trained electromagnetic recognition model can predict and output the signal type of the electromagnetic signal to be identified.
[0072] In summary, the electromagnetic signal recognition method based on distributed training of an electromagnetic recognition model provided in this embodiment implements an electromagnetic signal recognition model based on distributed training by constructing an electromagnetic signal sample set and a distributed architecture. In this distributed architecture, multiple computing nodes work together, each maintaining an electromagnetic recognition model, and are scheduled by a central node. This architecture fully leverages the advantages of distributed computing, distributing electromagnetic signal sample data across various computing nodes for processing. This significantly reduces the computational burden on individual nodes, significantly improves the efficiency of model training, and can quickly respond to the training needs of large-scale electromagnetic signal data.
[0073] During the training process, each computing node uses an adaptive comprehensive loss function to perform multiple training optimizations on the electromagnetic recognition model. This adaptive comprehensive loss function comprehensively considers the imbalance constraint loss function and the cross-entropy loss function. The imbalance constraint loss function is calculated based on the softened vectors from each training session, which in turn are obtained by selectively softening the feature extraction vectors. This selective softening operation fully accounts for the imbalanced classification set of sample sizes in the electromagnetic signal training sample set of the computing node. It can effectively alleviate the impact of category imbalance on model training, allowing the model to pay more attention to the signal characteristics of minority categories during the learning process, thereby improving the model's overall recognition accuracy for various electromagnetic signals.
[0074] Ultimately, the electromagnetic recognition model obtained after multiple rounds of distributed training can accurately predict and output the signal type after extracting the key features of the electromagnetic signal to be identified, providing an efficient and reliable solution for the rapid and accurate identification of electromagnetic signals, with good application prospects and practical value.
[0075] Those skilled in the art will appreciate that all or part of the process steps of the above-described embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a magnetic disk, an optical disk, a read-only memory, or a random access memory.
[0076] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.
Claims
1. A method for electromagnetic signal recognition based on a distributed training electromagnetic recognition model, characterized in that: The electromagnetic signal recognition method comprises: Constructing an electromagnetic signal sample set and a distributed architecture; wherein the electromagnetic signal sample set includes key features and signal type labels of each historical electromagnetic signal; the distributed architecture includes multiple computing nodes and a central node, each computing node maintains an electromagnetic recognition model, and the central node is used to schedule the computing nodes; Using a distributed architecture, the electromagnetic recognition model is trained based on the electromagnetic signal sample set to obtain a trained electromagnetic recognition model. The key features of the electromagnetic signal to be identified are extracted and input into the trained electromagnetic recognition model, and the electromagnetic recognition model predicts and outputs the signal type of the electromagnetic signal to be identified.
2. The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model according to claim 1, characterized in that: The trained electromagnetic recognition model is obtained through the following operations: During each round of training, each computing node uses the global training parameters sent by the central node to initialize the model parameters of the electromagnetic recognition model, and uses the adaptive comprehensive loss function to perform multiple training optimizations on the electromagnetic recognition model to obtain the local training parameters of the current round; the central node updates the global training parameters based on the local training parameters of the current round of each computing node and jumps to the next round of training; multiple rounds of training are repeated until the training end conditions are met, and a trained electromagnetic recognition model is obtained.
3. The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model according to claim 2, characterized in that: During each round of training, the computing node performs the following operations to obtain the local training parameters for the current round: Each time a training is performed, a feature extraction vector of the electromagnetic recognition model for this training is obtained, and a softened vector for this training is obtained after selective softening; According to the softening vector of the first training of the current round and the softening vector of this training, the imbalance constraint loss function of this training is obtained; According to the imbalance constraint loss function and cross entropy loss function of this training, the adaptive comprehensive loss function of this training is obtained to update the model parameters of the next training; Repeat the training for multiple times and use the model parameters updated in the last training of the current round as the local training parameters of the current round.
4. The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model according to claim 3 is characterized in that: During each training process, the computing node selectively softens the feature extraction vector of this training according to the classification category set with unbalanced sample size in the electromagnetic signal training sample set of the computing node to obtain the softened vector of this training.
5. The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model according to claim 4 is characterized in that: The adaptive comprehensive loss function L is expressed as: L=λL ce +(1-λ)L asc (1) Among them, λ represents the weight coefficient, λ∈[0,1], L ce represents the cross entropy loss function, L asc represents the imbalance constraint loss function.
6. The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model according to claim 5, characterized in that: The imbalance constraint loss function L for the kth training asc (k) is expressed as: Where T represents the softening temperature, C imb Represents a set of classification categories with unbalanced sample sizes in the electromagnetic signal training sample set of the computing node, Represents the jth element in the softened vector of the kth training; k ranges from 1 to K, where K represents the number of training times in each round of training.
7. The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model according to claim 6, characterized in that: Among them, p j (k) indicates that the feature extraction vector of the kth training is matched to C imb The jth element that needs to be softened.
8. The electromagnetic signal recognition method based on a distributed training electromagnetic recognition model according to any one of claims 2 to 7, characterized in that: The central node updates the global training parameters based on the local training parameters of each computing node in the current round, executing: The central node uses a distributed aggregation strategy to aggregate the current round of local training parameters of each computing node to obtain the updated global training parameters.
9. The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model according to claim 8, characterized in that: After each round of training, the central node verifies the prediction accuracy of the electromagnetic recognition model using the updated global training parameters as model parameters on the electromagnetic signal verification sample set. If the prediction accuracy is better than the historical optimal accuracy, the corresponding global training parameters will be stored as the global optimal parameters.
10. The electromagnetic signal recognition method based on the distributed training electromagnetic recognition model according to claim 9, characterized in that: If the fluctuation of the prediction accuracy of multiple consecutive rounds of training is less than the preset deviation, or the preset number of training rounds is reached, the training end condition is met, and the electromagnetic recognition model with the global optimal parameters as the model parameters is used as the trained electromagnetic recognition model.