Radiation source identification method, related device, equipment and storage medium

Abstract

The invention discloses a radiation source identification method, a related device, equipment and a storage medium, and the method comprises the steps: obtaining a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown equipment category; identifying the target radiation signal based on a target identification model to obtain a target equipment category of the target radiation source; wherein the target identification model takes the first equipment category of the sample radiation signal as a training target, and is obtained by training based on the sample signal features of the sample radiation signal and representative signal features of a plurality of sample equipment categories. According to the scheme, the accuracy of radiation source identification can be improved, and particularly the accuracy of radiation source identification in an unknown open set scene can be improved.

Description

Technical Field

[0001] The present application relates to the field of signal recognition technology, and in particular to a radiation source recognition method and related devices, equipment and storage media. Background Art

[0002] Radiation source identification refers to the process of distinguishing different radiation sources from a complex electromagnetic environment by utilizing the inherent properties of the radiation source through specific technical means.

[0003] At present, deep learning neural networks are used to extract the signal features of the radiation source emission signal, and by analyzing these signal features, the radiation source can be identified and the categories of different radiation sources can be distinguished. However, the training data of the neural network mainly comes from closed-set scenarios of known radiation sources. When facing open-set scenarios from unknown radiation sources, the neural network cannot adapt well to new unknown data, and the accuracy of radiation source identification is poor.

[0004] Therefore, how to improve the accuracy of radiation source identification, especially the accuracy of radiation source identification in unknown open-set scenarios, has become a problem that needs to be solved urgently. Summary of the invention

[0005] The main technical problem solved by the present application is to provide a radiation source identification method and related devices, equipment and storage media, which can improve the accuracy of radiation source identification, especially the accuracy of radiation source identification in unknown open set scenarios.

[0006] In order to solve the above technical problems, the first aspect of the present application provides a radiation source identification method, which includes: obtaining a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; identifying the target radiation signal based on a target recognition model to obtain the target device category of the target radiation source; wherein the target recognition model takes the first device category of the sample radiation signal as a training target, and is trained based on the sample signal features of the sample radiation signal and the representative signal features of several sample device categories, the first device category is determined by clustering each sample radiation signal, the sample signal features include feature information obtained by clustering the sample radiation signal, the sample radiation signal comes from the sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category.

[0007] In order to solve the above-mentioned technical problems, the second aspect of the present application provides a radiation source identification device, which includes an acquisition module and an identification module; the acquisition module is used to acquire a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; the identification module is used to identify the target radiation signal based on a target identification model to obtain the target device category of the target radiation source; wherein the target identification model takes the first device category of the sample radiation signal as the training target, and is trained based on the sample signal features of the sample radiation signal and the representative signal features of several sample device categories, the first device category is determined by clustering each sample radiation signal, the sample signal features include feature information obtained by clustering the sample radiation signal, the sample radiation signal comes from the sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category.

[0008] In order to solve the above technical problems, the third aspect of the present application provides an electronic device, including a memory and a processor coupled to each other, and the processor is used to execute program instructions stored in the memory to implement the radiation source identification method in the above first aspect.

[0009] In order to solve the above technical problem, the fourth aspect of the present application provides a computer-readable storage medium on which program instructions are stored. When the program instructions are executed by a processor, the radiation source identification method in the above first aspect is implemented.

[0010] The above scheme obtains a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; and identifies the target radiation signal based on a target recognition model to obtain a target device category of the target radiation source; wherein the target recognition model uses the first device category of the sample radiation signal as a training target, and is trained based on the sample signal features of the sample radiation signal and representative signal features of several sample device categories, wherein the first device category is determined by clustering the sample radiation signals, the sample signal features include feature information obtained by clustering the sample radiation signals, the sample radiation signal comes from the sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category. Therefore, by training the target recognition model through the sample signal features of the sample radiation signal and the representative signal features of various sample device categories, the target recognition model can simultaneously learn the category feature information of the first device category and the second device category, and capture the relationship between the sample signal features and the first device category and the second device category, which is beneficial to improving the adaptability of the target recognition model to sample device categories including the first device category and the second device category. Further, the target radiation signal can be identified through the target recognition model, and the target radiation signal initially identified as an unknown device category can be further identified, which can improve the accuracy of radiation source identification, especially the accuracy of radiation source identification in unknown open set scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a flow chart of an embodiment of a radiation source identification method of the present application; Figure 2 This is a schematic diagram of the training process of the target recognition model of this application; Figure 3 This is a schematic diagram of the framework of an embodiment of a radiation source identification device of the present application; Figure 4 A schematic diagram of a framework of an embodiment of an electronic device of the present application; Figure 5 A schematic diagram of a framework of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0012] The scheme of the embodiment of the present application is described in detail below in conjunction with the drawings of the specification.

[0013] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures, interfaces, and technologies are provided to facilitate a thorough understanding of the present application.

[0014] The term "multiple" herein means two or more than two. In addition, the term "at least one" herein means any combination of at least two of any one or more of a plurality of, for example, including at least one of A, B, and C, and may mean including any one or more elements selected from the set consisting of A, B, and C.

[0015] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of the radiation source identification method of the present application, which specifically includes the following steps: Step S11: Acquire a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category.

[0016] In one implementation scenario, a radiation source refers to a device used to emit ionizing radiation signals, which may include but is not limited to: radar, communication radio station, etc.

[0017] In one implementation scenario, the radiation signal emitted by the radiation source can be initially identified as an unknown device category, or can be initially identified as a known device category, and the radiation signal initially identified as an unknown device category is used as a target radiation signal, and the radiation source that emits the target radiation signal is used as a target radiation source. The number of target radiation signals can include one or more, and the multiple target radiation signals can come from the same target radiation source or from different target radiation sources. In addition, the known device category can further include several second device categories.

[0018] In a specific implementation scenario, radiation signals that cannot be identified as known device categories can be classified as unknown device categories, and radiation signals that are initially identified as unknown device categories can be saved in an unknown category library, which can be a storage medium such as a hard disk, and then a target number of target radiation signals can be obtained from the unknown category library according to the scenario requirements. Of course, the target radiation signal can also be collected in real time (e.g., received by a receiver), and the acquisition channel of the target radiation signal is not limited here.

[0019] In a specific implementation scenario, after acquiring the radiation signal emitted by the radiation source, the radiation signal can be directly analyzed to initially identify the type of device from which the radiation signal comes, or a specifically trained neural network can be used to identify the radiation signal to initially identify the type of device from which the radiation signal comes. The method for initially identifying the radiation signal is not specifically limited herein. Among them, the neural network used for initially identifying the radiation signal may include, but is not limited to: Resnet network, Inception network, etc. As a possible implementation method, the neural network used for initially identifying the radiation signal can be used as a basic model for training the target recognition model in subsequent embodiments.

[0020] Step S12: Identify the target radiation signal based on the target recognition model to obtain the target device category of the target radiation source.

[0021] In an embodiment of the present disclosure, a target recognition model takes a first device category of a sample radiation signal as a training target, and is trained based on sample signal features of the sample radiation signal and representative signal features of several sample device categories. The first device category is determined by clustering the sample radiation signals, the sample signal features include feature information obtained by clustering the sample radiation signals, the sample radiation signal comes from a sample radiation source, and the sample radiation signal is initially identified as an unknown device category. The several sample device categories include a first device category and at least one second device category representing a known device category.

[0022] It should be noted that the sample radiation signal is obtained in a similar manner to the target radiation signal in the aforementioned embodiment, which will not be described in detail here, and the radiation source that emits the sample radiation signal is taken as the sample radiation source.

[0023] In one implementation scenario, in order to improve the data processing efficiency of the target recognition model, before the target recognition model recognizes the target radiation signal, several electromagnetic signal features of the target radiation signal can be obtained, and the phases of various electromagnetic signal features can be adjusted respectively to obtain the phase offset features corresponding to the various electromagnetic signal features, and then the phase offset features corresponding to the several electromagnetic signal features can be fused to obtain the electromagnetic fusion features of the target radiation signal, and then the electromagnetic fusion features can be identified by the target recognition model to obtain the target device category of the target radiation source. Among them, there are at least two electromagnetic signal features with opposite phase offset directions. Therefore, by identifying the electromagnetic fusion features of the target radiation signal through the target recognition model, on the one hand, the signal feature information of the target radiation signal in different dimensions can be obtained, and on the other hand, by performing phase offset processing on at least two electromagnetic signal features of the target radiation signal in opposite directions to increase the distance between different types of electromagnetic signal features of the same target radiation signal, the signal feature information of different dimensions can be distinguished as clearly as possible, and the data validity can be improved by reducing the interference between the feature information of different dimensions, thereby improving the data processing efficiency of the target recognition model.

[0024] In a specific implementation scenario, several electromagnetic signal features can be obtained by extracting electromagnetic features from the radiation signal, and the types of electromagnetic signal features may include but are not limited to: time domain features, frequency domain features, power spectrum features, time-frequency spectrum features, wavelet transform features, etc. The extraction of electromagnetic signal features can refer to Fourier transform algorithm, fast Fourier transform algorithm, short-time Fourier transform algorithm, wavelet transform algorithm and other related technologies, and the extraction methods of different types of electromagnetic signal features will not be repeated here.

[0025] In a specific implementation scenario, the phase shift degrees of various electromagnetic signal features for phase adjustment may be the same or different, and the phase shift degrees may be obtained from the number of types of electromagnetic signal features. As a possible implementation, since the value of sin90 degrees is 1 and it is an orthogonal basis in complex numbers, 90 degrees can be decomposed into a corresponding number of degree components according to the number of types of electromagnetic signal features, and each degree component is used as the phase shift degree of various electromagnetic signal features. As a possible example, the number of types of electromagnetic signal features includes 5 types, namely: a, b, c, d, e, and the phase shift degrees of various electromagnetic signal features are the same. By dividing the orthogonal basis 90 degrees by the number of types 5, it can be obtained that the phase shift degrees of various electromagnetic signal features are 18 degrees. The five electromagnetic signal features are divided into 3 groups (such as a and b as the first group, c and d as the second group, and e alone as the third group). In the first group, a and b are phase adjusted with the same phase shift degrees and opposite phase shift directions, and the phase shift characteristics of a can be obtained. , b) Phase shift characteristics In the second group, phase adjustments are performed on c and d with the same phase shift degree and opposite phase shift direction, and the phase shift characteristic of c can be obtained. , d phase shift characteristics , in the third group, the phase of e is adjusted to obtain the phase shift characteristic of e . Therefore, the electromagnetic fusion feature can be expressed as: .

[0026] In one implementation scenario, in order to obtain the sample signal features of the sample radiation signal, the sample electromagnetic fusion features can be obtained by performing electromagnetic feature extraction on the sample radiation signal, and the sample electromagnetic fusion features of several sample radiation signals are clustered to obtain several first device categories, and the clustered sample electromagnetic fusion features are used as the sample signal features. Of course, the signal features of the sample radiation signal after clustering can also be directly obtained as the sample signal features.

[0027] In one implementation scenario, in order to obtain representative signal features of a sample device category, a signal feature that is closest to the feature mean of each signal feature in the sample device category can be selected as the representative signal feature, or a representative signal feature can be generated based on the feature mean of each signal feature in the sample device category. The method for obtaining the representative signal feature is not specifically limited herein.

[0028] In one implementation scenario, in order to train a target recognition model, the first predicted device category of the sample radiation source can be obtained through the sample signal features and the representative signal features, and the network parameters of the target recognition model can be adjusted through the difference between the first predicted device category and the first device category. Therefore, the first device category is used as the training target of the target recognition model, and on the basis that the target recognition model can recognize radiation sources of known device categories, the device categories that the target recognition model can recognize are increased, thereby improving the compatibility of the target recognition model in identifying radiation sources.

[0029] In a specific implementation scenario, in order to obtain the first predicted device category of the sample radiation source, graph structure data can be constructed based on the sample signal features and the representative signal features, and the sample fusion features of each node in the graph structure data are obtained, and then the node representing the sample signal features is selected as the target node, and the first predicted device category of the sample radiation source to which the sample signal features represented by the target node belong is determined based on the feature similarity between the sample fusion features of the target node and the sample fusion features corresponding to the node representing the representative signal features. Among them, each node in the graph structure data represents a different signal feature, and the sample fusion feature of the node at least includes the signal feature represented by the node and the feature information of other nodes connected to the node. Therefore, using the graph structure data, the signal features and the complex relationship between the signal features can be intuitively reflected, which helps to improve the efficiency of obtaining the sample fusion features of each node, and on the basis of the graph structure data, according to the feature relationship between the sample fusion features of each node, the first predicted device category of the sample radiation source is determined, which can improve the accuracy of obtaining the first predicted device category.

[0030] In a specific implementation scenario, on the basis of constructing graph structure data through sample signal features and representative signal features, in order to further obtain the sample fusion features of each node in the graph structure data, the graph structure data can be processed through the graph attention mechanism to obtain the aggregate features of the nodes in the graph structure data, and based on the signal features and aggregate features corresponding to the same node, the sample fusion features of the corresponding node are fused. Among them, the aggregate feature of a certain node represents the fusion information of signal feature fusion under the weight relationship between this node and other nodes. Since the weight relationship between each node and other nodes is different, there are also differences between the aggregate features of each node, which can be targeted to learn the different feature information of each node and other nodes, and improve the flexibility of each node to learn the feature information of other nodes. Therefore, on the basis of obtaining the aggregate feature of the node, based on the signal features and aggregate features corresponding to the same node, the sample fusion features of the corresponding node are fused as the new signal features represented by the node. According to the feature relationship between the new signal features represented by each node, the first predicted device category can be obtained, and the fusion learning of the category feature information between the first device category and the second device category can be promoted, and the compatibility of the target recognition model with the first device category and the second device category can be improved.

[0031] Exemplarily, the signal features corresponding to each node in the graph structure data are respectively represented as , to determine the nodes in the graph structure data As an example, through the node Signal characteristics Feature similarity with signal features of other nodes

[0032]

[0033] And through the softmax function Perform normalization to obtain the relationship between other nodes and node The attention weight :

[0034] The signal characteristics of each node are linearly transformed through the weight matrix, so that the signal characteristics of each node are linearly transformed according to the other nodes. The attention weight And the signal features after linear transformation of other nodes are fused to obtain the node Aggregation features:

[0035] in, represents the signal feature after linear transformation, and U is the weight matrix. The matrix form of the weight matrix can be Jordan standard form, etc. After obtaining the aggregate features of the node, the sample fusion features of the corresponding node are obtained by fusion based on the signal features and aggregate features corresponding to the same node. , such as node Sample fusion features It can be expressed as:

[0036] In a specific implementation scenario, based on the aforementioned construction of graph structure data through sample signal features and representative signal features, in order to further improve the accuracy of the first predicted device category, after selecting the target node representing the sample signal feature as described above, the sample fusion feature corresponding to the target node can be selected as the sample target feature, and the sample fusion feature other than the sample target feature can be selected as the sample candidate feature, and then the sample candidate feature whose feature similarity with the sample target feature meets the similarity requirement is selected as the sample matching feature, and finally the sample device category to which the sample matching feature belongs is used as the first predicted device category. Therefore, by selecting the sample candidate feature whose feature similarity with the sample target feature meets the similarity requirement as the sample matching feature, and using the sample device category to which the sample matching feature belongs as the first predicted device category, the device category that best matches the feature information contained in the sample signal feature can be selected in the sample device category, thereby improving the prediction accuracy of the first predicted device category.

[0037] For example, the sample fusion features of each node are obtained as described above. Based on the selection of nodes Sample fusion features As the sample target feature, by comparing The feature similarity between the sample fusion features (i.e., sample candidate features) of other nodes :

[0038] Select the feature similarity between the sample target feature and the Sample candidate features that meet the similarity requirements are used as sample matching features , and match the samples to features The sample device category is used as the first predicted device category. As a possible implementation, the sample matching feature The sample device category is the second device category, and the network parameters of the target recognition model are adjusted by the difference between the first predicted device category (the second device category) and the first device category. As another possible implementation, the first device category includes multiple devices, and the first device category to which the sample radiation signal belongs is used as the target device category, that is, the target recognition model uses the target device category as the training target, and uses the first device category other than the first device category to which the sample radiation signal belongs as the reference device category. When the category is the reference device category, the network parameters of the target recognition model are adjusted by the difference between the first predicted device category (reference device category) and the first device category (target device category). Therefore, by training the first predicted device category to be close to the first device category as the training target, the target recognition model can be gradually trained to have the ability to recognize radiation signals from radiation sources under the first device category, and by using multiple different first device categories as training targets to increase the types of device categories that the target recognition model can recognize, the accuracy of radiation source recognition can be improved, especially in scenarios with rich radiation signal sources.

[0039] It should be noted that the above feature similarity , feature similarity The angle between features (such as cosine similarity) is used as an example. It can also be measured by feature distance (such as Euclidean distance). The process of obtaining feature similarity is not specifically limited here, and feature similarity , feature similarity The same similarity measurement algorithm may be used, or different similarity measurement algorithms may be used. In addition, a preset angle threshold or a preset distance threshold of feature similarity may be used as a similarity requirement, and then by comparing the feature similarity with the preset angle threshold or the preset distance threshold, it is determined whether the similarity requirement is met. As a possible example, feature similarity The cosine similarity algorithm is used to measure the feature similarity. Compared with the preset angle threshold (such as 90 degrees, 95 degrees, etc.), that is, by judging the sample target characteristics With each sample candidate feature Whether the feature angle between them is greater than 90 degrees, and then judging whether the similarity requirement is met.

[0040] In a specific implementation scenario, the sample radiation signals include multiple ones, and each sample radiation signal corresponds to a sample signal feature. Based on the aforementioned construction of graph structure data based on the sample signal features and the representative signal features, and obtaining the sample fusion features of each node in the graph structure data, the sample fusion features corresponding to the nodes representing the sample signal features can be selected in turn as sample target features, and the sample fusion features other than the sample target features can be used as sample candidate features. In this case, the sample candidate features include the representative signal features and the sample fusion features corresponding to other sample signal features other than the sample target features, respectively. Then, by selecting the sample candidate features whose feature similarity with the sample target features meets the similarity requirements as sample matching features, and taking the sample device category to which the sample matching features belong as the first predicted device category, the first predicted device categories corresponding to the multiple sample radiation signals are obtained.

[0041] In another specific implementation scenario, unlike the aforementioned construction of graph structure data based on sample signal features and representative signal features, and then obtaining sample fusion features of each node in the graph structure data, self-attention mechanisms, multi-head attention mechanisms, etc. can also be used to directly process sample signal features and representative signal features to obtain sample fusion features corresponding to each signal feature, and based on the feature similarity between the sample fusion features corresponding to the sample signal features and the sample fusion features corresponding to the representative signal features, the first prediction device category of the sample radiation source to which the sample signal features belong is determined. Among them, mechanisms such as self-attention mechanisms and multi-head attention mechanisms assign different weights by calculating the correlation between elements to achieve effective integration of information, thereby also obtaining sample fusion features corresponding to each signal feature. The method for obtaining sample fusion features is not specifically limited here. The process of obtaining sample fusion features through attention mechanisms, multi-head attention mechanisms, etc. can refer to related technologies, and will not be described in detail here.

[0042] In an implementation scenario, after a target recognition model is obtained by training based on sample signal features and representative signal features, the sample signal features and representative signal features are processed based on the target recognition model to obtain a second predicted device category of the sample radiation source, wherein the process of obtaining the second predicted device category is similar to the process of obtaining the first predicted device, and will not be repeated here, that is, the first predicted device category and the second predicted device category are predicted by the target recognition model at different stages, the first predicted device category and the second predicted device category may be the same or different, and the second predicted device category is verified according to the real device category of the sample radiation signal to obtain a verification result, wherein the verification result includes: whether the second predicted device category is consistent with the real device category, and the real device category is obtained by the labeler by labeling the sample radiation signal based on experience, so that according to the verification result, the network parameters of the target recognition model are adjusted based on the difference between the second predicted device category and the real device category, or the second predicted device category is added as a new second device category. Therefore, after obtaining the target recognition model by training with the first device category as the training target, the trained target recognition model is tested, and the target recognition model is continued to be trained according to the test results, or the second predicted device category is saved as a new second device category, which can improve the accuracy and flexibility of the target recognition model training.

[0043] In a specific implementation scenario, when the verification result includes that the second predicted device category is consistent with the actual device category, the second predicted device category is added as a new second device category. In addition, the new second device category can also be saved in the known category library.

[0044] In a specific implementation scenario, when the verification result includes that the second predicted device category is inconsistent with the real device category, the target recognition model continues to be trained. Specifically, the real device category of the sample radiation signal is used as the training target of the target recognition model, and based on the difference between the second predicted device category and the real device category, the network parameters of the target recognition model are adjusted.

[0045] In a specific implementation scenario, the first device category of the sample radiation signal is predicted by the clustering model. When the verification result includes that the second predicted device category is inconsistent with the real device category, the network parameters of the clustering model can be adjusted according to the difference between the second predicted device category and the real device category. Therefore, since the second predicted device category is predicted by the target recognition model trained with the first device category as the training target, the difference between the second predicted device category and the real device category can reflect the difference between the first device category and the real device category, so that according to the difference between the second predicted device category and the real device category, the network parameters of the clustering model are adjusted, which can promote the accuracy of the clustering model in predicting the first device category of the sample radiation signal.

[0046] In a specific implementation scenario, the clustering model for obtaining the first device category includes at least a graph autoencoder network and a clustering network, wherein the graph autoencoder network is used to extract features of the sample radiation signal, and the clustering network is used to cluster the features extracted by the graph autoencoder network. The graph autoencoder includes a graph embedding layer, an encoder layer, and a decoder layer, and multiple input features of the graph autoencoder constitute input graph data, which is similar to the aforementioned graph structure data, and each node in the input graph data represents the signal features of the sample radiation signal before clustering. Among them, the graph embedding layer is used to map the nodes in the input graph data and the edges representing other nodes connected to the nodes into a low-dimensional space to obtain corresponding vector representations, and the node vector representation and the edge vector representation can be learned through different neural network layers respectively, or they can share parameters; the encoder layer is used to further process these vector representations, and learn to obtain node embedding features that capture the topological structure of the input graph data, and the encoder can be one or more fully connected layers, convolutional layers, or other types of neural network layers, depending on the structure of the graph and task requirements. In addition, the encoder can combine the attention mechanism to model the relationship between nodes, such as learning adjacent weights, so as to better represent the influence of unit nodes on the overall topological network; the decoder layer is used to reconstruct the structure of the input graph according to the node embedding features output by the encoder, and the decoder can adopt a neural network structure similar to that of the encoder but in the opposite direction, that is, the decoder is used to generate a high-dimensional graph representation from a low-dimensional node embedding, and the output of the decoder can be the coordinates of the node relocation, the reconstruction weight or probability of the edge, etc. Therefore, by capturing the features of the nodes and their structural positions in the graph and aggregating the information of neighboring nodes to learn and obtain the effective representation of each node (i.e., the node embedding features), the node embedding features output by the encoder in the graph autoencoder are used as the input data of the clustering network, and the clustering results output by the clustering network are obtained. Therefore, clustering the sample radiation signal through the clustering model to obtain the first device category can improve the efficiency of obtaining the first device category. In addition, the clustering model can also include a convolutional network and an output representation layer, the convolutional network is used to extract the features of the sample radiation signal, and the features of the sample radiation signal extracted by the convolutional network are used as the input features of the graph autoencoder, and the output representation layer is used to output the sample signal features containing the feature information obtained by clustering the sample radiation signal.

[0047] In a specific implementation scenario, when there are multiple real device categories, the inconsistency between the second predicted device category and the real device category includes: the number of types of the second predicted device category is inconsistent with the number of types of the real device category, or at least one of the second predicted device categories is inconsistent with its corresponding real device category. Therefore, the second predicted device category is verified from two levels: the difference in the number of types and the difference in type characteristics between the second predicted device category and the real device category, and different compensation strategies can be adopted according to different types of differences, thereby improving the flexibility of training the target recognition model or clustering model.

[0048] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the training process of the target recognition model for this application. Figure 2 The method includes an initial recognition module, a clustering model, an unknown category library, a known category library and a target recognition model, wherein the initial recognition module can be a basic model for training the target recognition model, and the acquired radiation signal is initially recognized by the initial recognition module. When the radiation signal is initially recognized as the second device category, the radiation signal initially recognized as the second device category is saved in the known category library, and the known category library includes the second device category (such as "known 1", "known 2", "known 3", "known 4"); when the radiation signal is initially recognized as the unknown device category, the radiation signal initially recognized as the unknown device category is saved in the unknown category library, and after the radiation signals in the unknown category library are accumulated to a certain number, the radiation signal is obtained from the unknown category library as a sample radiation signal, and several sample radiation signals are sent to the clustering model for clustering to obtain several first device categories (such as "unknown 1", "unknown 2", "unknown 3", "unknown 4", "unknown 5") of the several sample radiation signals. With the first device category as the training target, the basic model is trained as a target recognition model according to the sample signal features of the sample radiation signal and the representative signal features of several sample device categories. The trained target recognition model is used to identify the sample radiation signal again to obtain a second predicted device category ("known 5", "known 6", "known 7", "known 8", "known 9"). The second predicted device category is verified by the real device category annotated by the annotator. If the second predicted device category is consistent with the real device category, the second predicted device category is saved as a new second device category. If the second predicted device category is inconsistent with the real device category, the target recognition model and / or clustering model is trained with the real device category as the training target.

[0049] For example, see Figure 2, the second predicted device categories include "known 5", "known 6", "known 7", "known 8", and "known 9". When the real device categories include "known 5", "known 6", "known 7", "known 8", "known 9", and "known 10", the types of the second predicted device categories are less than the types of the real device categories, and it is judged as a case of "few clusters". Or, when the real device categories include "known 5", "known 6", "known 7", and "known 8", the types of the second predicted device categories are more than the types of the real device categories, and it is judged as a case of "multiple clusters". The network parameters of the clustering model can be adjusted according to the number of types of the real device categories, so that the clustering model can predict the first device category that is consistent with the number of types of the real device categories; for example, when the real device categories include "known 10", "known 6", "known 7", "known 8", and "known 9", among which the second predicted device category "known 5" is different from the real device category "known 10", it is judged as a case of "partial clustering error". According to the difference between the second predicted device category "known 5" and the real device category "known 10", the network parameters of the clustering model or the target recognition model can be adjusted.

[0050] Please continue reading Figure 2 After the target recognition model is trained, the target recognition model and / or clustering model can also be trained based on the radiation signals initially identified as known device categories and the radiation signals initially identified as unknown device categories as training data to enhance the compatibility of the target recognition model or clustering model with the new second device category and the original second device category.

[0051] In one implementation scenario, the difference measure between the first device category and the first predicted device category or the difference measure between the real device category and the second predicted device category can be implemented through loss functions such as cross entropy loss and arcface. The specific process of adjusting the network parameters of the target recognition model or clustering model can refer to related technologies such as gradient descent that use training loss to adjust network parameters, which will not be repeated here.

[0052] The above scheme obtains a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; and identifies the target radiation signal based on a target recognition model to obtain a target device category of the target radiation source; wherein the target recognition model uses the first device category of the sample radiation signal as a training target, and is trained based on the sample signal features of the sample radiation signal and representative signal features of several sample device categories, wherein the first device category is determined by clustering the sample radiation signals, the sample signal features include feature information obtained by clustering the sample radiation signals, the sample radiation signal comes from the sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category. Therefore, by training the target recognition model through the sample signal features of the sample radiation signal and the representative signal features of various sample device categories, the target recognition model can simultaneously learn the category feature information of the first device category and the second device category, and capture the relationship between the sample signal features and the first device category and the second device category, which is beneficial to improving the adaptability of the target recognition model to sample device categories including the first device category and the second device category. Further, the target radiation signal can be identified through the target recognition model, and the target radiation signal initially identified as an unknown device category can be further identified, which can improve the accuracy of radiation source identification, especially the accuracy of radiation source identification in unknown open set scenarios.

[0053] See also Figure 3 , Figure 3 This is a framework diagram of an embodiment of a radiation source identification device 30 of the present application, wherein the radiation source identification device 30 includes an acquisition module 31 and an identification module 32, wherein the acquisition module 31 is used to acquire a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; the identification module 32 is used to identify the target radiation signal based on a target identification model to obtain a target device category of the target radiation source; wherein the target identification model takes a first device category of a sample radiation signal as a training target, and is trained based on sample signal features of the sample radiation signal and representative signal features of several sample device categories, wherein the first device category is determined by clustering each sample radiation signal, the sample signal features include feature information obtained by clustering the sample radiation signal, the sample radiation signal comes from a sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include a first device category and at least one second device category representing a known device category.

[0054] In the above scheme, the radiation source identification device 30 obtains a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; and identifies the target radiation signal based on a target identification model to obtain a target device category of the target radiation source; wherein the target identification model takes the first device category of the sample radiation signal as a training target, and is trained based on the sample signal features of the sample radiation signal and representative signal features of several sample device categories, wherein the first device category is determined by clustering the sample radiation signals, the sample signal features include feature information obtained by clustering the sample radiation signals, the sample radiation signal comes from the sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category. Therefore, by training the target recognition model through the sample signal features of the sample radiation signal and the representative signal features of various sample device categories, the target recognition model can simultaneously learn the category feature information of the first device category and the second device category, and capture the relationship between the sample signal features and the first device category and the second device category, which is beneficial to improving the adaptability of the target recognition model to sample device categories including the first device category and the second device category. Further, the target radiation signal can be identified through the target recognition model, and the target radiation signal initially identified as an unknown device category can be further identified, which can improve the accuracy of radiation source identification, especially the accuracy of radiation source identification in unknown open set scenarios.

[0055] In some disclosed embodiments, the radiation source identification device 30 also includes a training module, which is used to train a target recognition model. The training module includes a prediction submodule and an adjustment submodule. The steps of training the target recognition model by the training module include: the prediction submodule is used to obtain a first predicted device category of the sample radiation source based on sample signal characteristics and representative signal characteristics; the adjustment submodule is used to adjust the network parameters of the target recognition model based on the difference between the first predicted device category and the first device category.

[0056] Therefore, the first device category is used as the training target of the target recognition model. On the basis that the target recognition model can recognize radiation sources of known device categories, the device categories that the target recognition model can recognize are increased, thereby improving the compatibility of the target recognition model in identifying radiation sources.

[0057] In some disclosed embodiments, the prediction submodule is specifically used to construct graph structure data based on sample signal features and representative signal features; wherein each node in the graph structure data represents a different signal feature; obtain sample fusion features of each node in the graph structure data; wherein the sample fusion features of the node at least include the signal features represented by the node and feature information of other nodes connected to the node; for the node representing the sample signal feature, execute: based on the feature similarity between the sample fusion features of the node and the sample fusion features corresponding to the node representing the representative signal feature, determine the first predicted device category of the sample radiation source to which the sample signal feature represented by the node belongs.

[0058] Therefore, the use of graph structure data can intuitively reflect the signal characteristics and the complex relationship between signal characteristics, which helps to improve the efficiency of obtaining sample fusion features of each node. On the basis of the graph structure data, according to the characteristic relationship between the sample fusion features of each node, the first predicted device category of the sample radiation source can be determined, which can improve the accuracy of obtaining the first predicted device category.

[0059] In some disclosed embodiments, when the prediction submodule is specifically used to determine the first predicted device category of the sample radiation source to which the sample signal feature represented by a node belongs based on the feature similarity between the sample fusion feature of the node and the sample fusion feature corresponding to the node representing the representative signal feature, the prediction submodule is also specifically used to select the sample fusion feature corresponding to the node representing the sample signal feature as the sample target feature, and use the sample fusion features other than the sample target feature as the sample candidate feature; select the sample candidate feature whose feature similarity with the sample target feature meets the similarity requirement as the sample matching feature, and use the sample device category to which the sample matching feature belongs as the first predicted device category.

[0060] Therefore, by selecting sample candidate features whose feature similarity with the sample target features meets the similarity requirements as sample matching features, and taking the sample device category to which the sample matching features belong as the first predicted device category, it is possible to select the device category that best matches the feature information contained in the sample signal features from the sample device category, thereby improving the prediction accuracy of the first predicted device category.

[0061] In some disclosed embodiments, when the prediction submodule is specifically used to obtain sample fusion features of each node in the graph structure data, the prediction submodule is also specifically used to process the graph structure data based on the graph attention mechanism to obtain the aggregation features of the nodes in the graph structure data; based on the signal features and aggregation features corresponding to the same nodes, the sample fusion features of the corresponding nodes are fused.

[0062] Therefore, on the basis of obtaining the aggregated features of the node, based on the signal features and the aggregated features corresponding to the same node, the sample fusion features of the corresponding node are fused to obtain the new signal features represented by the node. According to the feature relationship between the new signal features represented by each node, the first predicted device category can be obtained, and the fusion learning of the category feature information between the first device category and the second device category can be promoted, thereby improving the compatibility of the target recognition model with the first device category and the second device category.

[0063] In some disclosed embodiments, the training module also includes a verification submodule, which is used to process the sample signal characteristics and the representative signal characteristics based on the target recognition model to obtain a second predicted device category of the sample radiation source after the adjustment module adjusts the network parameters of the target recognition model based on the difference between the first predicted device category and the first device category; wherein the second predicted device category is the same as or different from the first predicted device category; the second predicted device category is verified according to the real device category of the sample radiation signal to obtain a verification result; wherein the verification result includes: whether the second predicted device category is consistent with the real device category; according to the verification result, the network parameters of the target recognition model are adjusted based on the difference between the second predicted device category and the real device category, or the second predicted device category is added as a new second device category.

[0064] Therefore, after obtaining the target recognition model by training with the first device category as the training target, the trained target recognition model is tested, and the target recognition model is continued to be trained according to the test results, or the second predicted device category is saved as a new second device category, which can improve the accuracy and flexibility of the target recognition model training.

[0065] In some disclosed embodiments, the first device category is predicted by a clustering model, and the verification submodule is specifically used to adjust the network parameters of the clustering model in response to the verification result including that the second predicted device category is inconsistent with the actual device category, based on the difference between the second predicted device category and the actual device category.

[0066] Therefore, since the second predicted device category is predicted by the target recognition model trained with the first device category as the training target, the difference between the second predicted device category and the real device category can reflect the difference between the first device category and the real device category. Therefore, according to the difference between the second predicted device category and the real device category, the network parameters of the clustering model are adjusted, which can promote the accuracy of the clustering model in predicting the first device category of the sample radiation signal.

[0067] In some disclosed embodiments, when there are multiple real device categories, the inconsistency between the second predicted device category and the real device category includes any of the following: the number of types of the second predicted device category is inconsistent with the number of types of the real device category, and at least one second predicted device category is inconsistent with its corresponding real device category.

[0068] Therefore, the second predicted device category is verified from two levels: the difference in the number of types and the difference in type characteristics between the second predicted device category and the actual device category. Different compensation strategies can be adopted according to different types of differences, thereby improving the flexibility of target recognition model or clustering model training.

[0069] In some disclosed embodiments, the radiation source identification device 30 also includes an electromagnetic feature extraction module, which is used to obtain several electromagnetic signal features of the target radiation signal before identifying the target radiation signal based on a target recognition model and obtaining the target device category of the target radiation source; performing phase adjustments on various electromagnetic signal features to obtain phase offset features corresponding to the various electromagnetic signal features; wherein at least two electromagnetic signal features have opposite phase offset directions; fusing the phase offset features corresponding to the several electromagnetic signal features to obtain the electromagnetic fusion features of the target radiation signal; and then the identification module identifies the electromagnetic fusion features based on the target recognition model to obtain the target device category of the target radiation source.

[0070] Therefore, by identifying the electromagnetic fusion characteristics of the target radiation signal through the target recognition model, on the one hand, the signal feature information of the target radiation signal in different dimensions can be obtained; on the other hand, by performing phase shift processing on at least two electromagnetic signal characteristics of the target radiation signal in opposite directions to increase the distance between different types of electromagnetic signal characteristics of the same target radiation signal, the signal feature information of different dimensions can be distinguished as clearly as possible, and the validity of the data can be improved by reducing the interference between the feature information of different dimensions, thereby improving the data processing efficiency of the target recognition model.

[0071] See also Figure 4 , Figure 4 Schematic diagram of the framework of an embodiment of an electronic device 40 of the present application. The electronic device 40 includes a memory 41 and a processor 42 coupled to each other, and the processor 42 is used to execute program instructions stored in the memory 41 to implement the steps in any of the above-mentioned radiation source identification method embodiments. Specifically, the electronic device 40 may include but is not limited to: a microcomputer, a server, a laptop computer, a tablet computer, a smart phone, etc., which are not limited here.

[0072] Specifically, the processor 42 is used to control itself and the memory 41 to implement the steps of any of the above-mentioned radiation source identification method embodiments. The processor 42 can also be called a CPU (Central Processing Unit). The processor 42 may be an integrated circuit chip with signal processing capabilities. The processor 42 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 42 can be implemented by an integrated circuit chip.

[0073] In the above scheme, the electronic device 40 obtains a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; and identifies the target radiation signal based on a target recognition model to obtain a target device category of the target radiation source; wherein the target recognition model takes the first device category of the sample radiation signal as a training target, and is trained based on the sample signal features of the sample radiation signal and representative signal features of several sample device categories, wherein the first device category is determined by clustering the sample radiation signals, the sample signal features include feature information obtained by clustering the sample radiation signals, the sample radiation signal comes from the sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category. Therefore, by training the target recognition model through the sample signal features of the sample radiation signal and the representative signal features of various sample device categories, the target recognition model can simultaneously learn the category feature information of the first device category and the second device category, and capture the relationship between the sample signal features and the first device category and the second device category, which is beneficial to improving the adaptability of the target recognition model to sample device categories including the first device category and the second device category. Further, the target radiation signal can be identified through the target recognition model, and the target radiation signal initially identified as an unknown device category can be further identified, which can improve the accuracy of radiation source identification, especially the accuracy of radiation source identification in unknown open set scenarios.

[0074] See also Figure 5 , Figure 5The schematic diagram of the framework of an embodiment of a computer-readable storage medium 50 of the present application. The computer-readable storage medium 50 stores program instructions 51 that can be executed by a processor, and the program instructions 51 are used to implement the steps of the above-mentioned radiation source identification method embodiment.

[0075] In the above scheme, the computer-readable storage medium 50 obtains a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; and identifies the target radiation signal based on a target recognition model to obtain a target device category of the target radiation source; wherein the target recognition model takes the first device category of the sample radiation signal as a training target, and is trained based on the sample signal features of the sample radiation signal and representative signal features of several sample device categories, wherein the first device category is determined by clustering the sample radiation signals, the sample signal features include feature information obtained by clustering the sample radiation signals, the sample radiation signal comes from the sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category. Therefore, by training the target recognition model through the sample signal features of the sample radiation signal and the representative signal features of various sample device categories, the target recognition model can simultaneously learn the category feature information of the first device category and the second device category, and capture the relationship between the sample signal features and the first device category and the second device category, which is beneficial to improving the adaptability of the target recognition model to sample device categories including the first device category and the second device category. Further, the target radiation signal can be identified through the target recognition model, and the target radiation signal initially identified as an unknown device category can be further identified, which can improve the accuracy of radiation source identification, especially the accuracy of radiation source identification in unknown open set scenarios.

[0076] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0077] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0078] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0079] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0080] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0081] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, the personal information processing rules are notified by obvious signs / information, and the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A radiation source identification method, characterized in that: include: Acquire a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; Identify the target radiation signal based on a target recognition model to obtain a target device category of the target radiation source; Among them, the target recognition model takes the first device category of the sample radiation signal as the training target, and is trained based on the sample signal characteristics of the sample radiation signal and the representative signal characteristics of several sample device categories, the first device category is determined by clustering each of the sample radiation signals, the sample signal characteristics include characteristic information of the sample radiation signal obtained by clustering, the sample radiation signal comes from a sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category.

2. The method according to claim 1, characterized in that The training process of the target recognition model includes: Based on the sample signal feature and the representative signal feature, obtaining a first predicted device category of the sample radiation source; Based on the difference between the first predicted device category and the first device category, a network parameter of the object recognition model is adjusted.

3. The method according to claim 2, characterized in that The obtaining, based on the sample signal feature and the representative signal feature, a first predicted device category of the sample radiation source comprises: Based on the sample signal features and the representative signal features, graph structure data is constructed; wherein each node in the graph structure data represents a different signal feature; Acquire sample fusion features of each node in the graph structure data; wherein the sample fusion features of the node at least include signal features represented by the node and feature information of other nodes connected to the node; For the node representing the sample signal feature, executing: based on the feature similarity between the sample fusion feature of the node and the sample fusion feature corresponding to the node representing the representative signal feature, determining the first predicted device category of the sample radiation source to which the sample signal feature represented by the node belongs.

4. The method according to claim 3, characterized in that The determining, based on the feature similarity between the sample fusion feature of the node and the sample fusion feature corresponding to the node representing the representative signal feature, a first prediction device category of the sample radiation source to which the sample signal feature represented by the node belongs, comprises: Selecting a sample fusion feature corresponding to a node representing the sample signal feature as a sample target feature, and using sample fusion features other than the sample target feature as sample candidate features; The sample candidate feature whose feature similarity with the sample target feature meets the similarity requirement is selected as the sample matching feature, and the sample device category to which the sample matching feature belongs is used as the first predicted device category.

5. The method according to claim 3, characterized in that: The obtaining of sample fusion features of each node in the graph structure data includes: Processing the graph structure data based on a graph attention mechanism to obtain aggregate features of the nodes in the graph structure data; Based on the signal features and aggregation features corresponding to the same node, the sample fusion features corresponding to the node are obtained by fusion.

6. The method according to claim 2, characterized in that After adjusting the network parameters of the target recognition model based on the difference between the first predicted device category and the first device category, the method further includes: Processing the sample signal feature and the representative signal feature based on the target recognition model to obtain a second predicted device category of the sample radiation source; wherein the second predicted device category is the same as or different from the first predicted device category; Verify the second predicted device category according to the real device category of the sample radiation signal to obtain a verification result; wherein the verification result includes: whether the second predicted device category is consistent with the real device category; According to the verification result, the network parameters of the target recognition model are adjusted based on the difference between the second predicted device category and the real device category, or the second predicted device category is added as a new second device category.

7. The method according to claim 6, characterized in that The first device category is predicted by a clustering model, and the second predicted device category is verified according to the real device category of the sample radiation signal. After obtaining the verification result, the method further includes: In response to the verification result including that the second predicted device category is inconsistent with the real device category, a network parameter of the clustering model is adjusted based on the difference between the second predicted device category and the real device category.

8. The method according to claim 6, characterized in that In the case that there are multiple real device categories, the inconsistency between the second predicted device category and the real device category includes any of the following: the number of types of the second predicted device category is inconsistent with the number of types of the real device category, and at least one of the second predicted device categories is inconsistent with its corresponding real device category.

9. The method according to claim 1, characterized in that: Before identifying the target radiation signal based on the target recognition model to obtain the target device category of the target radiation source, the method further includes: Acquiring several electromagnetic signal characteristics of the target radiation signal; Performing phase adjustments on various electromagnetic signal characteristics respectively to obtain phase offset characteristics corresponding to various electromagnetic signal characteristics respectively; wherein at least two of the electromagnetic signal characteristics have opposite phase offset directions; Based on the phase shift characteristics corresponding to the several electromagnetic signal characteristics, the electromagnetic fusion characteristics of the target radiation signal are obtained; The identifying the target radiation signal based on the target recognition model to obtain the target device category of the target radiation source includes: The electromagnetic fusion feature is identified based on the target recognition model to obtain a target device category of the target radiation source.

10. A radiation source identification device, characterized in that: include: An acquisition module, configured to acquire a target radiation signal from a target radiation source; wherein the target radiation signal is initially identified as an unknown device category; An identification module, used to identify the target radiation signal based on a target identification model to obtain a target device category of the target radiation source; Among them, the target recognition model takes the first device category of the sample radiation signal as the training target, and is trained based on the sample signal characteristics of the sample radiation signal and the representative signal characteristics of several sample device categories, the first device category is determined by clustering each of the sample radiation signals, the sample signal characteristics include characteristic information of the sample radiation signal obtained by clustering, the sample radiation signal comes from a sample radiation source, and the sample radiation signal is initially identified as an unknown device category, and the several sample device categories include the first device category and at least one second device category representing a known device category.

11. An electronic device, characterized in that: It comprises a memory and a processor coupled to each other, wherein the processor is used to execute program instructions stored in the memory to implement the radiation source identification method according to any one of claims 1 to 9.

12. A computer-readable storage medium having program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the radiation source identification method according to any one of claims 1 to 9 is implemented.

Images