Small sample radiation source individual identification model training method and device and electronic equipment

Abstract

The invention provides a small sample radiation source individual identification model training method and device and electronic equipment, and relates to the technical field of radiation source individual identification, and the method comprises the steps: obtaining an original signal sample, and carrying out the form conversion of the original signal sample, and obtaining a multi-domain signal sample; the multi-domain signal sample comprises same-direction-orthogonal component data, bispectrum data and Hilbert spectrum data; performing phase rotation on the multi-domain signal sample to obtain an enhanced signal sample; performing feature extraction on the enhanced signal sample, and performing feature fusion on the extracted features to obtain multi-domain embedded features; performing classification based on multi-domain embedded features to obtain a prediction result of an enhanced signal sample; and constructing multi-task loss based on the multi-domain embedded features, the prediction result and the sample label of the original signal sample, and performing model optimization based on the multi-task loss to obtain a trained radiation source individual identification model. According to the invention, radiation source individual identification in a small sample scene can be carried out, and the identification accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of radiation source individual identification, and in particular to a small sample radiation source individual identification model training method, device and electronic equipment. Background Art

[0002] Specific Emitter Identification (SEI) aims to identify specific individual emitters based on unintentional modulation features parasitic on wireless signals. Unintentional modulation characteristics, also known as Radio Frequency Fingerprint (RFF), originate from transmitter hardware damage during the manufacturing process, so they are unique to each transmitter and difficult to forge. The typical SEI process includes signal reception, feature extraction and classification. Among them, the received signal refers to the signal received from a transmitter that has passed through an unknown channel; feature extraction refers to extracting the RFF related to the transmitter; identification is to implement signal classification based on the RFF to obtain the transmitter identification result.

[0003] With the rapid development of Deep Learning (DL), it provides a new and effective solution for SEI. Generally speaking, DL-based SEI methods use a large number of labeled training samples and Deep Neural Networks (DNN) to simultaneously achieve feature extraction and recognition. Due to the powerful adaptive feature extraction capability of DL, these methods have achieved better performance. However, due to the data-driven nature of DL, the performance of these methods will be greatly reduced when there are insufficient training samples. In the field of signal recognition, signal samples are often difficult to obtain and the labeling cost is high, which limits the practical application of DL-based SEI. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a small sample radiation source individual identification model training method, device and electronic equipment, which can perform individual radiation source identification in a small sample scenario and improve the identification accuracy.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows: In the first aspect, the present invention provides a small sample radiation source individual recognition model training method, comprising: obtaining original signal samples, and converting the original signal samples into a multi-domain signal sample; wherein the multi-domain signal sample comprises: in-phase-orthogonal component data, bispectral data and Hilbert spectrum data; performing phase rotation on the multi-domain signal sample to obtain an enhanced signal sample; performing feature extraction on the enhanced signal sample, and performing feature fusion on the extracted features to obtain a multi-domain embedded feature; performing classification based on the multi-domain embedded feature to obtain a prediction result of the enhanced signal sample; constructing a multi-task loss based on the multi-domain embedded feature, the prediction result and the sample label of the original signal sample, and performing model optimization based on the multi-task loss to obtain a trained radiation source individual recognition model.

[0006] Optionally, the original signal samples are converted in form, including: converting the original signal samples into in-phase-quadrature component data, and performing power normalization on the in-phase-quadrature component data; performing a bi-spectrum transform on the normalized in-phase-quadrature component data, and obtaining a diagonal slice of the bi-spectrum as bi-spectrum data; performing a Hilbert-Huang transform on the normalized in-phase-quadrature component data, and performing amplitude-frequency mapping on the Hilbert spectrum to obtain the Hilbert spectrum data.

[0007] Optionally, performing phase rotation on the multi-domain signal samples to obtain enhanced signal samples includes: performing phase rotation on the in-phase-orthogonal component data, the bispectral data, and the Hilbert spectrum data respectively according to the following formula:

[0008] in, represents the phase rotation angle, and denote the real and imaginary parts of the multi-domain signal samples, respectively, and Represents the real and imaginary parts of the multi-domain signal samples after phase rotation.

[0009] Optionally, feature extraction is performed on the enhanced signal samples, and the extracted features are fused to obtain multi-domain embedded features, including: using a complex-valued convolutional neural network to extract features from the isotropic-orthogonal component data, bispectral data, and Hilbert spectrum data, respectively, and using a fusion device to fuse the extracted features to obtain multi-domain embedded features.

[0010] Optionally, a multi-task loss is constructed based on the multi-domain embedding features, the prediction results and the sample labels of the original signal samples, and a model is optimized based on the multi-task loss to obtain a trained radiation source individual recognition model, including: calculating the center loss and the triplet loss based on the multi-domain embedding features and the sample labels; calculating the cross entropy loss based on the prediction results and the sample labels; determining the multi-task loss based on the center loss, the triplet loss and the cross entropy loss; wherein the multi-task loss is: , is the multi-task loss, is the center loss, is the triplet loss, is the cross entropy loss, , and are the learning parameters; a multi-task learning strategy is used to adjust the parameters of each loss in the multi-task loss to obtain a trained radiation source individual recognition model.

[0011] In a second aspect, the present invention provides a method for identifying individual radiation sources, comprising: obtaining a received signal of a receiver; classifying the received signal based on a pre-trained radiation source individual identification model to obtain an individual radiation source category; wherein the radiation source individual identification model is trained using any of the small sample radiation source individual identification model training methods provided in the first aspect.

[0012] In the third aspect, the present invention provides a small sample radiation source individual identification model training device, including: a sample acquisition module, used to obtain original signal samples, and convert the original signal samples into a form to obtain multi-domain signal samples; wherein the multi-domain signal samples include: in-phase-orthogonal component data, bispectral data and Hilbert spectrum data; a signal enhancement module, used to phase rotate the multi-domain signal samples to obtain enhanced signal samples; a feature extraction module, used to extract features from the enhanced signal samples, and perform feature fusion on the extracted features to obtain multi-domain embedded features; a classification module, used to perform classification based on the multi-domain embedded features to obtain prediction results of the enhanced signal samples; a multi-task learning module, used to construct a multi-task loss based on the multi-domain embedded features, the prediction results and the sample labels of the original signal samples, and perform model optimization based on the multi-task loss to obtain a trained radiation source individual identification model.

[0013] In a fourth aspect, the present invention provides a radiation source individual identification device, comprising: a signal acquisition module, used to acquire a received signal of a receiver; an individual identification module, used to classify the received signal based on a pre-trained radiation source individual identification model to obtain the individual category of the radiation source; wherein the radiation source individual identification model is trained using any small sample radiation source individual identification model training method provided in the first aspect.

[0014] In a third aspect, the present invention provides an electronic device comprising a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement any one of the methods provided in the first aspect or the steps of the method provided in the second aspect.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program executes any one of the steps of the method provided in the first aspect or the second aspect.

[0016] The present invention brings the following beneficial effects: The above-mentioned small sample radiation source individual recognition model training method, device and electronic device provided by the present invention first obtain the original signal sample, and convert the original signal sample into a form to obtain a multi-domain signal sample (co-directional-orthogonal component data, bispectral data and Hilbert spectrum data); secondly, the multi-domain signal sample is phase rotated to obtain an enhanced signal sample; then the enhanced signal sample is feature extracted, and the extracted features are feature fused to obtain a multi-domain embedded feature; then classification is performed based on the multi-domain embedded feature to obtain the prediction result of the enhanced signal sample; finally, a multi-task loss is constructed based on the multi-domain embedded feature, the prediction result and the sample label of the original signal sample, and the model is optimized based on the multi-task loss to obtain a trained radiation source individual recognition model. In the above method, by converting the form and phase shifting the small sample training data set, the training data set can be expanded, thereby reducing the dependence on the auxiliary data set, improving the situation where the data is difficult to obtain and label, and improving the information volume and information reliability of the small sample data set; at the same time, through multi-domain feature fusion and multi-task loss learning, the discernibility of the multi-domain fusion feature is improved, thereby improving the recognition accuracy in the small sample scenario.

[0017] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the preferred embodiments are specifically mentioned below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A flowchart of a method for training a small sample radiation source individual identification model provided by an embodiment of the present invention; Figure 2 A framework diagram of a small sample radiation source individual identification based on multi-domain fusion and metric learning provided by an embodiment of the present invention; Figure 3 A flow chart of a method for identifying an individual radiation source provided by an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a small sample radiation source individual recognition model training device provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of a radiation source individual identification device provided by an embodiment of the present invention; Figure 6 A schematic structural diagram of a small electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] At present, the existing FS-SEI can be divided into FS-SEI based on pre-training-fine-tuning and FS-SEI based on data enhancement. FS-SEI based on pre-training-fine-tuning is limited by the pre-training process and auxiliary data sets. Since the cost of labeling signals in the field of signal recognition is high and difficult to achieve in non-cooperative communication scenarios, obtaining a suitable auxiliary data set with a large amount of data is not suitable for practical applications. Therefore, the dependence on auxiliary data sets is the main disadvantage of this technology. For FS-SEI based on data enhancement, since the effectiveness of virtual samples produced by data enhancement is extremely dependent on the quality of the original real samples, when the number of samples is extremely small, the overall quality of the samples is difficult to guarantee, resulting in performance far inferior to the previous method. Therefore, poor performance under extremely small samples is the main disadvantage of this technology. In addition, most of the above two schemes use single-domain signal representation as network input, lack multi-dimensional information guidance and fusion, resulting in room for improvement in performance in small sample scenarios.

[0023] Based on this, the embodiments of the present invention provide a small sample radiation source individual identification model training method, device and electronic equipment, which can perform individual radiation source identification in small sample scenarios and improve the identification accuracy.

[0024] To facilitate understanding of this embodiment, a small sample radiation source individual recognition model training method disclosed in an embodiment of the present invention is first introduced in detail. The method can be executed by an electronic device, such as a smart phone, a computer, a tablet computer, etc. Figure 1 The flowchart of a method for training a small sample radiation source individual identification model is shown, indicating that the method mainly includes the following steps S101 to S105: Step S101: obtaining original signal samples, and converting the original signal samples into a form to obtain multi-domain signal samples.

[0025] In one implementation, the original signal samples are first represented in the form of three different domains, namely, in-phase quadrature (IQ), bispectrum and Hilbert spectrum, that is, the original signal samples are converted into multi-domain signal samples: in-phase quadrature data, bispectrum data and Hilbert spectrum data. In a specific implementation, when the original signal samples are converted into a form, the following methods may be used, including but not limited to: First, the original signal samples are converted into in-phase-quadrature component data, and the in-phase-quadrature component data are power normalized.

[0026] In the specific implementation, for the received original signal sample , considering the rich information of the in-phase-quadrature component data, it is used as the first signal representation, that is, ,in , and Respectively In order to reduce the impact of power inconsistency on RFF extraction, power normalization is performed on each original signal sample:

[0027] Then, the normalized in-phase-quadrature component data are subjected to bispectral transformation, and diagonal slices of the bispectrum are obtained as bispectral data.

[0028] In a specific implementation, the bispectrum is a high-order spectral function of two frequencies, which can effectively suppress additive Gaussian noise and provide a measure of signal Gaussianity. Compared with the power spectrum, the bispectrum also carries phase information. Based on this, the bispectrum is used as a powerful tool for analyzing RFF in the embodiment of the present invention.

[0029] The bispectrum is defined as a two-dimensional Fourier transform of a third-order correlation function. In the embodiment of the present invention, a non-parametric method is used to estimate the bispectrum. Representing the estimation function, the estimated bispectrum can be obtained: .

[0030] In order to effectively reduce the dimensionality, the diagonal slices of the obtained bispectrum are used as signal representation (i.e., bispectral data), so that the relevant information can be preserved to a large extent. for The diagonal slice of , that is: .

[0031] It should be noted that ,in, and They are The real and imaginary parts of .

[0032] Finally, the normalized in-phase-quadrature component data are subjected to Hilbert-Huang transform, and the Hilbert spectrum is mapped to frequency to obtain the Hilbert spectrum data.

[0033] In specific implementation, Hilbert spectrum is a time-frequency signal representation method implemented by Hilbert-Huang Transform (HHT), which is an effective method for analyzing nonlinear and non-stationary signals. The method first decomposes the received signal into a set of intrinsic mode functions (IMFs) according to empirical mode decomposition (EMD), and then performs Hilbert transform on them. Through Hilbert transform, IMF generates instantaneous frequency functions that form frequency-time distribution.

[0034] Since HHT can only operate on real signals, the complex signal is divided into real and imaginary parts in the embodiment of the present invention, and HHT operation is performed on each of them to obtain the corresponding Hilbert spectrum. For example, suppose Representing the HHT function, we can get the Hilbert spectrum:

[0035] Considering that the RFF information in the time dimension is seriously polluted by the channel, in order to reduce the dimension, the Hilbert spectrum is mapped to the frequency domain to obtain the Hilbert spectrum data. for The amplitude-frequency projection of the Hilbert spectrum can be obtained: .

[0036] Similarly, we can get of . Combining these two parts into the final Hilbert spectrum representation .

[0037] In summary, we can get multi-domain signal samples, that is, multi-domain sample matrix ,in, , , They are , and Column vector form of .

[0038] Step S102: performing phase rotation on the multi-domain signal samples to obtain enhanced signal samples.

[0039] In one embodiment, each sample in the three fields is phase rotated. Phase rotation, also known as phase shift, is performed on the sample. , the phase rotation can be expressed as:

[0040] in, represents the phase rotation angle, and denote the real and imaginary parts of the multi-domain signal samples, respectively, and Represents the real and imaginary parts of the multi-domain signal samples after phase rotation.

[0041] Compared with the method of first rotating the phase and then performing domain transformation, the method adopted in the embodiment of the present invention can obtain a wider range of implicit samples, thereby better serving small sample feature extraction.

[0042] Step S103: extract features from the enhanced signal samples, and fuse the extracted features to obtain multi-domain embedded features.

[0043] In one implementation, a complex-valued convolutional neural network may be used to extract features from the isotropic-orthogonal component data, the bispectral data, and the Hilbert spectrum data, respectively, and a fuser may be used to fuse the extracted features to obtain multi-domain embedded features.

[0044] In specific implementation, , and represent the feature extraction functions of isotropic-orthogonal component data, bispectral data, and Hilbert spectrum data, respectively. represents the fuser function, Represents the multi-domain sample matrix after phase rotation, and then the multi-domain embedding features can be obtained: .

[0045] Based on this, for a multi-domain signal sample dataset , represents the overall feature embedding function, i The embedding of a sample can be expressed as .

[0046] Step S104: Classify based on the multi-domain embedded features to obtain the prediction result of the enhanced signal sample.

[0047] In one implementation, the classifier network may adopt a fully connected network, and the multi-domain embedded features are classified by the classifier network to obtain a prediction result.

[0048] Step S105: construct a multi-task loss based on the multi-domain embedding features, the prediction results and the sample labels of the original signal samples, and perform model optimization based on the multi-task loss to obtain a trained radiation source individual recognition model.

[0049] In one implementation, the center loss and triplet loss in metric learning are introduced with the cross entropy loss to construct a multi-task loss to form an overall optimization target, and the model is optimized to obtain a trained radiation source individual recognition model. The center loss and triplet loss are calculated using the multi-domain embedding features of the sample and the sample label, and the cross entropy loss is calculated using the prediction result and the sample label.

[0050] In specific implementation, when optimizing the model, the following methods may be used, including but not limited to: First, the center loss and triplet loss are calculated based on the multi-domain embedding features and sample labels.

[0051] Specifically, the center loss learns the clustering centers of the multi-domain embedding features of each category, while collecting the embeddings of the same category into their centers. The center loss can be expressed as:

[0052] in, express The cluster center of the embedded features of the class, is the Euclidean distance.

[0053] The triplet loss considers the intra-class relationship and the inter-class relationship, and uses triplet samples to construct the optimization target. The triplet sample contains an anchor sample , a positive sample and a negative sample ,in and Have the same tags, and There are different labels. The triplet loss is:

[0054] in, is an adjustable marginal parameter.

[0055] Secondly, the cross entropy loss is calculated based on the prediction results and sample labels.

[0056] In the specific implementation, it is assumed Represents the classifier function, then the prediction result vector can be obtained:

[0057] in, For the i In the sample j The predicted value of each category, C is the total number of categories.

[0058] The cross entropy loss reduces the difference between the true probability distribution and the predicted probability distribution and can be expressed as:

[0059] Then, the multi-task loss is determined based on the center loss, triplet loss, and cross entropy loss.

[0060] Among them, the multi-task loss is: , is the multi-task loss, is the center loss, is the triplet loss, is the cross entropy loss, , and are the learning parameters; Finally, a multi-task learning strategy is used to adjust the parameters of each loss in the multi-task loss to obtain a trained radiation source individual recognition model.

[0061] In the specific implementation, in order to integrate the above losses, a multi-task learning strategy is adopted. By estimating the uncertainty of each task, the weight of each task in the loss function is automatically adjusted to obtain a trained radiation source individual recognition model.

[0062] The above-mentioned small sample radiation source individual identification model training method provided by the present invention can expand the training data set by transforming the form and phase shifting the small sample training data set, thereby reducing the dependence on auxiliary data sets, improving the situation where data is difficult to obtain and label, and improving the information volume and information reliability of the small sample data set; at the same time, through multi-domain feature fusion and multi-task loss learning, the distinguishability of multi-domain fusion features is improved, thereby improving the recognition accuracy in small sample scenarios.

[0063] For ease of understanding, the present invention provides a method for training a small sample radiation source individual recognition model based on multi-domain fusion and metric learning. Figure 2 As shown in the figure, it mainly includes three parts: multi-domain representation, multi-domain data enhancement and multi-domain fusion metric learning. Specifically, the multi-domain representation first represents the signal samples in the form of three different domains, namely, in-phase-orthogonal components (, bispectrum and Hilbert spectrum. Among them, the bispectrum and Hilbert spectrum are respectively subjected to dimensionality reduction operations of diagonal slicing and amplitude-frequency mapping. Multi-domain data enhancement performs phase rotation on the three forms respectively. Multi-domain fusion metric learning extracts single-domain features from the samples after multi-domain data enhancement with three feature extractor networks, and then fuses and maps them to the feature space through the fusion network. Finally, the center loss and triplet loss in metric learning are introduced to construct multi-task losses with cross entropy loss to form an overall optimization goal. Among them, the center loss and triplet loss are calculated by sample embedding features and sample labels, and the cross entropy loss is calculated by predicting the result vector and sample labels. The feature extractor network adopts a complex-valued convolutional neural network; the fusion network is implemented in the form of channel splicing and convolution; the classifier network is a fully connected network.

[0064] The above method provided by the embodiment of the present invention represents signal samples into three signal representation forms of IQ, bispectrum, and Hilbert spectrum, and adopts diagonal slicing and amplitude-frequency mapping dimensionality reduction methods for the bispectrum and Hilbert spectrum respectively to achieve a unified complex representation form, thereby increasing the amount of data set information while maintaining a low data dimension, and avoiding the problem of overfitting caused by excessive network model parameters; the phase rotation data enhancement technology of the time domain IQ signal is introduced into the bispectrum and Hilbert spectrum, which can mine new effective potential information space; the multi-domain signal representation is implemented through a feature extractor network and a fusion network to achieve feature fusion, and the center loss and triplet loss are introduced as regularization terms of the cross entropy loss, and the three are effectively combined through the multi-task loss to automatically calculate the sum weight, thereby improving the generalization performance and recognition accuracy.

[0065] The present invention also provides a method for identifying individual radiation sources. Figure 3 The flowchart of a method for identifying a radiation source individual is shown, which illustrates that the method mainly includes the following steps S301 to S302: Step S301: Acquire a received signal of a receiver.

[0066] Step S302: classify the received signal based on the pre-trained radiation source individual recognition model to obtain the radiation source individual category.

[0067] In one embodiment, the radiation source individual identification model is trained using the small sample radiation source individual identification model training method provided in the above embodiment. First, the received signal is represented as three signal representation forms: IQ, bispectrum, and Hilbert spectrum, and the bispectrum and Hilbert spectrum are respectively reduced in dimension by diagonal slicing and amplitude-frequency mapping; then, the IQ, bispectrum, and Hilbert spectrum are respectively feature extracted, and feature fusion is performed to obtain embedded features; finally, the classifier is used to classify and obtain the individual category of the radiation source.

[0068] The above-mentioned identification method provided by the embodiment of the present invention uses a pre-trained radiation source individual identification model for identification, and increases the amount of information by representing the signal in three signal representation forms: IQ, bispectrum, and Hilbert spectrum; and realizes feature fusion through a feature extractor network and a fusion network, and classifies features through a classifier network, thereby improving the accuracy of individual identification.

[0069] For the small sample radiation source individual recognition model training method provided in the above embodiment, the embodiment of the present invention also provides a small sample radiation source individual recognition model training device, see Figure 4 The schematic diagram of the structure of a small sample radiation source individual recognition model training device is shown, indicating that the device mainly includes the following parts: The sample acquisition module 401 is used to acquire original signal samples and convert the original signal samples into a multi-domain signal sample; wherein the multi-domain signal sample includes: isotropic-orthogonal component data, bispectral data and Hilbert spectrum data; The signal enhancement module 402 is used to perform phase rotation on the multi-domain signal samples to obtain enhanced signal samples; The feature extraction module 403 is used to extract features from the enhanced signal samples and perform feature fusion on the extracted features to obtain multi-domain embedded features; A classification module 404, configured to perform classification based on the multi-domain embedded features to obtain a prediction result of the enhanced signal sample; The multi-task learning module 405 is used to construct a multi-task loss based on the multi-domain embedded features, the prediction results and the sample labels of the original signal samples, and to optimize the model based on the multi-task loss to obtain a trained radiation source individual recognition model.

[0070] The above-mentioned small sample radiation source individual identification model training device provided by the present invention can expand the training data set by converting the form and phase shifting the small sample training data set, thereby reducing the dependence on auxiliary data sets, improving the situation where data is difficult to obtain and label, and improving the information volume and information reliability of the small sample data set; at the same time, through multi-domain feature fusion and multi-task loss learning, the distinguishability of multi-domain fusion features is improved, thereby improving the recognition accuracy in small sample scenarios.

[0071] For the radiation source individual identification method provided in the above embodiment, the present invention also provides a radiation source individual identification device, see Figure 5 The schematic diagram of the structure of a radiation source individual identification device shown in FIG. 1 shows that the device mainly includes the following parts: The signal acquisition module 501 is used to acquire a received signal of a receiver; The individual identification module 502 is used to classify the received signal based on a pre-trained radiation source individual identification model to obtain the radiation source individual category; wherein the radiation source individual identification model is trained using any of the small sample radiation source individual identification model training methods provided in the first aspect.

[0072] The above-mentioned radiation source individual identification device provided by the embodiment of the present invention uses a pre-trained radiation source individual identification model for identification, and increases the amount of information by representing the signal in three signal representation forms: IQ, bispectrum, and Hilbert spectrum; and realizes feature fusion through a feature extractor network and a fusion network, and classifies features through a classifier network, thereby improving the accuracy of individual identification.

[0073] The device provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0074] An embodiment of the present invention further provides an electronic device. Specifically, the electronic device includes a processor and a storage device. The storage device stores a computer program, and when the computer program is executed by the processor, it executes the method described in any one of the above implementation methods.

[0075] Figure 6 A structural diagram of an electronic device provided in an embodiment of the present invention, the electronic device 100 includes: a processor 60, a memory 61, a bus 62 and a communication interface 63, wherein the processor 60, the communication interface 63 and the memory 61 are connected via the bus 62; the processor 60 is used to execute an executable module stored in the memory 61, such as a computer program.

[0076] The memory 61 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 63 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0077] The bus 62 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0078] Among them, the memory 61 is used to store programs, and the processor 60 executes the program after receiving the execution instruction. The method executed by the device for flow process definition disclosed in any embodiment of the above-mentioned embodiment of the present invention can be applied to the processor 60 or implemented by the processor 60.

[0079] The processor 60 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 60. The above processor 60 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to execute, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 61, and the processor 60 reads the information in the memory 61 and completes the steps of the above method in combination with its hardware.

[0080] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be referred to the previous method embodiments, which will not be repeated here.

[0081] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the 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.) to perform all or part of the steps of the methods described in various embodiments of the present invention. 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, etc., which can store program codes.

[0082] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for training a small sample radiation source individual identification model, characterized in that: include: Acquire original signal samples, and convert the original signal samples into a form to obtain multi-domain signal samples; wherein the multi-domain signal samples include: isotropic-orthogonal component data, bispectral data, and Hilbert spectrum data; Performing phase rotation on the multi-domain signal samples to obtain enhanced signal samples; Extracting features from the enhanced signal samples, and fusing the extracted features to obtain multi-domain embedded features; Performing classification based on the multi-domain embedded features to obtain a prediction result of the enhanced signal sample; A multi-task loss is constructed based on the multi-domain embedding features, the prediction results and the sample labels of the original signal samples, and a model is optimized based on the multi-task loss to obtain a trained radiation source individual recognition model.

2. The method according to claim 1, characterized in that The original signal samples are converted into a form, including: Converting the original signal samples into in-phase-quadrature component data, and performing power normalization processing on the in-phase-quadrature component data; Performing bispectral transformation on the normalized in-phase-quadrature component data, and obtaining diagonal slices of the bispectrum as bispectral data; The normalized in-phase-quadrature component data are subjected to Hilbert-Huang transform, and the Hilbert spectrum is subjected to amplitude-frequency mapping to obtain the Hilbert spectrum data.

3. The method according to claim 1, characterized in that The multi-domain signal samples are phase rotated to obtain enhanced signal samples, including: Phase rotation is performed on the in-phase-quadrature component data, the bispectral data and the Hilbert spectrum data respectively according to the following formulas: in, represents the phase rotation angle, and denote the real and imaginary parts of the multi-domain signal samples, respectively, and Represents the real and imaginary parts of the multi-domain signal samples after phase rotation.

4. The method according to claim 1, characterized in that: Feature extraction is performed on the enhanced signal sample, and the extracted features are subjected to feature fusion to obtain multi-domain embedded features, including: A complex-valued convolutional neural network is used to extract features from the isotropic-orthogonal component data, the bispectral data and the Hilbert spectrum data respectively, and a fuser is used to fuse the extracted features to obtain multi-domain embedded features.

5. The method according to claim 1, characterized in that Constructing a multi-task loss based on the multi-domain embedding features, the prediction results and the sample labels of the original signal samples, and performing model optimization based on the multi-task loss to obtain a trained radiation source individual recognition model, including: Calculate the center loss and the triplet loss based on the multi-domain embedding features and the sample labels; Calculate cross entropy loss based on the prediction result and the sample label; A multi-task loss is determined based on the center loss, the triplet loss, and the cross entropy loss; wherein the multi-task loss is: , is the multi-task loss, is the center loss, is the triplet loss, is the cross entropy loss, , and are the learning parameters; A multi-task learning strategy is adopted to adjust the parameters of each loss in the multi-task loss to obtain a trained radiation source individual recognition model.

6. A method for identifying individual radiation sources, characterized in that: include: Acquire a received signal of a receiver; The received signal is classified based on a pre-trained radiation source individual identification model to obtain a radiation source individual category; wherein the radiation source individual identification model is trained using the small sample radiation source individual identification model training method described in any one of claims 1 to 5.

7. A small sample radiation source individual identification model training device, characterized in that: include: A sample acquisition module is used to acquire original signal samples and convert the original signal samples into a multi-domain signal sample; wherein the multi-domain signal sample includes: isotropic-orthogonal component data, bispectral data and Hilbert spectrum data; A signal enhancement module, used to perform phase rotation on the multi-domain signal samples to obtain enhanced signal samples; A feature extraction module is used to extract features from the enhanced signal samples and perform feature fusion on the extracted features to obtain multi-domain embedded features; A classification module, used for performing classification based on the multi-domain embedding features to obtain a prediction result of the enhanced signal sample; The multi-task learning module is used to construct a multi-task loss based on the multi-domain embedding features, the prediction results and the sample labels of the original signal samples, and to perform model optimization based on the multi-task loss to obtain a trained radiation source individual recognition model.

8. A radiation source individual identification device, characterized in that: include: A signal acquisition module, used to acquire a received signal of a receiver; An individual identification module is used to classify the received signal based on a pre-trained radiation source individual identification model to obtain the radiation source individual category; wherein the radiation source individual identification model is trained using the small sample radiation source individual identification model training method described in any one of claims 1 to 5.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the steps of the method according to any one of claims 1 to 5 or claim 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 5 or claim 6 are executed.

Images