Radiation source individual identification method, device, equipment, medium and product

Abstract

The invention provides a radiation source individual identification method, device, equipment, medium and product, and the method comprises the steps: obtaining a physical signal feature and a deep learning signal feature corresponding to an intermediate frequency signal of a to-be-identified radiation source individual, the physical signal feature being obtained by extracting the physical feature of the intermediate frequency signal, and the deep learning signal feature being obtained by extracting the physical feature of the intermediate frequency signal; the deep learning signal features are obtained by performing deep learning-based feature extraction on the intermediate frequency signals; respectively performing convolution processing on the physical signal features and the deep learning signal features to obtain convolution physical signal features and convolution deep learning signal features; and fusing the convolutional physical signal feature and the convolutional deep learning signal feature to obtain a fused signal feature, and obtaining an identification result of the radiation source individual based on the fused signal feature. According to the invention, the radiation source individual identification accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of electromagnetic signal processing technology, and in particular to a method, device, equipment, medium and product for identifying individual radiation sources. Background Art

[0002] Radiation Source Individual Identification (RSII) is mainly used to identify and distinguish different radiation sources from electromagnetic signals. This technology has a wide range of applications in communication countermeasures, physical layer security, wireless communications, network security and other fields. The core of radiation source individual identification is to use the inherent physical characteristics of the radiation source (such as hardware characteristics, manufacturing differences, etc.) as fingerprints to distinguish and identify different emission sources.

[0003] At present, the technology of identifying individual radiation sources faces multiple challenges. On the one hand, the electromagnetic environment is becoming increasingly complex, with dense signals and severe interference, which makes it very difficult to accurately identify specific radiation sources from numerous signals. On the other hand, traditional signal processing and analysis methods often rely on advanced characteristics of the signal, such as modulation type, coding method, etc., but in practical applications, these characteristics may be intentionally or unintentionally masked or changed, resulting in a decrease in recognition accuracy. Summary of the invention

[0004] Based on the above technical status, the present application proposes a method, device, equipment, medium and product for individual identification of radiation sources, which can improve the accuracy of individual identification of radiation sources.

[0005] In order to achieve the above technical objectives, this application specifically proposes the following technical solutions:

[0006] In a first aspect, the present application proposes a method for identifying individual radiation sources, comprising: obtaining physical signal features and deep learning signal features corresponding to an intermediate frequency signal of an individual radiation source to be identified, wherein the physical signal features are obtained by extracting the physical features of the intermediate frequency signal, and the deep learning signal features are obtained by performing feature extraction based on deep learning on the intermediate frequency signal; performing convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; fusing the convolution physical signal features and the convolution deep learning signal features to obtain fused signal features, and obtaining an identification result of the individual radiation source based on the fused signal features.

[0007] In some implementations, the method of obtaining physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the radiation source individual to be identified; performing convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; fusing the convolution physical signal features and the convolution deep learning signal features to obtain fusion signal features, and obtaining the identification result of the radiation source individual based on the fusion signal features, includes: inputting the intermediate frequency signal of the radiation source individual to be identified into the radiation source individual identification model, and obtaining the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the radiation source individual to be identified by the radiation source individual identification model. Features; performing convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; fusing the convolution physical signal features and the convolution deep learning signal features to obtain fused signal features, and obtaining the identification results of the radiation source individuals based on the fused signal features; wherein the radiation source individual identification model is obtained by training the pre-training model with the first intermediate frequency signal samples of each radiation source individual and their corresponding labeled radiation source individuals, and the pre-training model is obtained by training the pre-training network with the long-time series feature samples and short-time local feature samples corresponding to the second intermediate frequency signal samples of each radiation source individual.

[0008] In some implementations, the radiation source individual recognition model includes a deep learning feature extraction module, a physical signal feature extraction module, a feature fusion module and a classification module; wherein the intermediate frequency signal of the radiation source individual to be identified is input into the radiation source individual recognition model, and the radiation source individual recognition model obtains the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the radiation source individual to be identified; the physical signal features and the deep learning signal features are convolved respectively to obtain convolution physical signal features and convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain fused signal features, and based on the fused signal The method obtains the identification result of the individual radiation source by using the features, including: obtaining the deep learning signal features of the intermediate frequency signal of the individual radiation source to be identified through the deep learning feature extraction module; obtaining the physical signal features of the intermediate frequency signal of the individual radiation source to be identified through the physical signal feature extraction module; performing convolution processing on the physical signal features and the deep learning signal features respectively through the feature fusion module to obtain the convolution physical signal features and the convolution deep learning signal features, and fusing the convolution physical signal features and the convolution deep learning signal features to obtain the fused signal features; and classifying the fused signal features through the classification module to obtain the identification result of the individual radiation source.

[0009] In some implementations, the feature fusion module includes a first convolution module, a second convolution module and at least two fully connected layers; wherein, the physical signal feature and the deep learning signal feature are respectively convolved by the feature fusion module to obtain a convolution physical signal feature and a convolution deep learning signal feature, and the convolution physical signal feature and the convolution deep learning signal feature are fused to obtain a fused signal feature, including: convolving the physical signal feature by the first convolution module to obtain a convolution physical signal feature; convolving the deep learning signal feature by the second convolution module to obtain a convolution deep learning signal feature; and performing dimensionality reduction processing and splicing processing on the convolution physical signal feature and the convolution deep learning signal feature in turn by the at least two fully connected layers to obtain a fused signal feature.

[0010] In some implementations, the pre-trained network includes a deep learning feature extraction module to be trained, a physical signal feature extraction module to be trained, a feature fusion module to be trained, a sampling point mask network and a loss function module; the pre-trained model is trained by the following steps: obtaining deep learning signal feature samples of intermediate frequency signal samples through the deep learning feature extraction module to be trained, and obtaining physical signal feature samples of the intermediate frequency signal samples through the physical signal feature extraction module to be trained; fusing the obtained physical signal feature samples and deep learning signal feature samples through the feature fusion module to be trained to obtain fused signal feature samples; encoding the fused signal feature samples through the sampling point mask network to obtain a target encoding result, and predicting the feature vector corresponding to the mask position in the fused signal feature sample based on the target encoding result to obtain a predicted feature vector, wherein the target encoding result includes the long-time series feature samples and short-time local feature samples of the second intermediate frequency signal sample; optimizing the pre-trained network based on the difference between the predicted feature vector and the original feature vector corresponding to the mask position through the loss function module to obtain the pre-trained model.

[0011] In some implementations, the sampling point mask network includes a Conformer module, and the Conformer module includes an encoding module, a third convolution module and a mask corresponding sampling point selection module; wherein, the sampling point mask network encodes the fused signal feature sample to obtain a target encoding result, and predicts the feature vector corresponding to the mask position in the fused signal feature sample based on the target encoding result to obtain a predicted feature vector, including: encoding the fused signal sample feature through the encoding module to obtain an initial encoding result of the second intermediate frequency signal sample, and the initial encoding result includes a long-time series feature sample of the second intermediate frequency signal sample; convolution processing is performed on the initial encoding result through the third convolution module to obtain a target encoding result, and the target encoding result includes a long-time series feature sample and a local feature sample of the second intermediate frequency signal sample; and the mask corresponding sampling point selection module predicts the feature vector corresponding to the mask position in the fused signal feature sample based on the target encoding result to obtain a predicted feature vector.

[0012] In some implementations, the physical signal feature includes at least one of an envelope feature, a fractal feature, a power spectrum feature, a modal feature, a high-order moment feature, and an information entropy feature.

[0013] The second aspect of the present application proposes a radiation source individual identification device, comprising: an acquisition unit, used to acquire physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified, the physical signal features are obtained by extracting the physical features of the intermediate frequency signal, and the deep learning signal features are obtained by performing feature extraction based on deep learning on the intermediate frequency signal; a fusion and identification unit, used to perform convolution processing on the physical signal features and the deep learning signal features, respectively, to obtain convolution physical signal features and convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain fused signal features, and the identification result of the individual radiation source is obtained based on the fused signal features.

[0014] The third aspect of the present application proposes an electronic device, comprising a memory and a processor; the memory is connected to the processor and is used to store programs; the processor is used to implement the radiation source individual identification method described in the first aspect and any one of the implementation methods of the first aspect by running the program in the memory.

[0015] In a fourth aspect, the present application proposes a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for identifying an individual radiation source as described in the first aspect and any one of the implementation methods of the first aspect is implemented.

[0016] In a fifth aspect, the present application proposes a computer program product, including computer program instructions, which, when executed by a processor, enable the processor to implement the radiation source individual identification method described in the first aspect and any one of the implementation methods of the first aspect.

[0017] The radiation source individual identification method, device, equipment, medium and product proposed in the present application obtain the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the radiation source individual to be identified, wherein the physical signal features are obtained by extracting the physical features of the intermediate frequency signal, and the deep learning signal features are obtained by extracting features based on deep learning on the intermediate frequency signal; the physical signal features and the deep learning signal features are convoluted respectively to obtain the convolution physical signal features and the convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain the fusion signal features, and the identification result of the radiation source individual is obtained based on the fusion signal features. The present application mines the features of the radiation source signal from different dimensions in different ways, and by fusing the physical signal features and the deep learning signal features, the expression accuracy of the fusion signal features can be improved. In addition, by first convoluting the two types of signal features and then fusing them, the balanced distribution of the physical signal features and the deep learning signal features in the fusion process is ensured, thereby further improving the expression accuracy of the signal features. Therefore, when the fusion signal features are used for radiation source individual identification, the accuracy of radiation source individual identification can be significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0019] Figure 1 A schematic diagram of a process for identifying a radiation source individual provided in an embodiment of the present application;

[0020] Figure 2 A schematic diagram of the framework of the radiation source individual identification model provided in the embodiment of the present application;

[0021] Figure 3 A schematic diagram of the framework of the pre-trained network provided in the embodiment of the present application;

[0022] Figure 4 A schematic diagram of a process of acquiring a first intermediate frequency signal sample and a second intermediate frequency signal sample provided in an embodiment of the present application;

[0023] Figure 5A schematic diagram of the structure of a radiation source individual identification device provided in an embodiment of the present application;

[0024] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solution proposed in the embodiment of the present application is applicable to any application scenario of individual identification of radiation sources. The technical solution of the embodiment of the present application can be applied to physical layer security, wireless communication network management, network security, spectrum monitoring, civil aviation safety and space communications.

[0026] Among them, applying individual identification of radiation sources to physical layer security can use subtle differences in the signals emitted by the device to verify the identity of the device, thereby strengthening the authentication mechanism by identifying the device of the legitimate user, ensuring the security of the wireless communication system and preventing unauthorized access.

[0027] Applying individual identification of radiation sources to wireless communication network management can monitor and manage devices in wireless communication networks, such as mobile phone base stations, Wi-Fi routers, etc., to identify illegally accessed devices and ensure stable operation of the network.

[0028] Applying radiation source individual identification to network security can identify and track potential threat sources and protect critical infrastructure from malicious attacks.

[0029] Applying individual identification of radiation sources to spectrum monitoring can identify and monitor spectrum usage, track unauthorized frequency bands, and avoid illegal occupation.

[0030] Applying individual identification of radiation sources to civil aviation safety can identify and eliminate communication equipment that may affect flight safety, ensuring that aircraft are not interfered with by illegal communication equipment.

[0031] Applying individual identification of radiation sources to space communications can use the unique characteristics of satellite transmission signals to identify and track specific satellites in satellite communications, thereby ensuring the security and stability of space communications.

[0032] The technical solution provided in the embodiments of the present application can be exemplarily applied to hardware devices such as processors, electronic devices, servers (including cloud servers), or packaged into software programs to be run. When the hardware device executes the processing of the technical solution of the embodiments of the present application, or the above-mentioned software program is run, the automatic splitting of the target task and the automatic calling of the application program interface required for the task can be achieved to complete the purpose of the target task. The embodiments of the present application only exemplarily introduce the specific processing of the technical solution of the present application, and do not limit the specific implementation form of the technical solution of the present application. Any technical implementation form that can execute the processing of the technical solution of the present application can be adopted by the embodiments of the present application.

[0033] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0034] Before introducing this application solution, the relevant technologies are first introduced:

[0035] Electromagnetic space is an invisible and cross-domain environment that contains a variety of electromagnetic wave signals. Through the detection and analysis of electromagnetic wave signals, the radiation source individual can be identified and applied to multiple fields such as physical layer security and wireless communication network management. Electromagnetic wave signals carry various attribute information of the target, but due to the complexity and uncertainty of the electromagnetic environment, it is difficult to accurately extract useful information from these signals.

[0036] At present, it is difficult to accurately identify and understand information such as target attributes, correlations, and capability systems contained in complex electromagnetic environments. Specifically, it is difficult to accurately identify the target platform behind the electromagnetic wave through single-point signal reception and analysis. This is because the complexity of the signal in the electromagnetic space makes the extraction of the entity target features behind the electromagnetic wave signal vague, which makes it difficult to accurately determine the attributes of the target, that is, the individual identification accuracy of the radiation source is low.

[0037] In view of this, the embodiments of the present application provide a method, device, equipment, medium and product for identifying individual radiation sources, by respectively convolving the physical signal features and deep learning signal features corresponding to the intermediate frequency signals of the individual radiation sources to be identified, and fusing the convolution physical signal features and the convolution deep learning signal features after the convolution processing, and performing individual radiation source identification based on the fused signal features after the fusion, to obtain an identification result of the individual radiation source, thereby improving the accuracy of individual radiation source identification.

[0038] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0039] The present application embodiment first proposes a method for identifying individual radiation sources. Figure 1 As shown, the method includes steps S101 to S103:

[0040] S101. Obtain physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified.

[0041] The physical signal feature is obtained by extracting the physical features of the intermediate frequency signal. The physical signal feature extraction method can be implemented by physical or mathematical methods, for example, by signal processing techniques such as spectrum analysis (such as Fourier transform), statistical analysis or time domain analysis. The physical signal feature includes at least one of envelope feature, fractal feature, power spectrum feature, modal feature, high-order moment feature, and information entropy feature.

[0042] Deep learning signal features are obtained by extracting features from intermediate frequency signals based on deep learning. Deep learning signal features refer to feature representations learned from intermediate frequency signals using deep learning. Deep learning features are often multi-layered and become more abstract as the network layers deepen.

[0043] Physical signal features and deep learning signal features are obtained by extracting features from the intermediate frequency signal from different signal feature extraction angles. Physical signal features reflect the intuitive understanding of the signal and are features with clear physical meanings. Deep learning signal features can capture nonlinear complex relationships from the intermediate frequency signal.

[0044] In some embodiments, the intermediate frequency signal of the individual radiation source to be identified can be input into the radiation source individual identification model, and the radiation source individual identification model can obtain the physical signal characteristics and deep learning signal characteristics corresponding to the intermediate frequency signal of the individual radiation source to be identified.

[0045] The following is a description of the specific implementation process of inputting the intermediate frequency signal of the individual radiation source to be identified into the radiation source individual identification model, and obtaining the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified by the radiation source individual identification model:

[0046] Figure 2 This is a schematic diagram of the framework of the radiation source individual identification model provided in the embodiment of the present application. Figure 2As shown, the radiation source individual identification model includes a deep learning feature extraction module 21 and a physical signal feature extraction module 22.

[0047] The deep learning feature extraction module 21 is used to extract the deep learning signal features of the intermediate frequency signal of the individual radiation source to be identified.

[0048] The physical signal feature extraction module 22 is used to extract the physical signal features of the intermediate frequency signal of the radiation source to be identified.

[0049] Among them, the intermediate frequency signal of the individual radiation source to be identified is input into the individual radiation source identification model, and the individual radiation source identification model obtains the physical signal characteristics and deep learning signal characteristics corresponding to the intermediate frequency signal of the individual radiation source to be identified, including: inputting the intermediate frequency signal of the individual radiation source to be identified into the deep learning feature extraction module 21, and extracting the deep learning signal characteristics of the intermediate frequency signal of the individual radiation source to be identified through the deep learning feature extraction module 21; inputting the intermediate frequency signal of the individual radiation source to be identified into the physical signal feature extraction module 22, and extracting the physical signal characteristics of the intermediate frequency signal of the individual radiation source to be identified through the physical signal feature extraction module 22.

[0050] Continue reading Figure 2 , the deep learning feature extraction module 21 includes a multi-layer convolution module, such as a 6-layer convolution module. The sizes of the 6-layer convolution modules are (512, 1024), (512, 512), (512, 256), (512, 128), (512, 64), and (512, 32) respectively. In other words, the intermediate frequency signal is sequentially processed by each convolution module in the multi-layer convolution module for dimensionality reduction, and a higher level of abstract features is extracted by gradually reducing the size of the features, and finally a deep learning signal feature of the size of (512, 32) is obtained.

[0051] Continue reading Figure 2 The physical signal feature extraction module 22 includes at least one of an envelope feature extraction unit, a fractal feature extraction unit, a power spectrum feature extraction unit, a modal feature extraction unit, a high-order moment feature extraction unit, and an information entropy feature extraction unit.

[0052] The envelope feature extraction unit is used to extract the envelope features of the intermediate frequency signal of the radiation source to be identified. The envelope features are a set of features obtained by analyzing the envelope of the signal. The envelope refers to the change trend of the amplitude of the signal over time. The envelope features can reflect the amplitude change characteristics of the signal and help identify the modulation type and modulation parameters of the signal. The envelope features can be obtained through Hilbert transform or absolute value operation.

[0053] The fractal feature extraction unit is used to extract the fractal features of the intermediate frequency signal of the individual radiation source to be identified. The fractal feature refers to the self-similar nature of the signal in time and frequency. The fractal dimension can be used to describe the complexity of the signal. The fractal feature can reveal the complexity and irregularity of the signal and can be used to distinguish transmitters with different hardware characteristics. The method of extracting the fractal feature includes calculating the fractal dimension of the signal, such as by box counting method or Hurst index.

[0054] The power spectrum feature extraction unit is used to extract the power spectrum features of the intermediate frequency signal of the individual radiation source to be identified. The power spectrum feature is a feature extracted from the power spectrum density of the signal, which reflects the energy distribution of the signal. The power spectrum feature can help identify important parameters such as the bandwidth and center frequency of the signal, and can be used to identify different types of signals. The power spectrum feature can be obtained by using the fast Fourier transform (FFT) to calculate the power spectrum density of the signal and then analyzing the main components of the power spectrum.

[0055] The modal feature extraction unit is used to extract the modal features of the intermediate frequency signal of the individual radiation source to be identified. Modal features refer to the characteristics of different frequency components in the signal, such as the peak and valley values ​​of the frequency response. Modal features can reveal the characteristics of different frequency components in the signal and can be used to identify the modulation mode and hardware characteristics of the signal. Modal features can be obtained by identifying the modal characteristics of the signal through spectrum analysis, such as peak frequency, peak power, etc.

[0056] The high-order moment feature extraction unit is used to extract the high-order moment features of the intermediate frequency signal of the individual radiation source to be identified. The high-order moment feature refers to the high-order statistics of the probability density function of the signal, such as skewness, kurtosis, etc. The high-order moment feature can capture the non-Gaussian characteristics of the signal and can be used to identify nonlinearly modulated signals. The high-order moment feature can be obtained by calculating the skewness (third-order moment) and kurtosis (fourth-order moment) of the signal.

[0057] The information entropy feature extraction unit is used to extract the information entropy feature of the intermediate frequency signal of the individual radiation source to be identified. The information entropy feature refers to the uncertainty measurement of the signal, which is usually used to describe the randomness and complexity of the signal. The information entropy feature can reveal the statistical characteristics of the signal and can be used to identify the random modulation and coding method of the signal. The information entropy feature can be obtained by calculating the probability distribution of the signal and then according to the probability distribution.

[0058] S102. Perform convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features.

[0059] In some embodiments, step S102 includes: performing convolution processing on the physical signal features and the deep learning signal features respectively through the radiation source individual identification model to obtain the convolution physical signal features and the convolution deep learning signal features.

[0060] Continue reading Figure 2 , the radiation source individual identification model includes a feature fusion module 23. Then, the radiation source individual identification model performs convolution processing on the physical signal feature and the deep learning signal feature respectively to obtain the convolution physical signal feature and the convolution deep learning signal feature, including: performing convolution processing on the physical signal feature and the deep learning signal feature respectively through the feature fusion module 23 to obtain the convolution physical signal feature and the convolution deep learning signal feature.

[0061] Continue reading Figure 2 The feature fusion module 23 includes a first convolution module 231, a second convolution module 232 and at least two fully connected layers 233. The first convolution module 231 is used to convolve the physical signal feature to obtain the convolution physical signal feature. The second convolution module 232 is used to convolve the deep learning signal feature to obtain the convolution deep learning signal feature. The convolution kernels of the first convolution module 231 and the second convolution module 232 are 3*3 respectively.

[0062] Among them, the physical signal features and the deep learning signal features are respectively convolved through the feature fusion module 23 to obtain the convolved physical signal features and the convolved deep learning signal features, including: convolving the physical signal features through the first convolution module 231 to obtain the convolved physical signal features; and convolving the deep learning signal features through the second convolution module 232 to obtain the convolved deep learning signal features.

[0063] S103. Fusing the convolution physical signal features and the convolution deep learning signal features to obtain a fused signal feature, and obtaining an identification result of the radiation source individual based on the fused signal feature.

[0064] In some embodiments, step S103 includes: fusing the convolution physical signal features and the convolution deep learning signal features through the radiation source individual identification model to obtain a fused signal feature, and obtaining an identification result of the radiation source individual based on the fused signal feature.

[0065] Among them, the feature fusion module 23 is also used to fuse the convolution physical signal feature and the convolution deep learning signal feature to obtain a fused signal feature. That is to say, the convolution physical signal feature and the convolution deep learning signal feature are fused through the radiation source individual recognition model to obtain a fused signal feature, including: fusing the convolution physical signal feature and the convolution deep learning signal feature through the feature fusion module 23 to obtain a fused signal feature.

[0066] Continue reading Figure 2 , the feature fusion module 23 also includes at least two fully connected layers 233. At least two fully connected layers 233 are used to sequentially perform dimensionality reduction processing and splicing processing on the convolution physical signal features and the convolution deep learning signal features to obtain fused signal features. Then the convolution physical signal features and the convolution deep learning signal features are fused by the feature fusion module 23 to obtain fused signal features, including: sequentially performing dimensionality reduction processing and splicing processing on the convolution physical signal features and the convolution deep learning signal features by at least two fully connected layers 233 to obtain fused signal features.

[0067] The sizes of the output data of at least two fully connected layers 233 are (512, 2048) and (512, 1024) in order from bottom right to top. That is, the convolution physical signal features and the convolution deep learning signal features are spliced ​​and processed into fused signal features with an output size of 1024 by the at least two fully connected layers 233. The splicing process can be performed at any layer of the at least two fully connected layers 233.

[0068] After obtaining the fused signal features, the identification result of the radiation source individual can be obtained based on the fused signal features. Specifically, obtaining the identification result of the radiation source individual based on the fused signal features includes: obtaining the identification result of the radiation source individual based on the fused signal features through a radiation source individual identification model.

[0069] Continue reading Figure 2 The radiation source individual identification model also includes a classification module 24. Obtaining the radiation source individual identification result based on the fusion signal feature through the radiation source individual identification model includes: obtaining the radiation source individual identification result by classifying the fusion signal feature through the classification module.

[0070] The above embodiment introduces the application process of the radiation source individual recognition model. However, before the model can be actually applied, it needs to be trained. Specifically, the radiation source individual recognition model is obtained by training the pre-trained model through the first intermediate frequency signal samples of each radiation source individual and their corresponding annotation information. The annotation information represents the recognition result of the radiation source individual.

[0071] Among them, the network framework structure of the pre-trained model can refer to the above Figure 2 The network framework structure of the radiation source individual identification model introduced in . That is, the pre-training model includes a physical signal feature extraction module, a deep learning feature extraction module, a feature fusion module, and a classification module. The radiation source individual identification model is obtained by training the pre-training model with the first intermediate frequency signal samples of each radiation source individual and its corresponding labeled radiation source individual, including steps a1 to a4:

[0072] Step a1: extracting the first physical signal feature samples corresponding to the first intermediate frequency signal samples of each radiation source individual through a physical signal feature extraction module.

[0073] Step a2: extracting the first deep learning feature samples corresponding to the first intermediate frequency signal samples of each radiation source individual through the deep learning feature extraction module.

[0074] Step a3: Perform feature fusion on the first physical signal feature and the first deep learning feature through a feature fusion module to obtain a first fused signal feature sample.

[0075] Step a4, performing radiation source individual identification on the first fusion signal feature sample through the classification module to obtain a predicted radiation source individual identification result, and based on the difference between the predicted radiation source individual identification result and the labeled radiation source individual, determining the radiation source individual identification loss, and optimizing the pre-trained model based on the radiation source individual identification loss to obtain a radiation source individual identification model.

[0076] The radiation source individual identification loss can be determined by using a cross entropy loss function. The pre-trained model is optimized based on the radiation source individual identification loss to obtain the radiation source individual identification model, including: back propagation based on the radiation source individual identification loss to adjust the network parameters of the physical signal feature extraction module, the deep learning feature extraction module, the feature fusion module, and the classification module, and training is continued until the radiation source individual identification loss is less than the preset loss or the number of iterations reaches the preset number, the training ends, and the pre-trained model at the end of the training is determined as the radiation source individual identification model.

[0077] As mentioned above, the pre-trained model is obtained by training the pre-trained network with the second intermediate frequency signal samples of each radiation source. The specific training process of the pre-trained model will be described in detail below with reference to the accompanying drawings:

[0078] Figure 3 The schematic diagram of the framework of the pre-trained network provided in the embodiment of the present application. Figure 3 As shown, the pre-trained network includes a deep learning feature extraction module 31 to be trained, a physical signal feature extraction module 32 to be trained, a feature fusion module 33 to be trained, a sampling point mask network 34 and a loss function module 35; the pre-trained model is obtained by training through the following steps b1 to b4:

[0079] Step b1, extracting deep learning signal feature samples of the second intermediate frequency signal samples through the deep learning feature extraction module 31 to be trained, and extracting physical signal feature samples of the second intermediate frequency signal samples through the physical signal feature extraction module 32 to be trained.

[0080] Among them, the deep learning feature extraction module 31 to be trained includes a multi-layer convolution module to be trained, such as a 6-layer convolution module. The sizes of the 6-layer convolution modules are (512, 1024), (512, 512), (512, 256), (512, 128), (512, 64), and (512, 32) respectively. In other words, the second intermediate frequency signal sample is sequentially subjected to dimensionality reduction processing by each convolution module in the multi-layer convolution module to be trained, and finally a deep learning signal feature sample of size (512, 32) is obtained.

[0081] And, the physical signal feature extraction module 32 to be trained includes at least one of an envelope feature extraction unit, a fractal feature extraction unit, a power spectrum feature extraction unit, a modal feature extraction unit, a high-order moment feature extraction unit, and an information entropy feature extraction unit. The envelope feature extraction unit, the fractal feature extraction unit, the power spectrum feature extraction unit, the modal feature extraction unit, the high-order moment feature extraction unit, and the information entropy feature extraction unit can be used to extract the envelope feature, fractal feature, power spectrum feature, modal feature, high-order moment feature, and information entropy feature corresponding to the second intermediate frequency signal sample, respectively, so as to obtain the deep learning signal feature sample corresponding to the second intermediate frequency signal sample.

[0082] Step b2: The obtained physical signal feature samples and deep learning signal feature samples are fused through the feature fusion module 33 to be trained to obtain fused signal feature samples.

[0083] Among them, the feature fusion module 33 to be trained includes a first convolution module 331 to be trained, a second convolution module 332 to be trained and at least two fully connected layers 333 to be trained. Among them, the first convolution module 331 to be trained is used to convolve the physical signal feature samples corresponding to the second intermediate frequency signal samples to obtain convolved physical signal feature samples. The second convolution module 332 to be trained is used to convolve the deep learning signal feature samples corresponding to the second intermediate frequency signal samples to obtain convolved deep learning signal feature samples. At least two fully connected layers 233 to be trained are used to perform splicing processing and at least two dimensionality reduction processing on the convolved physical signal feature samples and the convolved deep learning signal feature samples in turn to obtain fused signal feature samples.

[0084] In some embodiments, based on the network structure of the feature fusion module 33 to be trained, step b2 includes: convolving the physical signal features corresponding to the second intermediate frequency signal samples through the first convolution module 331 to be trained to obtain convolved physical signal feature samples; convolving the deep learning signal features corresponding to the second intermediate frequency signal samples through the second convolution module 332 to be trained to obtain convolved deep learning signal feature samples; and sequentially performing splicing processing and at least two dimensionality reduction processing on the convolved physical signal feature samples and the convolved deep learning signal feature samples through at least two fully connected layers 333 to be trained to obtain fused signal feature samples.

[0085] Among them, the sizes of the output data of the at least two fully connected layers 333 to be trained are (512, 2048) and (512, 1024) in the order from bottom right to top. That is to say, the convolution physical signal feature samples and the convolution deep learning signal feature samples are spliced ​​and dimensionally reduced through the at least two fully connected layers 333 to be trained to obtain fused signal feature samples with an output size of (512, 1024). The splicing process can be performed on any layer of the at least two fully connected layers 333 to be trained.

[0086] In some embodiments, the convolution kernels of the first convolution module 331 to be trained and the second convolution module 332 to be trained are 3*3 respectively.

[0087] Step b3, encode the fusion signal feature samples through the sampling point mask network 34 to obtain a target encoding result, and predict the feature vector corresponding to the mask position in the fusion signal feature sample based on the target encoding result to obtain a predicted feature vector; the target encoding result includes the long-time series feature samples and short-time local feature samples of the second intermediate frequency signal sample.

[0088] In some embodiments, continue to refer to Figure 3 The sampling point mask module 34 includes a Conformer module 345 and a mask corresponding sampling point selection layer 346. The Conformer module 345 includes a coding module 3451, a third convolution module 3452 and an output module 3453.

[0089] The encoding module 3451 is used to encode the fusion signal feature samples to obtain linear fusion signal feature samples, that is, initial encoding results. The linear fusion signal feature samples include the long-time series features corresponding to the second intermediate frequency signal samples.

[0090] The third convolution module 3452 is used to perform convolution processing based on the linear fusion signal feature samples, such as 3*3 convolution processing, to capture the local feature samples in the linear fusion signal feature samples, and obtain the convolution fusion signal feature samples, that is, the target coding result. The convolution fusion signal feature samples include not only the long-time series feature samples, but also the short-time local feature samples, which include the energy, frequency, time and space, modulation and other features of the electromagnetic wave pulse.

[0091] Based on the framework structure of the sampling point mask module 34 introduced above, step b3 includes: encoding the sample features of the fused signal through the encoding module 3451 in the Conformer module 345 to obtain an initial encoding result, which includes the long-time series feature samples of the second intermediate frequency sample signal; performing convolution processing on the initial encoding result through the third convolution module 3452 in the Conformer module 345 to obtain a target encoding result, which includes the long-time series feature samples and local feature samples of the second intermediate frequency signal samples; predicting the feature vector corresponding to the mask position in the feature sample of the fused signal based on the target encoding result through the mask corresponding sampling point selection layer 346 to obtain a predicted feature vector.

[0092] The following describes the encoding module 3451, the third convolution module 3452 and the output module 3453 included in the Conformer module 345 in turn:

[0093] In some embodiments, the encoding module 3451 in the Conformer module 345 includes an input layer, a KQV layer, and multiple stacked encoders, each encoder including a feedforward network module and a multi-head attention mechanism module.

[0094] The input layer is used to perform dimensionality reduction processing on the fusion signal feature samples and add position coding to obtain an embedding matrix; wherein the position coding is used to retain the position information in the sequence of the fusion signal feature samples.

[0095] The KQV layer is used to convert the embedding matrix into a key matrix K, a query matrix Q, and a value matrix V, i.e., K = EW K , Q=EW Q , V=EW V , where W K , W Q and W V is the learned matrix; then, the similarity score S between the key matrix K and the query matrix Q is calculated by dot product attention; and the score is converted into attention weight A using the softmax function, and the value V is weighted and summed using the attention weight A to obtain the weighted fusion signal feature sample Z.

[0096] The feedforward network module is used to perform nonlinear changes on the weighted fusion signal feature samples to obtain nonlinearly transformed fusion signal feature samples.

[0097] The multi-head attention mechanism module is used to divide the fused signal feature samples after nonlinear transformation into multiple subspaces (heads), and each head performs KQV calculation independently. The query, key and value of each head are obtained through their respective linear changes. After that, the KQV mechanism is executed on each head separately to obtain multiple weighted feature representations Z i , and concatenate the weighted feature representations of multiple heads, and obtain the final linear fusion signal feature sample through linear transformation, and the linear fusion signal feature sample includes a long-time series feature sample. The long-time series feature sample represents the change trend information of the electromagnetic signal in the second intermediate frequency signal sample.

[0098] The third convolution module 3452 in the Conformer module 345 can be a 3*3 convolution.

[0099] The input shape and output shape of the encoding module 3451 and the third convolution module 3452 are introduced below in conjunction with the table:

[0100] Table 1 Input shape and output shape of encoding module 3451 and third convolution module 3452

[0101]

[0102]

[0103] The input shape represents the size of the input data of each layer, and the output shape represents the size of the output data of each layer. The submodule represents the composition architecture of each layer. The input embedding layer and the position encoding layer in the table correspond to Figure 3 The input layer of the Conformer module in . Each of the 12 stacked encoders corresponds to Figure 3 The output layer is the final output of the Conformer module.

[0104] Each encoder is composed of 2 sub-layer connection structures, wherein the 1st sub-layer connection structure includes 1 multi-head self-attention sub-layer and a normalization layer and 1 residual connection, and the 2nd sub-layer connection structure includes 1 feed-forward fully connected sub-layer and a normalization layer and 1 residual connection. The sub-layer connection structure is input to each sub-layer and the normalization layer process, using residual links (jump connections), and this part of the structure as a whole is called sub-layer connection (representing sub-layers and their link structures). In each encoder layer, there are 2 sub-layers, and these 2 sub-layers plus the surrounding link structures form 2 sub-layer connection structures. The normalization layer is a standard network layer required by all deep network models, because with the increase of the number of network layers, the parameters may begin to appear too large or too small after the calculation of 12 layers, which may cause abnormalities in the learning process, and the model may converge very slowly. Therefore, this application is connected to the normalization layer after a certain number of layers to normalize the numerical value so that its feature value is within a reasonable range.

[0105] In the encoder stage, this application calculates a 256×128 attention matrix. Each value on the attention matrix corresponds to two points in the feature map space output by the backbone network, because the number of tokens is the same as the number of elements in the feature map space. The two points in the feature map space output by the backbone network refer to the two points of the fused signal features in the feature map space. For example, in the task of individual radiation source identification, each point on the feature map (fused signal feature) can be regarded as a token. The attention mechanism generates an attention matrix by calculating the similarity between these tokens (usually through dot products or other methods). Each element Aij in the matrix represents the degree of association between position i and position j on the feature map. Therefore, each value in the attention matrix actually reflects the strength or importance of the interaction between two points on the feature map. This mechanism enables the model to focus on the most relevant features, improving the model's understanding ability and prediction accuracy.

[0106] It should be noted that when the encoder is used in the embodiment of the present application, it plays the role of a decoder. Therefore, the decoder can also be understood as being composed of 12 decoder layers stacked together. The structure of each decoder layer is the same as the encoder structure, and is composed of 2 sub-layer connection structures. The first sub-layer connection structure includes 1 multi-head self-attention sub-layer and a normalization layer and 1 residual connection. The second sub-layer connection structure includes 1 multi-head attention sub-layer and a normalization layer and 1 residual connection. The third sub-layer connection structure includes a feed-forward fully connected sub-layer and a normalization layer and 1 residual connection.

[0107] In some embodiments, the output module 3453 in the Conformer module 345 includes: a feedforward network module, a second dropout layer, a layer normalization module, a GULE activation function layer, a third fully connected layer, a third dropout layer, a fourth fully connected layer and a layer normalization module, which are used to perform the following operations on the convolution fusion signal feature samples in sequence: nonlinear transformation; randomly discarding a portion of neurons to prevent the model from overfitting; standardizing all features of each sample in the same layer to ensure that the feature distribution between layers is relatively consistent; GULE (Generalized U-shaped Leaky ReLU) activation; mapping the received output of the previous layer to a new space; the third dropout layer, randomly discarding a portion of neurons to prevent the model from overfitting; mapping the input features to a larger space and then compressing them back to the original size; normalizing the features within a single sample and the features of the entire batch, which helps to improve the generalization ability and stability of the model.

[0108] After being processed by the output module 3453, the final encoding result corresponding to the fused signal feature sample can be obtained.

[0109] Through the above hierarchical structure, the Conformer module can effectively process the input fusion signal feature samples and capture the long-range dependencies and local features in the second IF signal samples.

[0110] The above embodiment introduces the Conformer module 345 in the sampling point mask module 34 and its function. In practice, the sampling point mask module 34 may also include other processing layers. Figure 3 In some embodiments, the sampling point mask module 34 further includes a prior probability statistics layer 341 , a first dropout layer 342 , a mask selection layer 343 , a normalization layer 344 , and a softmax activation function layer 347 .

[0111] The prior probability statistics layer 341 is used to perform statistics on the feature distribution of the fusion signal feature samples to obtain initial feature distribution statistics results.

[0112] The first dropout layer 342 is used to randomly discard some feature values ​​from the initial feature distribution statistics to obtain the feature distribution statistics to prevent overfitting.

[0113] The mask selection layer 343 is used to select the mask position based on the feature distribution statistics.

[0114] The normalization layer 344 is used to normalize the feature distribution statistical results to obtain normalized feature distribution statistical results.

[0115] The Conformer module 345 is used to encode based on the normalized feature distribution statistics to obtain a target encoding result, which includes the long-time series feature samples and short-time local feature samples of the second intermediate frequency signal sample. Among them, the Conformer module 345 includes multiple Conformer networks, such as 12. Here, it can be understood that the Conformer module 345 is based on the fusion signal feature sample encoded as the target encoding result after being processed in sequence by the prior probability statistics layer 341, the first dropout layer 342, the mask selection layer 343 and the normalization layer 344.

[0116] The mask corresponding sampling point selection layer 346 is used to select the context feature vector corresponding to the mask position from the target encoding result.

[0117] The softmax activation function layer 347 is used to predict the feature vector at the mask position based on the context feature vector to obtain a predicted feature vector.

[0118] After step b3, step b4 is also included.

[0119] b4. The pre-trained network is optimized based on the difference between the predicted feature vector and the original feature vector corresponding to the mask position through the loss function module 35 to obtain a pre-trained model.

[0120] Among them, the loss function module 35 adopts a cross entropy loss function. Specifically, the loss function module 35 determines the feature prediction loss based on the difference between the predicted feature vector and the original feature vector corresponding to the mask position, and performs back propagation based on the feature prediction loss to adjust the network parameters of the pre-trained network, and continues training until the feature prediction loss is less than the preset loss or the number of iterations reaches the preset number, the training ends, and the pre-trained network at the end of the training is determined as the pre-trained model. Among them, the original feature vector corresponding to the mask position can be obtained based on the original sampling module. The original sampling module includes a fourth dropout layer, an original feature vector acquisition layer, a quantization layer, and a softmax activation function layer. The role of the fourth dropout layer can refer to the above introduction to the first dropout layer, the second dropout layer, or the third dropout layer.

[0121] The original feature vector acquisition layer is used to obtain the original feature vector at the mask position.

[0122] The role of the quantization layer is similar to that of layer normalization. For details, please refer to the introduction to layer normalization in the previous content.

[0123] For the softmax activation function layer, please refer to the introduction of the softmax activation function layer in the previous content.

[0124] Through the above training steps, a pre-trained model can be obtained. After obtaining the pre-trained model, a classification module can be added on the basis of the trained deep learning feature extraction module, physical signal feature extraction module, and feature fusion module to train the radiation source individual recognition model. The training process of the radiation source individual recognition model has been introduced in the previous content and will not be repeated here.

[0125] In terms of implementation details, the loss function module 35 includes a linear layer and a logistic regression layer. The linear layer obtains the output of a specified dimension by linearly changing the previous step, that is, the role of converting dimensions. The logistic regression layer scales the numbers in the vector of the last dimension to the probability value range of 0-1 and satisfies that their sum is 1. Through multiple network trainings adjusted by the Loss value, a pre-trained model with fixed parameters is finally obtained.

[0126] In the above embodiment, the pre-trained model is trained in a self-supervised manner, which means that even in the absence of labels, the model can learn to extract implicit features in the data by masking part of the information of the training samples and trying to reconstruct the masked content.

[0127] After completing the training of the pre-trained model, in order to overcome the problem of insufficient data in the domain task, for example, in the short-wave communication radiation source identification system, since the data of mobile targets such as ships and aircraft are difficult to obtain, the model can be further trained on the domain small sample data set based on the pre-trained model using data with weak labels (that is, some samples have the first intermediate frequency signal samples with labeling information).

[0128] In addition, through fine-tuning migration technology, the model can continuously learn the electromagnetic data characteristics in different scenarios and continuously update its parameters to better adapt to electromagnetic target tasks in different environments.

[0129] In the end-side application of target tasks based on electromagnetic signals, considering the limited hardware resources of miniaturized terminal devices, directly transplanting and deploying large models trained on the server side to terminal devices faces many challenges. In view of this, for such end-side application scenarios, an integrated model compression framework can be designed to organically combine network pruning, knowledge distillation (knowledge transfer) and parameter quantization technologies to quantize the radiation source individual identification model. After this series of optimizations, the lightweight radiation source individual identification model can eventually be deployed on small terminal devices with limited resources.

[0130] During the training process, the present application uses physical signal feature samples and deep learning signal feature samples for feature fusion, and encodes based on the fused signal feature samples. The encoding result contains long-time series feature sample information and local feature sample information. The extraction accuracy of long-time series feature sample information and local feature sample information depends on the extraction accuracy of physical signal feature samples and deep learning signal feature samples. In other words, the more accurate the extraction of physical signal feature samples and deep learning signal feature samples, the more accurate the long-time series feature sample information and local feature sample information generated in the subsequent encoding process. Therefore, during the training process, the feature extraction capabilities of the physical signal feature extraction module and the deep learning signal feature extraction module are effectively improved through the long-time series feature sample information and the local feature sample information, which helps to improve the accuracy of individual identification of radiation sources during the application process.

[0131] Figure 4 Schematic diagram of the process of obtaining the first intermediate frequency signal sample and the second intermediate frequency signal sample provided in the embodiment of the present application. Figure 4 As shown, the first intermediate frequency signal sample and the second intermediate frequency signal sample in the above embodiment can be obtained in the following manner:

[0132] The receiver module converts the RF signal received by the antenna into a wideband intermediate frequency signal that is easy to process. In this process, conventional signal processing methods including signal denoising and frequency deviation correction are used.

[0133] Through the intelligent processing terminal, effective narrowband signals are extracted from the broadband intermediate frequency data, and the radiation source properties of these narrowband signals are identified to obtain the attribute information of each individual radiation source, including the individual, model and modulation method of the radiation source.

[0134] The attribute information output by the intelligent processing terminal is summarized and organized through the sample management platform to form a sample data set that can be used for training and verification.

[0135] The sources of the above sample data include but are not limited to shortwave and ultrashortwave data, open source data on the Internet, and data from the testing stations of the Radio Management Committee.

[0136] In summary, the technical solution proposed in this application obtains the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified, wherein the physical signal features are obtained by extracting the physical features of the intermediate frequency signal, and the deep learning signal features are obtained by extracting features based on deep learning on the intermediate frequency signal; the physical signal features and the deep learning signal features are convoluted respectively to obtain convolution physical signal features and convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain fusion signal features, and the identification result of the radiation source individual is obtained based on the fusion signal features. This application can improve the expression accuracy of the fusion signal features by fusing the physical signal features and the deep learning signal features. In addition, by first convoluting the two types of signal features and then fusing them, the balanced distribution of the physical signal features and the deep learning signal features in the fusion process is ensured, thereby further improving the expression accuracy of the signal features. Therefore, when using the fusion signal features to identify the individual radiation source, the accuracy of the individual radiation source identification can be significantly improved.

[0137] Corresponding to the above-mentioned radiation source individual identification method, the present application embodiment also proposes a radiation source individual identification device, such as Figure 5 As shown, the device includes: an acquisition unit 51 and a fusion and identification unit 52; the acquisition unit 51 is used to acquire the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified, the physical signal features are obtained by extracting the physical features of the intermediate frequency signal, and the deep learning signal features are obtained by performing feature extraction based on deep learning on the intermediate frequency signal; the fusion and identification unit 52 is used to perform convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain fused signal features, and the identification result of the individual radiation source is obtained based on the fused signal features.

[0138] In some embodiments, the acquisition unit 51 acquires the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the radiation source individual to be identified; the fusion and identification unit 52 performs convolution processing on the physical signal features and the deep learning signal features respectively to obtain the convolution physical signal features and the convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain the fused signal features, and the identification result of the radiation source individual is obtained based on the fused signal features, which specifically includes: inputting the intermediate frequency signal of the radiation source individual to be identified into the radiation source individual identification model, and the acquisition unit 51 of the radiation source individual identification model acquires the physical signal features and the deep learning signal features corresponding to the intermediate frequency signal of the radiation source individual to be identified. Features; the fusion and identification unit 52 of the radiation source individual identification model performs convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain fused signal features, and the identification results of the radiation source individuals are obtained based on the fused signal features; wherein the radiation source individual identification model is obtained by training the pre-training model with the first intermediate frequency signal samples of each radiation source individual and their corresponding labeled radiation source individuals, and the pre-training model is obtained by training the pre-training network with the long-time series feature samples and short-time local feature samples corresponding to the second intermediate frequency signal samples of each radiation source individual.

[0139] In some embodiments, the acquisition unit 51 includes a deep learning feature extraction module and a physical signal feature extraction module, and the fusion and identification unit 52 includes a feature fusion module and a classification module; wherein the acquisition unit 51 acquires the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the radiation source individual to be identified; the fusion and identification unit 52 performs convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain fused signal features, and the identification result of the radiation source individual is obtained based on the fused signal features. Specifically, it includes: the acquisition unit 51 acquires the deep learning signal features of the intermediate frequency signal of the individual radiation source to be identified through the deep learning feature extraction module, and acquires the physical signal features of the intermediate frequency signal of the individual radiation source to be identified through the physical signal feature extraction module; the fusion and identification unit 52 performs convolution processing on the physical signal features and the deep learning signal features respectively through the feature fusion module to obtain convolution physical signal features and convolution deep learning signal features, and fuses the convolution physical signal features and the convolution deep learning signal features to obtain fused signal features, and classifies the fused signal features through the classification module to obtain the identification result of the individual radiation source.

[0140] In some embodiments, the feature fusion module includes a first convolution module, a second convolution module and at least two fully connected layers; wherein the fusion and recognition unit 52 performs convolution processing on the physical signal feature and the deep learning signal feature respectively through the feature fusion module to obtain a convolution physical signal feature and a convolution deep learning signal feature, and fuses the convolution physical signal feature and the convolution deep learning signal feature to obtain a fused signal feature, specifically including: the fusion and recognition unit 52 convolves the physical signal feature through the first convolution module to obtain a convolution physical signal feature; convolves the deep learning signal feature through the second convolution module to obtain a convolution deep learning signal feature; performs dimensionality reduction processing and splicing processing on the convolution physical signal feature and the convolution deep learning signal feature in turn through the at least two fully connected layers to obtain a fused signal feature.

[0141] In some embodiments, the pre-trained network includes a deep learning feature extraction module to be trained, a physical signal feature extraction module to be trained, a feature fusion module to be trained, a sampling point mask network and a loss function module; the pre-trained model is trained by the following steps: obtaining deep learning signal feature samples of intermediate frequency signal samples through the deep learning feature extraction module to be trained, and obtaining physical signal feature samples of the intermediate frequency signal samples through the physical signal feature extraction module to be trained; fusing the obtained physical signal feature samples and deep learning signal feature samples through the feature fusion module to be trained to obtain fused signal feature samples; encoding the fused signal feature samples through the sampling point mask network to obtain a target encoding result, and predicting the feature vector corresponding to the mask position in the fused signal feature sample based on the target encoding result to obtain a predicted feature vector, wherein the target encoding result includes the long-time series feature samples and short-time local feature samples of the second intermediate frequency signal sample; optimizing the pre-trained network based on the difference between the predicted feature vector and the original feature vector corresponding to the mask position through the loss function module to obtain the pre-trained model.

[0142] In some embodiments, the sampling point mask network includes a Conformer module, and the Conformer module includes an encoding module, a third convolution module and a mask corresponding sampling point selection module; wherein, the sampling point mask network encodes the fused signal feature sample to obtain a target encoding result, and predicts the feature vector corresponding to the mask position in the fused signal feature sample based on the target encoding result to obtain a predicted feature vector, including: encoding the fused signal sample feature through the encoding module to obtain an initial encoding result of the second intermediate frequency signal sample, and the initial encoding result includes a long-time series feature sample of the second intermediate frequency signal sample; convolution processing is performed on the initial encoding result through the third convolution module to obtain a target encoding result, and the target encoding result includes a long-time series feature sample and a local feature sample of the second intermediate frequency signal sample; and the mask corresponding sampling point selection module predicts the feature vector corresponding to the mask position in the fused signal feature sample based on the target encoding result to obtain a predicted feature vector.

[0143] In some embodiments, the physical signal feature includes at least one of an envelope feature, a fractal feature, a power spectrum feature, a modal feature, a high-order moment feature, and an information entropy feature.

[0144] The radiation source individual identification device provided in this embodiment belongs to the same application concept as the radiation source individual identification method provided in the above embodiments of this application, and can execute the radiation source individual identification method provided in any of the above embodiments of this application, and has the corresponding functional modules and beneficial effects of executing the radiation source individual identification method. For the technical details not described in detail in this embodiment, please refer to the specific processing content of the radiation source individual identification method provided in the above embodiments of this application, and will not be repeated here.

[0145] The functions implemented by the various units in the above radiation source individual identification device can be implemented by the same or different processors respectively, and the embodiments of the present application are not limited thereto.

[0146] It should be understood that the units in the above radiation source individual identification device can be implemented in the form of a processor calling software. For example, the device includes a processor, the processor is connected to a memory, and instructions are stored in the memory. The processor calls the instructions stored in the memory to implement any of the above methods or realize the functions of each unit of the device, wherein the processor can be a general-purpose processor, such as a CPU or a microprocessor, etc., and the memory can be a memory in the device or a memory outside the device. Alternatively, the unit in the device can be implemented in the form of a hardware circuit, and the functions of some or all units can be realized by designing the hardware circuit. The hardware circuit can be understood as one or more processors; for example, in one implementation, the hardware circuit is an ASIC, and the functions of some or all of the above units are realized by designing the logical relationship of the components in the circuit; for another example, in another implementation, the hardware circuit can be implemented by PLD, taking FPGA as an example, which can include a large number of logic gate circuits, and the connection relationship between the logic gate circuits is configured by a configuration file, so as to realize the functions of some or all of the above units. All units of the above device can be implemented in the form of a processor calling software, or in the form of a hardware circuit, or in part by a processor calling software, and the remaining part is implemented in the form of a hardware circuit.

[0147] In an embodiment of the present application, a processor is a circuit with the ability to process signals. In one implementation, the processor may be a circuit with the ability to read and run instructions, such as a CPU, a microprocessor, a GPU, or a DSP; in another implementation, the processor may implement certain functions through the logical relationship of a hardware circuit, and the logical relationship of the hardware circuit is fixed or reconfigurable, such as a hardware circuit implemented by an ASIC or PLD, such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document to implement the hardware circuit configuration can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as an NPU, TPU, DPU, etc.

[0148] It can be seen that each unit in the above device can be one or more processors (or processing circuits) configured to implement the above method, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor forms.

[0149] In addition, all or part of the units in the above device can be integrated together, or can be implemented independently. In one implementation, these units are integrated together and implemented in the form of a SOC. The SOC may include at least one processor for implementing any of the above methods or implementing the functions of each unit of the device. The type of the at least one processor may be different, for example, including a CPU and an FPGA, a CPU and an artificial intelligence processor, a CPU and a GPU, etc.

[0150] An embodiment of the present application further provides a control device, which includes a processor and an interface circuit. The processor in the control device is connected to a data input component via the interface circuit of the control device.

[0151] The data input component specifically refers to a functional component capable of inputting or collecting voice data, such as a microphone, etc.

[0152] The above-mentioned interface circuit can be any interface circuit that can realize the data communication function, for example, it can be a USB interface circuit, a Type-C interface circuit, a serial port circuit, a PCIE circuit, etc.

[0153] The processor in the control device is also a circuit with signal processing capability, which executes the radiation source individual identification method introduced in the above embodiment. The specific implementation of the processor can refer to the above processor implementation, and the embodiment of the present application is not strictly limited.

[0154] Another embodiment of the present application also provides an electronic device, see Figure 6 As shown, the device includes:

[0155] Memory 200 and processor 210;

[0156] The memory 200 is connected to the processor 210 and is used to store programs;

[0157] The processor 210 is used to implement the radiation source individual identification method disclosed in any of the above embodiments by running the program stored in the memory 200.

[0158] Specifically, the electronic device may further include: a bus, a communication interface 220 , an input device 230 and an output device 240 .

[0159] The processor 210, the memory 200, the communication interface 220, the input device 230 and the output device 240 are connected to each other via a bus.

[0160] A bus may include a pathway that transfers information between components of a computer system.

[0161] The processor 210 may be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the scheme of the present invention. It may 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 gate or transistor logic devices, discrete hardware components.

[0162] The processor 210 may include a main processor, and may also include a baseband chip, a modem, and the like.

[0163] The memory 200 stores a program for executing the technical solution of the present invention, and may also store an operating system and other key services. Specifically, the program may include a program code, and the program code includes a computer operation instruction. More specifically, the memory 200 may include a read-only memory (ROM), other types of static storage devices that can store static information and instructions, a random access memory (RAM), other types of dynamic storage devices that can store information and instructions, a disk storage, a flash, and the like.

[0164] The input device 230 may include a device for receiving data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer, or a gravity sensor.

[0165] Output device 240 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0166] The communication interface 220 may include any transceiver or the like to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0167] The processor 210 executes the program stored in the memory 200 and calls other devices, which can be used to implement each step of any radiation source individual identification method provided in the above embodiments of the present application.

[0168] An embodiment of the present application also proposes a chip, which includes a processor and a data interface. The processor reads and runs a program stored in a memory through the data interface to execute the radiation source individual identification method introduced in any of the above embodiments. The specific processing process and its beneficial effects can be found in the introduction to the embodiment of the radiation source individual identification method mentioned above.

[0169] In addition to the above methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the radiation source individual identification method described in any of the above embodiments of this specification.

[0170] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0171] In addition, an embodiment of the present application may also be a storage medium on which a computer program is stored, and the computer program is executed by a processor to execute the steps of the radiation source individual identification method described in any of the above embodiments of this specification.

[0172] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0173] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0174] The steps in the methods of each embodiment of the present application can be adjusted in order, combined and deleted according to actual needs, and the technical features recorded in each embodiment can be replaced or combined.

[0175] The modules and sub-modules in the devices and terminals of the various embodiments of the present application can be combined, divided and deleted according to actual needs.

[0176] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the terminal embodiments described above are only schematic, for example, the division of modules or submodules is only a logical function division, and there may be other division methods in actual implementation, for example, multiple submodules or modules can be combined or integrated into another module, 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 an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0177] The modules or submodules described as separate components may or may not be physically separated, and the components of the modules or submodules may or may not be physical modules or submodules, that is, they may be located in one place, or they may be distributed on multiple network modules or submodules. Some or all of the modules or submodules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0178] In addition, each functional module or submodule in each embodiment of the present application may be integrated into one processing module, or each module or submodule may exist physically separately, or two or more modules or submodules may be integrated into one module. The above-mentioned integrated modules or submodules may be implemented in the form of hardware or in the form of software functional modules or submodules.

[0179] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0180] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly by hardware, software units executed by a processor, or a combination of the two. The software units may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0181] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0182] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying an individual radiation source, characterized in that: include: Obtaining physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified, wherein the physical signal features are obtained by extracting the physical features of the intermediate frequency signal, and the deep learning signal features are obtained by performing feature extraction based on deep learning on the intermediate frequency signal; Performing convolution processing on the physical signal feature and the deep learning signal feature respectively to obtain a convolution physical signal feature and a convolution deep learning signal feature; The convolution physical signal feature and the convolution deep learning signal feature are fused to obtain a fused signal feature, and an identification result of the radiation source individual is obtained based on the fused signal feature.

2. The method according to claim 1, characterized in that The step of obtaining physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the radiation source individual to be identified; performing convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; fusing the convolution physical signal features and the convolution deep learning signal features to obtain fusion signal features, and obtaining the identification result of the radiation source individual based on the fusion signal features, includes: Inputting the intermediate frequency signal of the individual radiation source to be identified into the individual radiation source identification model, the individual radiation source identification model obtains the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified; performing convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; fusing the convolution physical signal features and the convolution deep learning signal features to obtain fusion signal features, and obtaining the identification result of the individual radiation source based on the fusion signal features; Among them, the radiation source individual identification model is obtained by training the pre-training model through the first intermediate frequency signal samples of each radiation source individual and their corresponding labeled radiation source individuals, and the pre-training model is obtained by training the pre-training network through the long-time series feature samples and short-time local feature samples corresponding to the second intermediate frequency signal samples of each radiation source individual.

3. The method according to claim 2, characterized in that The radiation source individual identification model includes a deep learning feature extraction module, a physical signal feature extraction module, a feature fusion module and a classification module; Among them, the intermediate frequency signal of the individual radiation source to be identified is input into the individual radiation source identification model, and the individual radiation source identification model obtains the physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified; the physical signal features and the deep learning signal features are convolved respectively to obtain convolution physical signal features and convolution deep learning signal features; the convolution physical signal features and the convolution deep learning signal features are fused to obtain fusion signal features, and the identification result of the individual radiation source is obtained based on the fusion signal features, including: Acquire the deep learning signal features of the intermediate frequency signal of the individual radiation source to be identified through the deep learning feature extraction module; The physical signal feature extraction module is used to obtain the physical signal feature of the intermediate frequency signal of the radiation source to be identified; The feature fusion module is used to convolve the physical signal feature and the deep learning signal feature respectively to obtain a convolutional physical signal feature and a convolutional deep learning signal feature, and the convolutional physical signal feature and the convolutional deep learning signal feature are fused to obtain a fused signal feature. The fusion signal features are classified by the classification module to obtain the identification result of the radiation source individual.

4. The method according to claim 3, characterized in that The feature fusion module includes a first convolution module, a second convolution module and at least two fully connected layers; Among them, the physical signal feature and the deep learning signal feature are respectively convolved by the feature fusion module to obtain a convolution physical signal feature and a convolution deep learning signal feature, and the convolution physical signal feature and the convolution deep learning signal feature are fused to obtain a fused signal feature, including: Convolving the physical signal feature through the first convolution module to obtain a convolved physical signal feature; Convolving the deep learning signal feature through the second convolution module to obtain a convolved deep learning signal feature; The at least two fully connected layers are used to sequentially perform dimensionality reduction processing and splicing processing on the convolutional physical signal features and the convolutional deep learning signal features to obtain fused signal features.

5. The method according to claim 2, characterized in that: The pre-trained network includes a deep learning feature extraction module to be trained, a physical signal feature extraction module to be trained, a feature fusion module to be trained, a sampling point mask network and a loss function module; the pre-trained model is trained by the following steps: Acquire deep learning signal feature samples of intermediate frequency signal samples through the deep learning feature extraction module to be trained, and acquire physical signal feature samples of the intermediate frequency signal samples through the physical signal feature extraction module to be trained; The obtained physical signal feature samples and deep learning signal feature samples are fused by the feature fusion module to be trained to obtain fused signal feature samples; Encoding the fused signal feature sample through the sampling point mask network to obtain a target encoding result, and predicting the feature vector corresponding to the mask position in the fused signal feature sample based on the target encoding result to obtain a predicted feature vector, wherein the target encoding result includes the long-time series feature samples and the short-time local feature samples of the second intermediate frequency signal sample; The pre-trained network is optimized by the loss function module based on the difference between the predicted feature vector and the original feature vector corresponding to the mask position to obtain the pre-trained model.

6. The method according to claim 5, characterized in that The sampling point mask network includes a Conformer module, and the Conformer module includes an encoding module, a third convolution module, and a mask corresponding sampling point selection module; The sampling point mask network encodes the fused signal feature sample to obtain a target encoding result, and predicts a feature vector corresponding to a mask position in the fused signal feature sample based on the target encoding result to obtain a predicted feature vector, including: Encoding the fusion signal sample features through the encoding module to obtain an initial encoding result of the second intermediate frequency signal sample, wherein the initial encoding result includes a long time series feature sample of the second intermediate frequency signal sample; Performing convolution processing on the initial coding result through the third convolution module to obtain a target coding result, wherein the target coding result includes a long-time series feature sample and a local feature sample of the second intermediate frequency signal sample; The mask corresponding sampling point selection module predicts the feature vector corresponding to the mask position in the fusion signal feature sample based on the target encoding result to obtain a predicted feature vector.

7. The method according to any one of claims 1 to 6, characterized in that: The physical signal feature includes at least one of an envelope feature, a fractal feature, a power spectrum feature, a modal feature, a high-order moment feature and an information entropy feature.

8. A radiation source individual identification device, characterized in that: include: An acquisition unit, used to acquire physical signal features and deep learning signal features corresponding to the intermediate frequency signal of the individual radiation source to be identified, wherein the physical signal features are obtained by extracting the physical features of the intermediate frequency signal, and the deep learning signal features are obtained by performing feature extraction based on deep learning on the intermediate frequency signal; The fusion and identification unit is used to perform convolution processing on the physical signal features and the deep learning signal features respectively to obtain convolution physical signal features and convolution deep learning signal features; fuse the convolution physical signal features and the convolution deep learning signal features to obtain fusion signal features, and obtain the identification result of the radiation source individual based on the fusion signal features.

9. An electronic device, characterized in that: including memory and processor; The memory is connected to the processor and is used to store programs; The processor is configured to implement the method according to any one of claims 1 to 7 by running the program in the memory.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

11. A computer program product, characterized in that The method comprises computer program instructions, which, when executed by a processor, cause the processor to implement the method as claimed in any one of claims 1 to 7.

Images