A radiation source identification method and system

Abstract

The present application relates to the field of radiation source identification, and relates to a radiation source identification method and system, the method comprising: acquiring a radiation source signal; preprocessing the radiation source signal to obtain a preprocessed radiation source signal; constructing a time-frequency feature according to the preprocessed radiation source signal to obtain first feature information and second feature information, the first feature information comprising a time domain feature, and the second feature information comprising a frequency domain feature; inputting the first feature information and the second feature information into a KAN constructed by a mixed base function to perform feature mapping to obtain a fused feature vector; and sending the fused feature vector to an identification model for identification to obtain a radiation source identification result. The present application effectively solves the problems of insufficient generalization ability of artificial design feature rules, weak interpretability of deep learning models, and insufficient description of signal frequency domain structure by original KAN networks, and low efficiency caused by large parameter quantity.

Description

Technical Field

[0001] This invention relates to the field of radiation source identification, and more specifically, to a radiation source identification method and system. Background Technology

[0002] With the increasing complexity of electromagnetic environments and the emergence of new signal systems, higher demands are being placed on the accuracy, interpretability, and engineering deployment efficiency of identification methods. Traditional identification methods mainly rely on manually designed feature extraction rules and prior knowledge bases. Their recognition performance is severely limited by expert experience, and their generalization ability and adaptability are insufficient when facing real-world scenarios with complex and variable signal waveforms and subtle individual fingerprint differences. In recent years, end-to-end learning methods, represented by deep learning, have overcome the limitations of traditional methods to some extent and improved recognition accuracy through their powerful automatic feature learning capabilities. However, these models typically suffer from inherent defects such as complex structures, large numbers of parameters, and weak interpretability. Their "black box" nature makes the model's decision-making process difficult to trace and analyze, and also poses challenges for deployment on resource-constrained edge computing platforms. Summary of the Invention

[0003] The purpose of this invention is to provide a radiation source identification method and system to improve the above-mentioned problems.

[0004] To achieve the above objectives, the embodiments of this application provide the following technical solutions:

[0005] On the one hand, embodiments of this application provide a method for identifying radiation sources, the method comprising:

[0006] Acquire radiation source signals;

[0007] The radiation source signal is preprocessed to obtain the preprocessed radiation source signal;

[0008] Based on the preprocessed radiation source signal, time-frequency features are constructed to obtain first feature information and second feature information. The first feature information includes time-domain features, and the second feature information includes frequency-domain features.

[0009] The first feature information and the second feature information are input into the KAN constructed by the mixing basis function for feature mapping to obtain the fused feature vector.

[0010] The fused feature vector is sent to the recognition model for identification to obtain the radiation source identification result.

[0011] Secondly, embodiments of this application provide a radiation source identification system, the system comprising:

[0012] The acquisition module is used to acquire radiation source signals;

[0013] The preprocessing module is used to preprocess the radiation source signal to obtain the preprocessed radiation source signal;

[0014] The construction module is used to construct time-frequency features based on the preprocessed radiation source signal to obtain first feature information and second feature information. The first feature information includes time-domain features, and the second feature information includes frequency-domain features.

[0015] The processing module is used to input the first feature information and the second feature information into the KAN constructed by the mixing basis function for feature mapping to obtain the fused feature vector;

[0016] The identification module is used to send the fused feature vector to the identification model for identification, and obtain the radiation source identification result.

[0017] Thirdly, embodiments of this application provide a radiation source identification device, the device including a memory and a processor. The memory is used to store a computer program; the processor is used to execute the computer program to implement the steps of the radiation source identification method described above.

[0018] Fourthly, embodiments of this application provide a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the radiation source identification method described above.

[0019] The beneficial effects of this invention are as follows:

[0020] This invention improves data quality by preprocessing radiation source signals, then constructs first and second feature information containing time and frequency domain characteristics, providing a more comprehensive basis for identification. The first and second feature information are then subjected to hybrid basis function feature mapping, capturing the transient nonlinearity and global periodicity features of the signal through heterogeneous function bases, achieving deep feature fusion. Finally, the fused feature vector is sent to the recognition model for identification. This invention addresses the problems of insufficient generalization ability of manually designed feature rules and weak interpretability of deep learning models in existing technologies by introducing a KAN network and adaptively improving its insufficient characterization of the signal frequency domain structure and parameter redundancy. While improving the accuracy and generalization ability of radiation source identification, this invention also has the ability to assist in interpreting model behavior and analyzing feature interval responses, providing auxiliary reference for model optimization in practical applications. The optimization method reduces model complexity and computational overhead, making it more suitable for deployment in practical engineering.

[0021] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the radiation source identification method described in an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of the radiation source identification device described in an embodiment of the present invention.

[0025] The diagram is labeled as follows: 800, Radiation source identification device; 801, Processor; 802, Memory; 803, Multimedia component; 804, I / O interface; 805, Communication component. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0027] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0028] Example 1:

[0029] This embodiment provides a radiation source identification method. It can be understood that this embodiment can be used to set up a scenario, such as: in an area with an extremely complex electromagnetic environment, multiple new communication devices are working simultaneously, and the signals are affected by multipath propagation and noise superposition. This causes the air signals to exhibit similar transient characteristics in the time domain and rapid changes or spectral aliasing in the frequency domain. Traditional identification methods based on fixed rules or single deep learning are difficult to reliably extract discriminative features from the signals that have both subtle hardware fingerprint differences and global modulation structures. This results in misjudgment of legitimate devices and underreporting of malicious signals. Furthermore, the model decision-making process is opaque and computationally expensive, making it impossible to achieve real-time and accurate control on edge devices.

[0030] See Figure 1 The figure shows that the method includes steps S1-S5.

[0031] Step S1: Obtain the radiation source signal;

[0032] Step S2: Preprocess the radiation source signal to obtain the preprocessed radiation source signal;

[0033] Step S2 further includes steps S21-S23, which specifically include:

[0034] Step S21: Process the radiation source signal using the moving average method to obtain a first signal;

[0035] Step S22: Perform amplitude normalization processing on the first signal to obtain the second signal;

[0036] In this step, we iterate through all the sampling points of the first signal to find the maximum value of its absolute amplitude, and then divide the value of each sampling point in the first signal by this maximum value to normalize the amplitude.

[0037] Step S23: Use the energy threshold detection algorithm to extract the effective signal segment from the second signal to obtain the preprocessed signal.

[0038] In radiation source identification, signals often exist in the form of burst pulses, with a large number of invalid noise gaps between pulses. Therefore, this step uses an energy threshold detection algorithm to extract the effective signal segments in the second signal and distinguish the noise in the entire long sequence, which not only eliminates a lot of interference but also saves computing resources.

[0039] In this embodiment, the fixed DC bias introduced by the receiver hardware is filtered out by the moving average method. This elimination of incoherent interference prevents the control point resources of the spline basis function in the subsequent KAN network from being used to model such meaningless constant offsets. By normalizing the maximum value mapping, the amplitude of all signals is unified, eliminating the difference in absolute signal strength caused by environmental factors such as propagation path loss. This allows the domain and weights of the spline basis function to respond stably and consistently to the relative shape changes of the signal waveform, rather than its absolute energy. Finally, the pulse effective segments with energy significantly higher than the background noise are accurately located and extracted from the normalized signal, thereby completely eliminating pure noise segments in the signal gaps. This ensures that the data input to the subsequent KAN network consists almost entirely of high contribution intervals, i.e., effective signal segments carrying radiation sources. This allows the limited local fitting ability and parameter capacity of the spline basis function to be fully concentrated on characterizing the subtle features such as transient nonlinear distortions in these key intervals, rather than being scattered and consumed on modeling a large number of noise segments with no information. This fundamentally improves the response probability and modeling accuracy of spline mapping to the high contribution intervals in the signal that truly have discriminative value.

[0040] Step S3: Construct time-frequency features based on the preprocessed radiation source signal to obtain first feature information and second feature information. The first feature information includes time-domain features, and the second feature information includes frequency-domain features.

[0041] In this step, the preprocessed radiation source signal is input into the time domain branch for further processing, preserving the transient characteristics of the radiation source signal on the time axis. Specifically, the preprocessed radiation source signal is expanded into a two-channel real vector. It should be noted that, This represents the complex baseband signal value at the nth sampling point, which includes the amplitude and phase information of the signal; This represents the in-phase component of the signal at the nth sampling point; This represents the orthogonal component of the signal at the nth sampling point; This represents a two-channel real number vector formed by expanding a complex number sequence; express; This represents the value of the in-phase component from the 1st to the Nth sampling point; it also represents the value of the quadrature component from the 1st to the Nth sampling point. This represents the number of sampling points within a single observation window.

[0042] Step S3 further includes steps S31-S33, which specifically include:

[0043] Step S31: Perform a short-time Fourier transform on the preprocessed radiation source signal to obtain the frequency domain signal;

[0044] Step S32: Calculate the time-frequency distribution matrix of the frequency domain signal using a preset sliding window function;

[0045] Step S33: Calculate the magnitude of the time-frequency distribution matrix to obtain the second feature information.

[0046] In this embodiment, a one-dimensional time-domain signal is converted into a two-dimensional time-frequency joint representation by short-time Fourier transform, thereby simultaneously capturing the dynamic law of the signal frequency components changing with time. This solves the limitation of traditional Fourier transform in being unable to handle non-stationary signals. On this basis, a time-frequency distribution matrix is ​​generated by calculating the time-frequency distribution matrix of the frequency domain signal using a preset sliding window function, which preserves the time-frequency energy distribution structure that better reflects the inherent characteristics of the radiation source hardware.

[0047] Step S4: Input the first feature information and the second feature information into the KAN constructed by the mixing basis function for feature mapping to obtain the fused feature vector;

[0048] Step S4 further includes steps S41-S45, which specifically include:

[0049] Step S41: Project and align the first feature information and the second feature information to obtain the time-domain projection feature vector and the frequency-domain projection feature vector;

[0050] In this step, two linear projection matrices are preset to map the first feature information and the second feature information to a space of the same dimension, resulting in a time-domain projection feature vector and a frequency-domain projection feature vector.

[0051] Step S42: Process the temporal projection feature vector using a KAN layer based on B-spline function activation to obtain the third feature information;

[0052] In this step, a B-spline function is used as the activation function for each component of the temporal projection feature vector, using a total of [number missing] activation functions. Cascading architecture This represents the total number of layers in the neural network used in the time-domain B-spline branch, where the nth layer is the nth layer. The layer mapping operator is defined as:

[0053]

[0054] in, This represents the mapping of the input to the output of the l-th layer network; This represents the trainable weight coefficients associated with the j-th B-spline basis function in the l-th layer; Represents B-spline basis functions; For the first The control point vector of the layer spline function. Utilizing the local support properties of B-splines, high-sensitivity fitting is achieved for subtle fingerprint features in time-domain signals, such as transient fluctuations and pulse rising-edge distortion caused by hardware nonlinearity of the radiation source. It should be noted that the B-spline is a cubic spline.

[0055] Step S43: Process the frequency domain projection feature vector using a KAN layer based on Fourier series activation to obtain the fourth feature information;

[0056] In this step, a Fourier series is used as the activation function for each component in the frequency domain projection feature vector, totaling [number] functions. Cascading architecture This represents the total number of layers in the neural network used in the frequency domain Fourier series branch, where the th layer... The layer mapping operator is defined as:

[0057]

[0058] in, Represents the first Layered networks map inputs to outputs; This represents the constant term in a Fourier series; This represents the highest order of the Fourier series expansion; , They represent the first In the layer, corresponding and The First harmonic coefficient; The summation index represents the ergodicity from the 1st fundamental frequency to the Mth harmonic. Utilizing the global orthogonality of trigonometric functions helps to characterize global periodic features such as the signal spectral envelope at shallower layers.

[0059] Step S44: Perform feature interaction fusion processing on the third feature information and the fourth feature information to obtain the fifth feature information;

[0060] In this step, the branch using the B-spline function as the activation function outputs the third feature information, and the branch using the Fourier series as the activation function outputs the fourth feature information. Orthogonal constraints are applied between the parallel branches to ensure that the outputs of the two branches extract non-redundant features, promoting complementary representation of time and frequency domain features. A 1×1 convolutional layer is set before each branch inputs to the KAN module and before feature fusion, without changing the time or frequency dimension, to adjust the dimensionality and reorganize the channels of the input features. The feature interaction fusion of the third and fourth feature information is specifically as follows:

[0061]

[0062] In the above formula, This represents the fifth characteristic information; Indicates third characteristic information; This represents the fourth feature information; This represents the Hadamard product. In radiation source signals, many fingerprints vary with modulation. By adding an interaction term to describe this superposition effect, a deep interaction and fusion of time-domain and frequency-domain features can be achieved.

[0063] Step S45: Send the fifth feature information to the KAN layer based on the hybrid activation function for processing to obtain the fused feature vector.

[0064] In this step, the obtained fifth feature information is input into a KAN layer using a hybrid activation function, meaning each edge has two basis function activations, along with a global bias residual term:

[0065]

[0066] In the above formula, Indicates a hybrid activation function; , These represent the trainable weight coefficients assigned to the B-spline function components and the trainable weight coefficients assigned to the Fourier series components, respectively. Represents a B-spline function; Represents a Fourier series; This represents the trainable weight coefficients assigned to the residual term; Representing the residual term, it is responsible for conveying the basic linear information in the feature vector, reducing the model's sensitivity to parameter initialization in the early stages of training, and improving the stability of model training.

[0067] This embodiment achieves the precise extraction of local nonlinear transient features and global periodic structural features from aligned time-frequency features through projection alignment, heterogeneous basis function mapping, feature interaction fusion, and deep hybrid mapping. The two are then deeply coupled to generate a more discriminative fused feature vector, providing a more complete feature representation for high-precision recognition.

[0068] Step S5: Send the fused feature vector to the recognition model for recognition to obtain the radiation source recognition result.

[0069] Before step S5, steps S51-S53 are also included, which specifically include:

[0070] Step S51: Perform differential smoothing and sparsity processing based on the control point coefficients of the B-spline function to construct differential constraint terms;

[0071] Because B-spline functions are extremely sensitive to waveform details and easily affected by residual noise, resulting in meaningless micro-fluctuations, a difference constraint term is constructed in this step:

[0072]

[0073] in, Indicates the difference constraint term; Indicates the first The first spline edge Coefficients for each control point; Indicates the first The first spline edge Control point coefficients.

[0074] Step S52: Perform weighted difference sparsity processing on the harmonic coefficients of the Fourier series to construct gradient constraint terms;

[0075] In this step, the gradient constraint term constructed is specifically as follows:

[0076]

[0077] In the above formula, Represents the gradient constraint term; Indicates with order The weighting factors for exponential growth, among which Indicates the exponential decay factor. Indicates the order of harmonics; , They represent the first The first Fourier side Harmonic coefficients of the cosine and sine terms; , They represent the first The first Fourier side Harmonic coefficients of the cosine and sine terms.

[0078] Step S53: Construct a joint loss function based on the difference constraint term and the gradient constraint term, and use the joint loss function to train the recognition model.

[0079] In this step, the joint loss function is specifically as follows:

[0080]

[0081] In the above formula, Represents the joint loss function; Represents the cross-entropy classification loss; , All of these are hyperparameters that are greater than zero, used to adjust the relative weights of each constraint in the total loss, in order to achieve the best trade-off between classification accuracy, model smoothness and sparsity. This represents the first-order difference constraint term; This represents the gradient constraint term.

[0082] In this embodiment, a first-order difference constraint term is constructed for the control point coefficients of the time-domain spline branch. This aims to force the B-spline curve to become smoother in non-critical feature regions and suppress minor, meaningless fluctuations caused by noise. For the harmonic coefficients of the frequency-domain Fourier branch, an order-weighted difference constraint term is constructed. This aims to guide spectral energy towards the dominant low-frequency modulation features that contribute more to recognition by penalizing changes in higher-order harmonic components, adaptively truncating redundant high-frequency components. Then, the two sparse constraints specifically designed for the time-frequency branch are combined with the cross-entropy loss, which measures classification performance, to construct a joint loss function. This allows for simultaneous optimization of classification accuracy and parameter efficiency during the backpropagation process of model training. While maintaining the recognition model's ability to jointly model the time-frequency features of the signal, the parameter redundancy and model complexity of the time-domain spline and frequency-domain Fourier branches are effectively reduced, improving training efficiency and parameter utilization efficiency. Ultimately, an optimized recognition model with simpler parameters and easier engineering deployment is obtained.

[0083] Step S5 further includes steps S54-S56, which specifically include:

[0084] Step S54: Calculate the category score based on the fused feature vector to obtain the category score corresponding to each radiation source category;

[0085] In this step, the specific process for calculating the category score is as follows:

[0086]

[0087] In the above formula, Indicates category score; The first feature vector represents the fused feature vector. The component and the first Trainable weight coefficients between scores of each category; The eigenvector representing the fused feature vector is the first... One component; Representing the There are several categories.

[0088] Step S55: Perform decision processing based on the category score corresponding to each radiation source category to obtain the radiation source identification result;

[0089] In this step, the softmax function is used to calculate the membership probability of each category, and the category with the highest probability is output as the radiation source identification result.

[0090] Step S56: Perform model decision-making source tracing analysis based on the feature component that contributes the most in the radiation source identification results to obtain the analysis results.

[0091] Because the KAN structure possesses a certain degree of functional visualization capability, it can provide auxiliary basis for model analysis. For example, the category corresponding to the maximum membership probability is... At that time, take out the first Each component of the class score, i.e., contribution. The top N basis function indices with the highest contribution are obtained by sorting. Based on the position of the basis functions in the function mapping layer, their respective feature channels and corresponding intervals are determined. Combining the linear transformation weights of the front-end 1×1 convolution, linear backtracking analysis is performed on the feature channels to determine their main relevant components in the original time-domain or frequency-domain input. If the basis function belongs to the time-domain branch, its main response time interval range can be estimated based on the linear mapping relationship; if it belongs to the frequency-domain branch, its corresponding frequency interval or spectral component range can be located based on the mapping relationship to assist in analyzing the model's response trend to different feature intervals.

[0092] Example 2:

[0093] This embodiment provides a radiation source identification system, which includes an acquisition module, a preprocessing module, a construction module, a processing module, and an identification module, specifically including:

[0094] The acquisition module is used to acquire radiation source signals;

[0095] The preprocessing module is used to preprocess the radiation source signal to obtain the preprocessed radiation source signal;

[0096] The construction module is used to construct time-frequency features based on the preprocessed radiation source signal to obtain first feature information and second feature information. The first feature information includes time-domain features, and the second feature information includes frequency-domain features.

[0097] The processing module is used to input the first feature information and the second feature information into the KAN constructed by the mixing basis function for feature mapping to obtain the fused feature vector;

[0098] The identification module is used to send the fused feature vector to the identification model for identification, and obtain the radiation source identification result.

[0099] In one specific embodiment of this disclosure, the preprocessing module includes a first processing unit, a second processing unit, and a third processing unit, specifically comprising:

[0100] The first processing unit is used to process the radiation source signal using the moving average method to obtain a first signal;

[0101] The second processing unit is used to perform amplitude normalization processing on the first signal to obtain the second signal;

[0102] The third processing unit is used to extract the effective signal segment from the second signal using an energy threshold detection algorithm to obtain the preprocessed signal.

[0103] In one specific embodiment of this disclosure, the construction module includes a fourth processing unit, a fifth processing unit, and a sixth processing unit, specifically including:

[0104] The fourth processing unit is used to perform a short-time Fourier transform on the preprocessed radiation source signal to obtain a frequency domain signal;

[0105] The fifth processing unit is used to calculate the time-frequency distribution matrix of the frequency domain signal using a preset sliding window function;

[0106] The sixth processing unit is used to calculate the magnitude of the time-frequency distribution matrix to obtain the second feature information.

[0107] In one specific embodiment of this disclosure, the processing module includes a seventh processing unit, an eighth processing unit, a ninth processing unit, a tenth processing unit, and an eleventh processing unit, specifically comprising:

[0108] The seventh processing unit is used to project and align the first feature information and the second feature information to obtain a time-domain projection feature vector and a frequency-domain projection feature vector;

[0109] The eighth processing unit is used to process the temporal projection feature vector using a KAN layer based on Fourier series activation to obtain the third feature information;

[0110] The ninth processing unit is used to process the frequency domain projection feature vector using a KAN layer based on Fourier series activation to obtain the fourth feature information.

[0111] The tenth processing unit is used to perform feature interaction fusion processing on the third feature information and the fourth feature information to obtain the fifth feature information;

[0112] The eleventh processing unit is used to send the fifth feature information to the KAN layer based on the hybrid activation function for processing to obtain the fused feature vector.

[0113] In one specific embodiment of this disclosure, the identification module includes a twelfth processing unit, a thirteenth processing unit, and a fourteenth processing unit, specifically comprising:

[0114] The twelfth processing unit is used to calculate the category score based on the fused feature vector to obtain the category score corresponding to each radiation source category.

[0115] The thirteenth processing unit is used to perform decision processing based on the category score corresponding to each radiation source category to obtain the radiation source identification result.

[0116] The fourteenth processing unit is used to perform model decision-making source tracing analysis based on the feature component that contributes the most in the radiation source identification results, and obtain the analysis results.

[0117] In one specific embodiment of this disclosure, the identification module includes a fifteenth processing unit, a sixteenth processing unit, and a seventeenth processing unit, specifically comprising:

[0118] The fifteenth processing unit is used to perform differential smoothing and sparsity processing based on the control point coefficients of the B-spline function, and to construct differential constraint terms.

[0119] The sixteenth processing unit is used to perform weighted difference sparsity processing based on the harmonic coefficients of the Fourier series and construct gradient constraint terms.

[0120] The seventeenth processing unit is used to construct a joint loss function based on the difference constraint term and the gradient constraint term, and the joint loss function is used to train the recognition model.

[0121] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0122] Example 3:

[0123] Corresponding to the above method embodiments, this embodiment also provides a radiation source identification device. The radiation source identification device described below and the radiation source identification method described above can be referred to each other.

[0124] Figure 2 This is a block diagram illustrating a radiation source identification device 800 according to an exemplary embodiment. (e.g.) Figure 2 As shown, the radiation source identification device 800 may include: a processor 801 and a memory 802. The radiation source identification device 800 may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0125] The processor 801 controls the overall operation of the radiation source identification device 800 to complete all or part of the steps in the radiation source identification method described above. The memory 802 stores various types of data to support the operation of the radiation source identification device 800. This data may include, for example, instructions for any application or method operating on the radiation source identification device 800, and application-related data such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. Multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in memory 802 or transmitted via communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as a keyboard, mouse, buttons, etc. These buttons may be virtual or physical buttons. Communication component 805 is used for wired or wireless communication between the radiation source identification device 800 and other devices. Wireless communication may include Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination of these. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, or an NFC module.

[0126] In an exemplary embodiment, the radiation source identification device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the radiation source identification method described above.

[0127] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the radiation source identification method described above. For example, the computer-readable storage medium may be the memory 802 including the program instructions described above, which may be executed by the processor 801 of the radiation source identification device 800 to complete the radiation source identification method described above.

[0128] Example 4:

[0129] Corresponding to the above method embodiments, this embodiment also provides a readable storage medium. The readable storage medium described below can be referred to in conjunction with the radiation source identification method described above.

[0130] A readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the radiation source identification method described in the above method embodiments.

[0131] Specifically, the readable storage medium can be a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other readable storage medium capable of storing program code.

[0132] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0133] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying radiation sources, characterized in that, include: Acquire radiation source signals; The radiation source signal is preprocessed to obtain the preprocessed radiation source signal; Based on the preprocessed radiation source signal, time-frequency features are constructed to obtain first feature information and second feature information. The first feature information includes time-domain features, and the second feature information includes frequency-domain features. The first feature information and the second feature information are input into the KAN constructed by the mixing basis function for feature mapping to obtain the fused feature vector. The fused feature vector is sent to the recognition model for identification to obtain the radiation source identification result.

2. The radiation source identification method according to claim 1, characterized in that, Preprocessing the radiation source signal includes: The radiation source signal is processed using the moving average method to obtain a first signal; The first signal is normalized to obtain the second signal; The effective signal segment in the second signal is extracted using an energy threshold detection algorithm to obtain the preprocessed signal.

3. The radiation source identification method according to claim 1, characterized in that, Constructing time-frequency features based on the preprocessed radiation source signal includes: The preprocessed radiation source signal is subjected to a short-time Fourier transform to obtain the frequency domain signal; The time-frequency distribution matrix of the frequency domain signal is calculated using a preset sliding window function; The magnitude of the time-frequency distribution matrix is ​​calculated to obtain the second feature information.

4. The radiation source identification method according to claim 1, characterized in that, The first feature information and the second feature information are input into a KAN constructed from a mixed basis function for feature mapping, including: The first feature information and the second feature information are projected and aligned to obtain the time-domain projection feature vector and the frequency-domain projection feature vector. The temporal projection feature vector is processed using a KAN layer based on B-spline function activation to obtain the third feature information; The frequency domain projection feature vector is processed using a KAN layer based on Fourier series activation to obtain the fourth feature information; The third feature information and the fourth feature information are subjected to feature interaction fusion processing to obtain the fifth feature information; The fifth feature information is sent to a KAN layer based on a hybrid activation function for processing to obtain the fused feature vector.

5. The radiation source identification method according to claim 1, characterized in that, The fused feature vector is sent to the recognition model for recognition, including: Based on the fused feature vector, a category score is calculated to obtain the category score corresponding to each radiation source category; Decision processing is performed based on the category score corresponding to each radiation source category to obtain the radiation source identification result; Based on the feature component that contributes the most in the radiation source identification results, model decision-making source tracing analysis is performed to obtain the analysis results.

6. A radiation source identification system, characterized in that, include: The acquisition module is used to acquire radiation source signals; The preprocessing module is used to preprocess the radiation source signal to obtain the preprocessed radiation source signal; The construction module is used to construct time-frequency features based on the preprocessed radiation source signal to obtain first feature information and second feature information. The first feature information includes time-domain features, and the second feature information includes frequency-domain features. The processing module is used to input the first feature information and the second feature information into the KAN constructed by the mixing basis function for feature mapping to obtain the fused feature vector; The identification module is used to send the fused feature vector to the identification model for identification, and obtain the radiation source identification result.

7. The radiation source identification system according to claim 6, characterized in that, The preprocessing module includes: The first processing unit is used to process the radiation source signal using the moving average method to obtain a first signal; The second processing unit is used to perform amplitude normalization processing on the first signal to obtain the second signal; The third processing unit is used to extract the effective signal segment from the second signal using an energy threshold detection algorithm to obtain the preprocessed signal.

8. The radiation source identification system according to claim 6, characterized in that, The building module includes: The fourth processing unit is used to perform a short-time Fourier transform on the preprocessed radiation source signal to obtain a frequency domain signal; The fifth processing unit is used to calculate the time-frequency distribution matrix of the frequency domain signal using a preset sliding window function; The sixth processing unit is used to calculate the magnitude of the time-frequency distribution matrix to obtain the second feature information.

9. The radiation source identification system according to claim 6, characterized in that, The processing module includes: The seventh processing unit is used to project and align the first feature information and the second feature information to obtain a time-domain projection feature vector and a frequency-domain projection feature vector; The eighth processing unit is used to process the temporal projection feature vector using a KAN layer based on B-spline function activation to obtain the third feature information; The ninth processing unit is used to process the frequency domain projection feature vector using a KAN layer based on Fourier series activation to obtain the fourth feature information. The tenth processing unit is used to perform feature interaction fusion processing on the third feature information and the fourth feature information to obtain the fifth feature information; The eleventh processing unit is used to send the fifth feature information to the KAN layer based on the hybrid activation function for processing to obtain the fused feature vector.

10. The radiation source identification system according to claim 6, characterized in that, The identification module includes: The twelfth processing unit is used to calculate the category score based on the fused feature vector to obtain the category score corresponding to each radiation source category. The thirteenth processing unit is used to perform decision processing based on the category score corresponding to each radiation source category to obtain the radiation source identification result. The fourteenth processing unit is used to perform model decision-making source tracing analysis based on the feature component that contributes the most in the radiation source identification results, and obtain the analysis results.

Images