A method and device for identifying individual radiation sources using a small sample

Abstract

The present invention discloses a method and device for identifying a small sample of radiation sources, the method comprising: obtaining a small sample data set of radiation sources; the small sample data set of radiation sources comprises N small sample data information of radiation sources, N being a positive integer; extracting features from the small sample data information of radiation sources to obtain individual feature information of radiation sources; expanding the individual feature information of radiation sources to obtain expanded individual feature information of radiation sources; using the expanded individual feature information of radiation sources to train a preset small sample individual recognition model of radiation sources to obtain an optimized small sample individual recognition model of radiation sources; using the optimized small sample individual recognition model of radiation sources to process the small sample data information of radiation sources to be identified to obtain a small sample data recognition result of radiation sources. The method of the present invention can realize rapid reconnaissance and accurate identification of individual radiation sources, and provide a realistic basis for subsequent decisions such as tracking and interference with individuals.

Description

Technical Field

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

[0002] The individual identification of radiation sources starts with feature extraction. After an in-depth study of the individual characteristics of radiation sources, the transient and steady-state characteristics of the radiation sources are extracted. These characteristics include intra-pulse unintentional modulation characteristics and inter-pulse modulation characteristics. It is generally believed that only individual characteristics that do not change over time can be applied to the individual identification of radiation sources. In practical applications, due to problems such as random startup of radiation source targets, low sidelobe radiation, and scanning working characteristics of interception equipment, the number of intercepted radiation source signal samples that can be used for individual identification is small. Therefore, how to achieve high-performance individual identification of radiation sources under small sample conditions is an urgent need.

[0003] Traditional deep network-based radiation source individual identification generally uses intermediate frequency data as direct input and uses a typical deep convolutional network for network design and training. This method can achieve high recognition performance under large sample conditions, but its performance degrades severely under small sample conditions. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and device for identifying a small sample of radiation sources, so as to achieve high-performance identification of radiation sources under small sample conditions. The present invention selects features that are time-invariant and can reflect the unintentional modulation of the radiation source emission components through the optimization of radiation source features; on the basis of the selected fingerprint features, the sample size is expanded by adding noise to the small sample data; on this basis, a radiation source small sample individual identification model is designed, and the model is trained by a metric learning method based on a distance constraint (Center Loss).

[0005] In order to solve the above technical problems, the first aspect of the embodiments of the present invention discloses a method for identifying a small sample of radiation sources, the method comprising:

[0006] S1, obtaining a small sample data set of radiation sources; the small sample data set of radiation sources includes N small sample data information of radiation sources, where N is a positive integer;

[0007] S2, extracting features from the small sample data information of the radiation source to obtain individual feature information of the radiation source;

[0008] S3, expanding the radiation source individual characteristic information to obtain expanded radiation source individual characteristic information;

[0009] S4, using the expanded radiation source individual characteristic information, training a preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model;

[0010] S5, using the optimized radiation source small sample individual identification model, processing the radiation source small sample data information to be identified, and obtaining the radiation source small sample data identification result.

[0011] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the step of extracting features from the small sample data information of the radiation source to obtain individual feature information of the radiation source includes:

[0012] S21, extracting features from the small sample data information of the radiation source to obtain intra-pulse fingerprint feature information;

[0013] S22, extracting features from the small sample data information of the radiation source to obtain pulse-to-pulse fingerprint feature information;

[0014] S23, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain individual feature information of the radiation source.

[0015] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the step of extracting features from the radiation source small sample data information to obtain intra-vein fingerprint feature information includes:

[0016] S211, performing a first feature extraction on the small sample data information of the radiation source to obtain radiation source envelope feature information;

[0017] S212, performing a second feature extraction on the radiation source small sample data information to obtain the radiation source intra-pulse instantaneous frequency feature information;

[0018] S213, performing a third feature extraction on the radiation source small sample data information to obtain radiation source spectrum feature information;

[0019] S214, performing feature fusion on the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information to obtain intra-pulse fingerprint feature information.

[0020] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the feature fusion of the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information to obtain the intra-pulse fingerprint feature information includes:

[0021] S2141, fusing the radiation source envelope characteristic information and the radiation source intra-pulse instantaneous frequency characteristic information to obtain first intra-pulse fingerprint characteristic information;

[0022] S2142, performing feature fusion on the instantaneous frequency feature information within the radiation source pulse and the spectrum feature information of the radiation source to obtain second intra-pulse fingerprint feature information;

[0023] S2143, performing feature fusion on the radiation source envelope feature information and the radiation source spectrum feature information to obtain third intra-pulse fingerprint feature information;

[0024] S2144, fusing the first intra-pulse fingerprint feature information, the second intra-pulse fingerprint feature information and the third intra-pulse fingerprint feature information to obtain intra-pulse fingerprint feature information.

[0025] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the screening of the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain the radiation source individual feature information includes:

[0026] S231, processing the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain correlation coefficient information;

[0027] S232, aligning the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the correlation coefficient information to obtain first intra-pulse fingerprint feature information and first inter-pulse fingerprint feature information;

[0028] S233, performing vector distance analysis on the first intra-vein fingerprint feature information and the first inter-vein fingerprint feature information to obtain vector distance information;

[0029] S234, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the vector distance information to obtain individual feature information of the radiation source.

[0030] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the expanding the radiation source individual characteristic information to obtain the expanded radiation source individual characteristic information includes:

[0031] S31, filtering the radiation source individual characteristic information to obtain filtered radiation source individual characteristic information;

[0032] S32, calculating the signal-to-noise ratio of the individual characteristic information of the filtered radiation source to obtain signal-to-noise ratio information;

[0033] S33, expanding the radiation source individual characteristic information according to the signal-to-noise ratio information to obtain expanded radiation source individual characteristic information.

[0034] As an optional implementation, in the first aspect of the embodiment of the present invention, the use of the expanded radiation source individual feature information to train a preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model includes:

[0035] S41, presetting a radiation source small sample individual identification model according to the data characteristics of the radiation source small sample data to be identified;

[0036] S42, using a loss function based on a distance constraint, training the preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model;

[0037] The distance constraint-based loss function includes a cross entropy loss function and a center distance loss function.

[0038] A second aspect of an embodiment of the present invention discloses a device for identifying a small sample of radiation sources, the device comprising:

[0039] A data acquisition module is used to acquire a small sample data set of radiation sources; the small sample data set of radiation sources includes N small sample data information of radiation sources, where N is a positive integer;

[0040] A feature extraction module is used to extract features from the small sample data information of the radiation source to obtain individual feature information of the radiation source;

[0041] An information expansion module, used to expand the individual characteristic information of the radiation source to obtain expanded individual characteristic information of the radiation source;

[0042] A model training module is used to train a preset radiation source small sample individual recognition model using the expanded radiation source individual characteristic information to obtain an optimized radiation source small sample individual recognition model;

[0043] The radiation source small sample data identification module is used to process the radiation source small sample data information to be identified using the optimized radiation source small sample individual identification model to obtain the radiation source small sample data identification result.

[0044] As an optional implementation manner, in the second aspect of the embodiment of the present invention, the feature extraction of the small sample data information of the radiation source to obtain the individual feature information of the radiation source includes:

[0045] S21, extracting features from the small sample data information of the radiation source to obtain intra-pulse fingerprint feature information;

[0046] S22, extracting features from the small sample data information of the radiation source to obtain pulse-to-pulse fingerprint feature information;

[0047] S23, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain individual feature information of the radiation source.

[0048] As an optional implementation, in the second aspect of the embodiment of the present invention, the feature extraction of the radiation source small sample data information to obtain the intra-vein fingerprint feature information includes:

[0049] S211, performing a first feature extraction on the small sample data information of the radiation source to obtain radiation source envelope feature information;

[0050] S212, performing a second feature extraction on the radiation source small sample data information to obtain the radiation source intra-pulse instantaneous frequency feature information;

[0051] S213, performing a third feature extraction on the radiation source small sample data information to obtain radiation source spectrum feature information;

[0052] S214, performing feature fusion on the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information to obtain intra-pulse fingerprint feature information.

[0053] As an optional implementation manner, in the second aspect of the embodiment of the present invention, the feature fusion of the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information to obtain the intra-pulse fingerprint feature information includes:

[0054] S2141, fusing the radiation source envelope characteristic information and the radiation source intra-pulse instantaneous frequency characteristic information to obtain first intra-pulse fingerprint characteristic information;

[0055] S2142, performing feature fusion on the instantaneous frequency feature information within the radiation source pulse and the spectrum feature information of the radiation source to obtain second intra-pulse fingerprint feature information;

[0056] S2143, performing feature fusion on the radiation source envelope feature information and the radiation source spectrum feature information to obtain third intra-pulse fingerprint feature information;

[0057] S2144, fusing the first intra-pulse fingerprint feature information, the second intra-pulse fingerprint feature information and the third intra-pulse fingerprint feature information to obtain intra-pulse fingerprint feature information.

[0058] As an optional implementation manner, in the second aspect of the embodiment of the present invention, the screening of the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain the radiation source individual feature information includes:

[0059] S231, processing the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain correlation coefficient information;

[0060] S232, aligning the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the correlation coefficient information to obtain first intra-pulse fingerprint feature information and first inter-pulse fingerprint feature information;

[0061] S233, performing vector distance analysis on the first intra-vein fingerprint feature information and the first inter-vein fingerprint feature information to obtain vector distance information;

[0062] S234, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the vector distance information to obtain individual feature information of the radiation source.

[0063] As an optional implementation manner, in the second aspect of the embodiment of the present invention, the expanding the radiation source individual characteristic information to obtain the expanded radiation source individual characteristic information includes:

[0064] S31, filtering the radiation source individual characteristic information to obtain filtered radiation source individual characteristic information;

[0065] S32, calculating the signal-to-noise ratio of the individual characteristic information of the filtered radiation source to obtain signal-to-noise ratio information;

[0066] S33, expanding the radiation source individual characteristic information according to the signal-to-noise ratio information to obtain expanded radiation source individual characteristic information.

[0067] As an optional implementation, in the second aspect of the embodiment of the present invention, the use of the expanded radiation source individual feature information to train a preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model includes:

[0068] S41, presetting a radiation source small sample individual identification model according to the data characteristics of the radiation source small sample data to be identified;

[0069] S42, using a loss function based on a distance constraint, training the preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model;

[0070] The distance constraint-based loss function includes a cross entropy loss function and a center distance loss function.

[0071] The third aspect of the present invention discloses another device for identifying a small sample of radiation sources, the device comprising:

[0072] A memory storing executable program code;

[0073] a processor coupled to the memory;

[0074] The processor calls the executable program code stored in the memory to execute part or all of the steps in the method for identifying individual radiation source small samples disclosed in the first aspect of the embodiment of the present invention.

[0075] The fourth aspect of the present invention discloses a computer-storable medium, which stores computer instructions. When the computer instructions are called, they are used to execute some or all of the steps in the method for identifying individual radiation sources in a small sample disclosed in the first aspect of an embodiment of the present invention.

[0076] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0077] In view of the problem that traditional neural networks cannot adapt to high-performance recognition under small sample conditions, the present invention adopts three technical means, namely, radiation source individual feature optimization, noise sample expansion, and model training based on distance constraints, to realize the design and implementation of a small sample individual recognition model of radiation sources under small sample conditions. Feature optimization is a part of feature engineering. The present invention adopts the preferred intra-pulse instantaneous frequency feature, which has the characteristic of large inter-class differences. The noise processing for the preferred features expands the number of samples and provides a high-quality sample set. The features extracted by the training of the radiation source small sample individual recognition model based on distance constraints significantly increase the intra-class spacing and reduce the intra-class spacing. Under small sample conditions, a high recognition accuracy rate can still be achieved. After testing, the present invention can reduce the sample demand by 2 / 3 under the condition of eight individual radiation sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0079] Figure 1 It is a flow chart of a method for identifying a small sample of radiation sources disclosed in an embodiment of the present invention;

[0080] Figure 2 It is a flow chart of another method for identifying a small sample of radiation sources disclosed in an embodiment of the present invention;

[0081] Figure 3 is an alternative radiation source fingerprint characteristic diagram disclosed in an embodiment of the present invention;

[0082] Figure 4 It is a preferred instantaneous frequency characteristic diagram of the radiation source pulse disclosed in the embodiment of the present invention;

[0083] Figure 5 It is a comparison diagram of network training based on distance constraint and cross entropy disclosed in an embodiment of the present invention;

[0084] Figure 6It is an individual feature distribution diagram after network training based on distance constraints disclosed in an embodiment of the present invention;

[0085] Figure 7 It is a schematic diagram of the structure of a device for identifying a small sample of radiation sources disclosed in an embodiment of the present invention;

[0086] Figure 8 It is a schematic diagram of the structure of another device for identifying an individual with a small sample of radiation source disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0087] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0088] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or equipment that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or equipment.

[0089] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0090] The present invention discloses a method and device for identifying a small sample of radiation sources, the method comprising: obtaining a small sample data set of radiation sources; the small sample data set of radiation sources comprises N small sample data information of radiation sources, N being a positive integer; extracting features of the small sample data information of radiation sources to obtain individual feature information of radiation sources; expanding the individual feature information of radiation sources to obtain expanded individual feature information of radiation sources; using the expanded individual feature information of radiation sources to train a preset small sample individual recognition model of radiation sources to obtain an optimized small sample individual recognition model of radiation sources; using the optimized small sample individual recognition model of radiation sources to process the small sample data information of radiation sources to be identified, and obtain the recognition result of the small sample data of radiation sources. The method of the present invention can realize the rapid reconnaissance and accurate identification of individual radiation sources, provide a realistic basis for subsequent decisions such as tracking and interference of individuals, and is used in the fields of electronic countermeasures and threat monitoring. The following are detailed descriptions.

[0091] Embodiment 1

[0092] See also Figure 1 , Figure 1 : is a flow chart of a method for identifying a small sample of radiation sources disclosed in an embodiment of the present invention. Figure 1 The described method for identifying individual radiation sources with a small sample size is applied to fields such as electronic countermeasures and threat monitoring, and can realize rapid reconnaissance and accurate identification of individual radiation sources, providing a realistic basis for subsequent decisions such as tracking and interference with individuals, which is not limited in the embodiments of the present invention.

[0093] like Figure 1 As shown, the radiation source small sample individual identification method may include the following operations:

[0094] S1, obtaining a small sample data set of radiation sources; the small sample data set of radiation sources includes N small sample data information of radiation sources, where N is a positive integer;

[0095] S2, extracting features from the small sample data information of the radiation source to obtain individual feature information of the radiation source;

[0096] S3, expanding the radiation source individual characteristic information to obtain expanded radiation source individual characteristic information;

[0097] S4, using the expanded radiation source individual characteristic information, training a preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model;

[0098] S5, using the optimized radiation source small sample individual identification model, processing the radiation source small sample data information to be identified, and obtaining the radiation source small sample data identification result.

[0099] Optionally, extracting features from the small sample data information of the radiation source to obtain individual feature information of the radiation source includes:

[0100] S21, extracting features from the small sample data information of the radiation source to obtain intra-pulse fingerprint feature information;

[0101] S22, extracting features from the small sample data information of the radiation source to obtain pulse-to-pulse fingerprint feature information;

[0102] The pulse-to-pulse fingerprint feature information (repetition interval feature) is an important feature of radar pulse-to-pulse fingerprint, and the calculation method is:

[0103]

[0104] Where N s is the total number of observed pulses, TOA(n) is the arrival time of the nth pulse, and PRI is the inter-pulse fingerprint feature information.

[0105] S23, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain individual feature information of the radiation source.

[0106] Optionally, extracting features from the small sample data information of the radiation source to obtain intra-pulse fingerprint feature information includes:

[0107] S211, performing a first feature extraction on the radiation source small sample data information to obtain radiation source envelope feature information; the specific method is:

[0108] For the small sample data information (intermediate frequency signal) x(n) of the radiation source, the calculation process of its envelope characteristics is as follows:

[0109] First, perform an orthogonal transformation, and the calculation formula is as follows.

[0110] s(n)=hilbert(x(n))

[0111] hilbert means taking the Hilbert transform. For the orthogonal signal s(n), the absolute value can be taken to calculate the radiation source envelope characteristic information Env(n). abs means taking the absolute value:

[0112] Env(n)=abs(s(n))

[0113] S212, performing a second feature extraction on the radiation source small sample data information to obtain the radiation source intra-pulse instantaneous frequency feature information;

[0114] For the orthogonally transformed signal s(n), find its instantaneous phase θ(n).

[0115] θ(n)=atan(s(n))

[0116] atan means to find the inverse tangent. The instantaneous frequency of the signal (instantaneous frequency characteristic information in the radiation source pulse) f(n) can be obtained by the phase difference method:

[0117]

[0118] Where θ(n+1) is the instantaneous phase at time n+1.

[0119] S213, performing a third feature extraction on the radiation source small sample data information to obtain radiation source spectrum feature information;

[0120] For the radiation source small sample data information (intermediate frequency signal) x(n), the radiation source spectrum characteristic information Sp(n) is obtained by the fast Fourier transform method. fft means fast Fourier transform, abs means absolute value, and the calculation formula is as follows:

[0121] Sp(n)=abs(fft(x(n)))

[0122] S214, performing feature fusion on the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information to obtain intra-pulse fingerprint feature information.

[0123] Optionally, feature clustering is performed before feature fusion. The feature clustering is performed using an agglomerative hierarchical clustering method. Agglomerative hierarchical clustering (AHC) has a bottom-up hierarchical tree structure. The algorithm first places all data on the first layer, and each data object is considered as a minimum cluster. Then the similarity between different clusters is calculated, and the two clusters with the largest similarity are merged into a large cluster and placed on the second layer. As the number of layers increases, the number of clusters decreases until all data is merged into one cluster or the number of clusters reaches a preset value.

[0124] Optionally, the step of fusing the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information, and the radiation source spectrum feature information to obtain intra-pulse fingerprint feature information includes:

[0125] S2141, fusing the radiation source envelope characteristic information and the radiation source intra-pulse instantaneous frequency characteristic information to obtain first intra-pulse fingerprint characteristic information;

[0126] S2142, performing feature fusion on the instantaneous frequency feature information within the radiation source pulse and the spectrum feature information of the radiation source to obtain second intra-pulse fingerprint feature information;

[0127] S2143, performing feature fusion on the radiation source envelope feature information and the radiation source spectrum feature information to obtain third intra-pulse fingerprint feature information;

[0128] S2144, fusing the first intra-pulse fingerprint feature information, the second intra-pulse fingerprint feature information and the third intra-pulse fingerprint feature information to obtain intra-pulse fingerprint feature information.

[0129] The steps of feature fusion are:

[0130] The two feature matrices that need to be fused are X and Y, X is an n′×m′ dimensional matrix, Y is a p×m′ dimensional matrix, m′ represents the number of samples, n′ and p represent the dimensions of the two features, and the two matrices are projected to 1D for linear representation, corresponding to the projection vector a respectively 1 and b 1 , then the feature matrix after projection becomes:

[0131]

[0132] The purpose is to maximize the correlation coefficient between X' and Y', so as to obtain the projection vector a when the correlation coefficient is maximum. 1 and b 1 ,Right now:

[0133]

[0134] Before projection, the data is standardized. The purpose of standardization is to make the mean of the data 0 and the variance 1. We can get:

[0135]

[0136]

[0137]

[0138] D(X)=cov(X,X)=E(X T X)

[0139] D(Y)=cov(Y,Y)=E(Y T Y)

[0140] cov(X,Y)=E(XY T ),cov(Y,X)=E(YX T )

[0141] S XX =cov(X,X), then the solution objective is transformed into:

[0142]

[0143] Step 1: Calculate the variance S of X and Y XX and S YY , the covariance S of XY and YX XY =S YX T ;

[0144] Step 2: Calculate the matrix

[0145] Step 3: Solve the singular values ​​of M′ to obtain the largest singular value and its preceding and following singular vectors u, v;

[0146] Step 4: Projection vector a of X and Y 1 and b 1 They are:

[0147]

[0148] Step 5: Fusion vector

[0149] Using the above-mentioned feature fusion steps, the radiation source envelope feature information and the radiation source intra-pulse instantaneous frequency feature information are fused to obtain the first intra-pulse fingerprint feature information; the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information are feature fused to obtain the second intra-pulse fingerprint feature information; the radiation source envelope feature information and the radiation source spectrum feature information are feature fused to obtain the third intra-pulse fingerprint feature information.

[0150] The method of fusing the first intra-pulse fingerprint feature information, the second intra-pulse fingerprint feature information and the third intra-pulse fingerprint feature information to obtain the intra-pulse fingerprint feature information is:

[0151] fusing the first intra-pulse fingerprint feature information and the second intra-pulse fingerprint feature information to obtain first feature information;

[0152] fusing the first intra-pulse fingerprint feature information and the third intra-pulse fingerprint feature information to obtain second feature information;

[0153] The first characteristic information and the second characteristic information are fused to obtain intra-vein fingerprint characteristic information.

[0154] Optionally, the filtering of the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain radiation source individual feature information includes:

[0155] S231, processing the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain correlation coefficient information;

[0156] C ENVi=corr(ENV t ,ENV Li )

[0157]

[0158] C Spi =corr(Sp t ,Sp Li )

[0159] Where ENV t is the radiation source envelope feature information of the tested sample, ENVLi is the radiation source envelope feature information of the training sample, C ENVi is the correlation coefficient between the two. is the instantaneous frequency characteristic information of the radiation source pulse of the tested sample, is the instantaneous frequency characteristic information of the radiation source pulse of the training sample, is the correlation coefficient between the two. t is the radiation source spectrum characteristic information of the tested sample, Sp Li is the radiation source spectrum characteristic information of the training sample, CS pi is the correlation coefficient between the two, corr represents the correlation operation, which is the prior art.

[0160] S232, aligning the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the correlation coefficient information, by first normalizing the radiation source envelope feature information, and then performing time alignment according to the 50% leading edge. The radiation source intra-pulse instantaneous frequency feature information is aligned according to the time alignment method of the radiation source envelope feature information. The radiation source spectrum feature information is aligned according to the normalization method. Obtaining the first intra-pulse fingerprint feature information and the first inter-pulse fingerprint feature information;

[0161] S233, performing vector distance analysis on the first intra-vein fingerprint feature information and the first inter-vein fingerprint feature information to obtain vector distance information;

[0162]

[0163]

[0164]

[0165] The vector distance information includes: radiation source envelope vector distance information R ENVi , the instantaneous frequency vector distance information R of the radiation source pulse fifi , radiation source spectrum feature vector distance information R Spi , N is the length of the vector;

[0166] S234, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the vector distance information to obtain individual feature information of the radiation source.

[0167] A vector distance information threshold is preset, and the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information are screened using the threshold, and the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information exceeding the preset threshold are retained to obtain the individual feature information of the radiation source.

[0168] Optionally, the expanding the radiation source individual characteristic information to obtain expanded radiation source individual characteristic information includes:

[0169] S31, filtering the radiation source individual characteristic information to obtain filtered radiation source individual characteristic information;

[0170] Use filtering to improve the signal-to-noise ratio. The filtering method is:

[0171] Set the initial threshold δ th The target interval is [α, β], where 0≤α<β≤max{S(m,n)}, and S(m,n) is the individual characteristic information of the radiation source;

[0172] Divide the target interval [α, β] into L equal parts;

[0173] The initial threshold δ th Set to α, according to The step size is incremented and iterated to obtain the threshold

[0174] Utilizing Thresholds Calculate the individual characteristic information of the radiation source after filtering at each iteration

[0175]

[0176] Among them, S(m,n) represents the individual characteristic information of the original radiation source, represents the individual characteristic information of the radiation source after filtering, and a and b represent the time and frequency parameters in the threshold;

[0177] Calculate the difference ε between each iteration and the individual characteristic information of the radiation source obtained in the previous iteration i1 ,i 1 =1,2,…,i 1 is the number of iterations;

[0178] The difference Perform curve fitting to obtain curve C. When curve C has an inflection point, the corresponding threshold is the optimal threshold.

[0179] Using the optimal threshold, performing optimal threshold filtering on the original radiation source individual characteristic information S(m,n), filtering the radiation source individual characteristic information;

[0180] S32, calculating the signal-to-noise ratio of the individual characteristic information of the filtered radiation source to obtain signal-to-noise ratio information;

[0181] S33, expanding the radiation source individual characteristic information according to the signal-to-noise ratio information to obtain expanded radiation source individual characteristic information.

[0182] The expansion method is: preset a signal-to-noise ratio threshold, add noise to the signal with a signal-to-noise ratio higher than the preset signal-to-noise ratio threshold, and the added noise is Gaussian white noise. The radiation source samples after noise addition are evenly distributed in the range of 10dB to 40dB, and the individual characteristic information of the expanded radiation source is obtained.

[0183] Optionally, the method for expanding the radiation source individual characteristic information to obtain the expanded radiation source individual characteristic information is:

[0184] The preset data enhancement model is used to expand the individual characteristic information of the radiation source, and the expanded individual characteristic information of the radiation source is constructed. The data enhancement model consists of two modules, namely the generator and the discriminator. The generator inputs random variables. In order to improve the reliability of training and the speed of inference, the generator uses 5 convolutions of different scales to perform feature self-encoding. Among them, convolutions Conv1 and Conv2 are used to implement the feature Encoder, mapping the input data to a low-dimensional feature space; Conv3, Conv4 and Conv5 are used to implement the Decoder, reconstructing the low-dimensional space feature vector into generated data. The discriminator uses a multi-scale CBL module to identify the generated data. The CBL module consists of a convolution layer Conv, a BN batch normalization layer and a Leaky Relu activation function layer. In the CBL module, the generated data and the real data are first convolved to extract features; the extracted features are batch normalized to obtain the features of the specified feature space; after batch normalization, the activation function is used to improve the expressiveness of the model. In order to prevent the problem of gradient disappearance, the Leaky Relu activation function is used in the CBL module:

[0185]

[0186] After feature extraction, the discriminator uses the fully connected layer and the SoftMax activation function to discriminate the generated data. The expression of the SoftMax activation function is as follows:

[0187]

[0188] The loss function used by the discriminator is the Sigmoid loss function:

[0189]

[0190] Optionally, the use of the expanded radiation source individual characteristic information to train a preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model includes:

[0191] S41, presetting a radiation source small sample individual identification model according to the data characteristics of the radiation source small sample data to be identified;

[0192] The preset radiation source small sample individual identification model is the CTPN model;

[0193] The CTPN model includes a VGG16 convolutional model, a bidirectional LSTM model, and a fully connected layer;

[0194] Inputting the expanded radiation source individual feature information into the VGG16 convolutional model; the VGG16 network includes 13 convolutional layers, 5 maximum pooling layers and 3 fully connected layers;

[0195] In the convolution layer, a two-dimensional convolution kernel w∈R is used. 3×3 Extract the feature information of the expanded radiation source individual characteristics to obtain the feature matrix C n ;

[0196]

[0197] Among them, n represents the number of convolution operations, m represents the number of convolution kernels, and p i represents the i-th feature matrix obtained, f represents the nonlinear activation function, · represents the corresponding operation of the shared weight of the convolution kernel and the feature matrix, w represents the weight of the convolution kernel, b represents the bias value, R 3×3 represents a 3×3 real matrix;

[0198] The maximum pooling method is used in the pooling layer, and the feature extraction formula is expressed as follows:

[0199] p u =Max 2×2 [C n ]

[0200] Among them, u represents the number of pooling times, Max 2×2 represents the maximum pooling operation method of a 2×2 matrix, p u is the extracted feature; after multiple convolution and pooling operations, the Reshape processed data stream is input into the bidirectional LSTM model to obtain a feature vector with time series attributes, which is expressed as follows:

[0201] o t =g(V st+V s ' T′ )

[0202]

[0203]

[0204] Among them, s t represents the output of the forward sequence at time t, s t ' represents the output of the reverse sequence at time t, U Xt Represents the initial input of the forward sequence, U′ Xt Represents the initial input of the reverse sequence, Represents the input at the previous moment in the forward sequence. The input of the next moment in reverse time sequence, o t Represents the output at time t. After feature extraction by the time series model, the spatial + sequence feature vector is input into the RPN network and passes through two feature extraction networks. One of them obtains positive and negative feedback classification through the softmax classification feature vector set, and the other is used to calculate the boundary value regression offset for the feature vector set to obtain accurate prediction;

[0205] S42, using a loss function based on a distance constraint, training the preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model;

[0206] The loss function expression based on distance constraint is:

[0207]

[0208] in is the cross entropy loss function, is the center distance loss function, where x i is the i-th sampling value of the measured signal, is the network fitting value; is the feature cluster center value, Loss is the loss function based on distance constraint, m is the dimension of the sampled value of the measured signal, e is a natural constant, λ is a preset hyperparameter, n is the number of sample categories, b yi is the preset bias parameter, To predict the signal, are the preset parameters.

[0209] It can be seen that the present invention aims at the problem that traditional neural networks cannot adapt to high-performance recognition under small sample conditions, and adopts three technical means, namely, radiation source individual feature optimization, noise sample expansion, and model training based on distance constraints, to realize the design and implementation of a small sample individual recognition model of radiation sources under small sample conditions. Feature optimization is part of feature engineering. The present invention adopts the preferred intra-pulse instantaneous frequency feature, which has the characteristic of large inter-class differences. The noise processing for the preferred features expands the number of samples and provides a high-quality sample set. The features extracted by the training of the radiation source small sample individual recognition model based on distance constraints significantly increase the intra-class spacing and reduce the intra-class spacing. Under small sample conditions, a high recognition accuracy can still be achieved. After testing, the present invention can reduce the sample demand by 2 / 3 under the condition of eight individual radiation sources.

[0210] Embodiment 2

[0211] See also Figure 2 , Figure 2 : is a flow chart of another method for identifying a small sample of radiation sources disclosed in an embodiment of the present invention. Figure 2 The described method for identifying individual radiation sources with a small sample size is applied to fields such as electronic countermeasures and threat monitoring, and can realize rapid detection and accurate identification of individual radiation sources, providing a realistic basis for subsequent tracking and interference decisions on individuals, which is not limited in the embodiments of the present invention. Figure 2 As shown, the radiation source small sample individual identification method may include the following operations:

[0212] Step 1: Optimization of individual features of radiation sources. Optimization of individual features is part of feature engineering. It is achieved through the following steps.

[0213] (1) Extract the intrapulse fingerprint features of the radiation source, including the radiation source envelope features, the radiation source intrapulse instantaneous frequency features, and the spectrum features. After feature extraction, normalization is required. The maximum value of the envelope is normalized to 400, and the center value of the instantaneous frequency feature is normalized to 0.

[0214] (2) Extract the inter-pulse fingerprint features of the radiation source and envelop the repetitive interval features of the radiation source.

[0215] (3) Calculate the feature spread technique to estimate the distance between features. The distance is calculated as vector distance. First, calculate the correlation coefficient of two vectors and align the two vectors. Recalculate the vector distance using the aligned vectors.

[0216] (4) Selecting individual features of radiation sources with small intra-class spacing and large inter-class spacing. The present invention selects the intra-pulse instantaneous frequency feature, which has a stable time-invariant difference characteristic for most radiation sources. Figure 3 It is an alternative radiation source fingerprint feature diagram disclosed in the embodiment of the present invention. Figure 4 It is a preferred instantaneous frequency characteristic diagram of the radiation source pulse disclosed in the embodiment of the present invention.

[0217] Step 2: Sample expansion. In order to adapt to the small number of radiation source samples intercepted by reconnaissance equipment and the uneven distribution of the signal-to-noise ratio of the samples, this method adds noise to the high signal-to-noise ratio signal, and the added noise style is Gaussian white noise. The radiation source samples after noise addition are evenly distributed in the range of 10dB to 40dB. When the signal-to-noise ratio of the intercepted signal is low, filtering is first used to improve the signal-to-noise ratio, and then the signal-to-noise ratio is estimated to determine the signal-to-noise ratio after filtering. After the noise addition process, the number of samples is expanded.

[0218] Step 3: Network training based on distance constraints. Traditional network training is based on the cross entropy loss function. The cross entropy loss function can separate data of different classes, but it does not have an explicit aggregation effect on samples of the same class, because the distribution of samples of the same class in the feature space after the softmax classifier may be scattered. Therefore, the present invention has innovated this step and adopted network training based on distance constraints. The network trained with distance constraints can increase the inter-class spacing of individual radiation sources and reduce the spacing within a class. A higher recognition performance can be achieved with a smaller sample. The specific steps are as follows.

[0219] (1) Design a deep convolutional neural network (CNN) based on the input signal length and the number of classifications and recognitions; the deep convolutional neural network can use basic network structures such as VGGNet, AlexNet, InceptionNet and ResNet. Different network structures can be used according to the computing power of the operating platform. The basic deep network architecture used in the present invention is VGG16.

[0220] (2) A distance-constrained loss function is used for network training. The loss function expression is as follows.

[0221]

[0222] in is the cross entropy loss function, is the center distance loss function, where x i is the sample value of the measured signal, is the fitting value of the deep convolutional (CNN) network; is the feature cluster center value.

[0223] (3) The deep neural network is trained. When the gradient of the loss function is less than a fixed threshold, the network training ends. Figure 5 This is a comparison diagram of network training based on distance constraint and cross entropy disclosed in the embodiment of the present invention. After network training based on distance constraint, the distribution of individual characteristics of samples is as follows Figure 6 shown.

[0224] Embodiment 3

[0225] See also Figure 7 , Figure 7 : is a schematic diagram of the structure of a small sample individual identification device for a radiation source disclosed in an embodiment of the present invention. Figure 7 The described radiation source small sample individual identification device is used in the fields of electronic countermeasures, threat monitoring, etc., and can realize rapid reconnaissance and accurate identification of individual radiation sources, providing a realistic basis for subsequent decisions such as tracking and interference of individuals, and the embodiments of the present invention are not limited.

[0226] like Figure 7 As shown, the radiation source small sample individual identification device may include the following operations:

[0227] S301, a data acquisition module, used to acquire a small sample data set of radiation sources; the small sample data set of radiation sources includes N small sample data information of radiation sources, where N is a positive integer;

[0228] S302, a feature extraction module is used to extract features from the small sample data information of the radiation source to obtain individual feature information of the radiation source;

[0229] S303, an information expansion module, used to expand the individual characteristic information of the radiation source to obtain expanded individual characteristic information of the radiation source;

[0230] S304, a model training module, used to train a preset radiation source small sample individual recognition model using the expanded radiation source individual characteristic information to obtain an optimized radiation source small sample individual recognition model;

[0231] S305, a radiation source small sample data identification module is used to process the radiation source small sample data information to be identified using the optimized radiation source small sample individual identification model to obtain a radiation source small sample data identification result.

[0232] Embodiment 4

[0233] See also Figure 8 , Figure 8 Schematic diagram of another device for identifying a small sample of radiation sources according to an embodiment of the present invention. Figure 8 The described radiation source small sample individual identification device is used in the fields of electronic countermeasures, threat monitoring, etc., and can realize rapid reconnaissance and accurate identification of individual radiation sources, providing a realistic basis for subsequent decisions such as tracking and interference of individuals, and the embodiments of the present invention are not limited.

[0234] like Figure 8 As shown, the radiation source small sample individual identification device may include the following operations:

[0235] A memory 401 storing executable program codes;

[0236] a processor 402 coupled to the memory 401;

[0237] The processor 402 calls the executable program code stored in the memory 401 to execute the steps of the radiation source small sample individual identification method described in the first and second embodiments.

[0238] Embodiment 5

[0239] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the method for identifying an individual of a small sample of radiation sources described in Embodiments 1 and 2.

[0240] The device embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0241] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution can be essentially or partly contributed to the prior art in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0242] Finally, it should be noted that the method and device for identifying individual radiation sources with a small sample size disclosed in the embodiments of the present invention disclose only the preferred embodiments of the present invention, and are only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features therein may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for identifying a small sample of radiation sources, characterized in that: The method comprises: S1, obtaining a small sample data set of radiation sources; the small sample data set of radiation sources includes N small sample data information of radiation sources, where N is a positive integer; S2, extracting features from the small sample data information of the radiation source to obtain individual feature information of the radiation source, including: S21, extracting features from the radiation source small sample data information to obtain intra-pulse fingerprint feature information, including: S211, performing a first feature extraction on the small sample data information of the radiation source to obtain radiation source envelope feature information; S212, performing a second feature extraction on the radiation source small sample data information to obtain the radiation source intra-pulse instantaneous frequency feature information; S213, performing a third feature extraction on the radiation source small sample data information to obtain radiation source spectrum feature information; S214, performing feature fusion on the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information to obtain intra-pulse fingerprint feature information; The steps of feature fusion are: The two feature matrices that need to be fused are X and Y, X is an n′×m′ dimensional matrix, Y is a p×m′ dimensional matrix, m′ represents the number of samples, n′ and p represent the dimensions of the two features, and the two matrices are projected to 1 dimension for linear representation, corresponding to the projection vectors a1 and b1 respectively, then the feature matrix after projection becomes: The purpose is to maximize the correlation coefficient between X′ and Y′, so as to obtain the projection vectors a1 and b1 when the correlation coefficient is maximum, that is: Before projection, the data is standardized. The purpose of standardization is to make the mean of the data 0 and the variance 1. We can get: D(X)=cov(X,X)=E(X T X) D(Y)=cov(Y,Y)=E(Y T Y) cov(X,Y)=E(XY T ),cov(Y,X)=E(YX T ) S XX =cov(X,X), then the solution objective is transformed into: Step 1: Calculate the variance S of X and Y XX and S YY , the covariance S of XY and YX XY =S YX T ; Step 2: Calculate the matrix Step 3: Solve the singular values ​​of M′ to obtain the largest singular value and its preceding and following singular vectors u, v; Step 4: The projection vectors a1 and b1 of X and Y are: Step 5: Fusion vector S22, extracting features from the small sample data information of the radiation source to obtain pulse-to-pulse fingerprint feature information; S23, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain radiation source individual feature information, including: S231, processing the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain correlation coefficient information; C ENVi =corr(ENV t ,ENV Li ) C Spi =corr(Sp t ,Sp Li ) Where ENV t ENV is the radiation source envelope characteristic information of the tested sample. Li is the radiation source envelope characteristic information of the training sample, C ENVi is the correlation coefficient between the two; is the instantaneous frequency characteristic information of the radiation source pulse of the tested sample, is the instantaneous frequency characteristic information of the radiation source pulse of the training sample, is the correlation coefficient between the two; Sp t is the radiation source spectrum characteristic information of the tested sample, Sp Li is the radiation source spectrum characteristic information of the training sample, C Spi is the correlation coefficient between the two; S232, aligning the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the correlation coefficient information to obtain first intra-pulse fingerprint feature information and first inter-pulse fingerprint feature information; The alignment method is to first normalize the radiation source envelope characteristic information, and then perform time alignment according to the 50% frontier; the instantaneous frequency characteristic information in the radiation source pulse is aligned according to the time alignment method of the radiation source envelope characteristic information; the radiation source spectrum characteristic information is aligned according to the normalization method; S233, performing vector distance analysis on the first intra-vein fingerprint feature information and the first inter-vein fingerprint feature information to obtain vector distance information; The vector distance information includes: radiation source envelope vector distance information R ENVi , the instantaneous frequency vector distance information of the radiation source pulse Radiation source spectrum characteristic vector distance information R Spi , N is the length of the vector; S234, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the vector distance information to obtain individual feature information of the radiation source; Preset a vector distance information threshold, use the threshold to filter the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information, retain the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information that exceed the preset threshold, and obtain the radiation source individual feature information; S3, expanding the radiation source individual characteristic information to obtain expanded radiation source individual characteristic information, including: S31, filtering the radiation source individual characteristic information to obtain filtered radiation source individual characteristic information; The filtering method is: Set the initial threshold δ th The target interval is [α, β], where 0≤α<β≤max{S(m,n)}, and S(m,n) is the individual characteristic information of the radiation source; Divide the target interval [α, β] into L equal parts; The initial threshold δ th Set to α, according to The step size is incremented and iterated to obtain the threshold Utilizing Thresholds Calculate the individual characteristic information of the radiation source after filtering at each iteration Among them, S(m,n) represents the individual characteristic information of the original radiation source, represents the individual characteristic information of the radiation source after filtering, and a and b represent the time and frequency parameters in the threshold; Calculate the difference between each iteration and the individual characteristic information of the radiation source obtained in the previous iteration i1=1,2,…, i1 is the number of iterations; The difference Perform curve fitting to obtain curve C. When curve C has an inflection point, the corresponding threshold is the optimal threshold. Using the optimal threshold, performing optimal threshold filtering on the original radiation source individual characteristic information S(m,n), filtering the radiation source individual characteristic information; S32, calculating the signal-to-noise ratio of the individual characteristic information of the filtered radiation source to obtain signal-to-noise ratio information; S33, expanding the radiation source individual characteristic information according to the signal-to-noise ratio information to obtain expanded radiation source individual characteristic information; The preset data enhancement model is used to expand the data of the individual characteristic information of the radiation source, and the expanded individual characteristic information of the radiation source is constructed; The data augmentation model consists of two modules, namely the generator and the discriminator. The generator inputs random variables. In order to improve the reliability of training and the speed of inference, the generator uses 5 convolutions of different scales to perform feature self-encoding. Among them, convolutions Conv1 and Conv2 are used to implement the feature encoder, mapping the input data to a low-dimensional feature space. Conv3, Conv4 and Conv5 are used to implement the decoder, reconstructing the low-dimensional space feature vector into generated data. The discriminator uses a multi-scale CBL module to identify the generated data. The CBL module consists of a convolution layer Conv, a BN batch normalization layer and a Leaky Relu activation function layer. In the CBL module, the generated data and the real data are first convolved to extract features. The extracted features are batch normalized to obtain the features of the specified feature space. S4, using the expanded radiation source individual characteristic information, training a preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model; The preset radiation source small sample individual identification model is the CTPN model; The CTPN model includes a VGG16 convolutional model, a bidirectional LSTM model, and a fully connected layer; Inputting the expanded radiation source individual feature information into the VGG16 convolutional model; the VGG16 network includes 13 convolutional layers, 5 maximum pooling layers and 3 fully connected layers; In the convolution layer, a two-dimensional convolution kernel w∈R is used. 3×3 Extract the feature information of the expanded radiation source individual characteristics to obtain the feature matrix C n ; Among them, n represents the number of convolution operations, m represents the number of convolution kernels, and p i represents the i-th feature matrix obtained, f represents the nonlinear activation function, · represents the corresponding operation of the shared weight of the convolution kernel and the feature matrix, w represents the weight of the convolution kernel, b represents the bias value, R 3×3 represents a 3×3 real matrix; The maximum pooling method is used in the pooling layer, and the feature extraction formula is expressed as follows: p u =Max 2×2 [C n ] Among them, u represents the number of pooling times, Max 2×2 represents the maximum pooling operation method of a 2×2 matrix, p u is the extracted feature; after multiple convolution and pooling operations, the Reshape processed data stream is input into the bidirectional LSTM model to obtain a feature vector with time series attributes, which is expressed as: o t =g(V st +V s ′ T′ ) Among them, s t represents the output of the forward sequence at time t, s t ′ represents the output of the reverse sequence at time t, U Xt Represents the initial input of the forward sequence, U′ Xt Represents the initial input of the reverse sequence, Represents the input at the previous moment in the forward sequence, The input of the next moment in reverse time sequence, o t Represents the output at time t. After feature extraction by the time series model, the spatial + sequence feature vector is input into the RPN network and passes through two feature extraction networks. One of them obtains positive and negative feedback classification through the softmax classification feature vector set, and the other is used to calculate the boundary value regression offset for the feature vector set to obtain accurate prediction; S5, using the optimized radiation source small sample individual identification model, processing the radiation source small sample data information to be identified, and obtaining the radiation source small sample data identification result.

2. The method for identifying a small sample of radiation sources according to claim 1, characterized in that: The step of fusing the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information to obtain intra-pulse fingerprint feature information includes: S2141, fusing the radiation source envelope characteristic information and the radiation source intra-pulse instantaneous frequency characteristic information to obtain first intra-pulse fingerprint characteristic information; S2142, performing feature fusion on the instantaneous frequency feature information within the radiation source pulse and the spectrum feature information of the radiation source to obtain second intra-pulse fingerprint feature information; S2143, performing feature fusion on the radiation source envelope feature information and the radiation source spectrum feature information to obtain third intra-pulse fingerprint feature information; S2144, fusing the first intra-pulse fingerprint feature information, the second intra-pulse fingerprint feature information and the third intra-pulse fingerprint feature information to obtain intra-pulse fingerprint feature information.

3. The method for identifying a small sample of radiation sources according to claim 1, characterized in that: The method uses the expanded radiation source individual characteristic information to train a preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model, including: S41, presetting a radiation source small sample individual identification model according to the data characteristics of the radiation source small sample data to be identified; S42, using a loss function based on a distance constraint, training the preset radiation source small sample individual recognition model to obtain an optimized radiation source small sample individual recognition model; The distance constraint-based loss function includes a cross entropy loss function and a center distance loss function.

4. A radiation source small sample individual identification device, characterized in that: The device comprises: A data acquisition module is used to acquire a small sample data set of radiation sources; the small sample data set of radiation sources includes N small sample data information of radiation sources, where N is a positive integer; The feature extraction module is used to extract features from the small sample data information of the radiation source to obtain individual feature information of the radiation source, including: S21, extracting features from the radiation source small sample data information to obtain intra-pulse fingerprint feature information, including: S211, performing a first feature extraction on the small sample data information of the radiation source to obtain radiation source envelope feature information; S212, performing a second feature extraction on the radiation source small sample data information to obtain the radiation source intra-pulse instantaneous frequency feature information; S213, performing a third feature extraction on the radiation source small sample data information to obtain radiation source spectrum feature information; S214, performing feature fusion on the radiation source envelope feature information, the radiation source intra-pulse instantaneous frequency feature information and the radiation source spectrum feature information to obtain intra-pulse fingerprint feature information; The steps of feature fusion are: The two feature matrices that need to be fused are X and Y, X is an n′×m′ dimensional matrix, Y is a p×m′ dimensional matrix, m′ represents the number of samples, n′ and p represent the dimensions of the two features, and the two matrices are projected to 1 dimension for linear representation, corresponding to the projection vectors a1 and b1 respectively, then the feature matrix after projection becomes: The purpose is to maximize the correlation coefficient between X′ and Y′, so as to obtain the projection vectors a1 and b1 when the correlation coefficient is maximum, that is: Before projection, the data is standardized. The purpose of standardization is to make the mean of the data 0 and the variance 1. We can get: D(X)=cov(X,X)=E(X T X) D(Y)=cov(Y,Y)=E(Y T Y) cov(X,Y)=E(XY T ),cov(Y,X)=E(YX T ) S XX =cov(X,X), then the solution objective is transformed into: Step 1: Calculate the variance S of X and Y XX and S YY , the covariance S of XY and YX XY =S YX T ; Step 2: Calculate the matrix Step 3: Solve the singular values ​​of M′ to obtain the largest singular value and its preceding and following singular vectors u, v; Step 4: The projection vectors a1 and b1 of X and Y are: Step 5: Fusion vector S22, extracting features from the small sample data information of the radiation source to obtain pulse-to-pulse fingerprint feature information; S23, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain radiation source individual feature information, including: S231, processing the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information to obtain correlation coefficient information; C ENVi =corr(ENV t ,ENV Li ) C Spi =corr(Sp t ,Sp Li ) Where ENV t ENV is the radiation source envelope characteristic information of the tested sample. Li is the radiation source envelope characteristic information of the training sample, C ENVi is the correlation coefficient between the two; is the instantaneous frequency characteristic information of the radiation source pulse of the tested sample, is the instantaneous frequency characteristic information of the radiation source pulse of the training sample, is the correlation coefficient between the two; Sp t is the radiation source spectrum characteristic information of the tested sample, Sp Li is the radiation source spectrum characteristic information of the training sample, C Spi is the correlation coefficient between the two; S232, aligning the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the correlation coefficient information to obtain first intra-pulse fingerprint feature information and first inter-pulse fingerprint feature information; The alignment method is to first normalize the radiation source envelope characteristic information, and then perform time alignment according to the 50% frontier; the instantaneous frequency characteristic information in the radiation source pulse is aligned according to the time alignment method of the radiation source envelope characteristic information; the radiation source spectrum characteristic information is aligned according to the normalization method; S233, performing vector distance analysis on the first intra-vein fingerprint feature information and the first inter-vein fingerprint feature information to obtain vector distance information; The vector distance information includes: radiation source envelope vector distance information R ENVi , the instantaneous frequency vector distance information of the radiation source pulse Radiation source spectrum characteristic vector distance information R Spi , N is the length of the vector; S234, screening the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information according to the vector distance information to obtain individual feature information of the radiation source; Preset a vector distance information threshold, use the threshold to filter the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information, retain the intra-pulse fingerprint feature information and the inter-pulse fingerprint feature information that exceed the preset threshold, and obtain the radiation source individual feature information; The information expansion module is used to expand the individual characteristic information of the radiation source to obtain expanded individual characteristic information of the radiation source, including: S31, filtering the radiation source individual characteristic information to obtain filtered radiation source individual characteristic information; The filtering method is: Set the initial threshold δ th The target interval is [α, β], where 0≤α<β≤max{S(m,n)}, and S(m,n) is the individual characteristic information of the radiation source; Divide the target interval [α, β] into L equal parts; The initial threshold δ th Set to α, according to The step size is incremented and iterated to obtain the threshold Utilizing Thresholds Calculate the individual characteristic information of the radiation source after filtering at each iteration Among them, S(m,n) represents the individual characteristic information of the original radiation source, represents the individual characteristic information of the radiation source after filtering, and a and b represent the time and frequency parameters in the threshold; Calculate the difference between each iteration and the individual characteristic information of the radiation source obtained in the previous iteration i1=1,2,…, i1 is the number of iterations; The difference Perform curve fitting to obtain curve C. When curve C has an inflection point, the corresponding threshold is the optimal threshold. Using the optimal threshold, performing optimal threshold filtering on the original radiation source individual characteristic information S(m,n), filtering the radiation source individual characteristic information; S32, calculating the signal-to-noise ratio of the individual characteristic information of the filtered radiation source to obtain signal-to-noise ratio information; S33, expanding the radiation source individual characteristic information according to the signal-to-noise ratio information to obtain expanded radiation source individual characteristic information; The preset data enhancement model is used to expand the data of the individual characteristic information of the radiation source, and the expanded individual characteristic information of the radiation source is constructed; The data augmentation model consists of two modules, namely the generator and the discriminator. The generator inputs random variables. In order to improve the reliability of training and the speed of inference, the generator uses 5 convolutions of different scales to perform feature self-encoding. Among them, convolutions Conv1 and Conv2 are used to implement the feature encoder, mapping the input data to a low-dimensional feature space. Conv3, Conv4 and Conv5 are used to implement the decoder, reconstructing the low-dimensional space feature vector into generated data. The discriminator uses a multi-scale CBL module to identify the generated data. The CBL module consists of a convolution layer Conv, a BN batch normalization layer and a Leaky Relu activation function layer. In the CBL module, the generated data and the real data are first convolved to extract features. The extracted features are batch normalized to obtain the features of the specified feature space. A model training module is used to train a preset radiation source small sample individual recognition model using the expanded radiation source individual characteristic information to obtain an optimized radiation source small sample individual recognition model; The preset radiation source small sample individual identification model is the CTPN model; The CTPN model includes a VGG16 convolutional model, a bidirectional LSTM model, and a fully connected layer; Inputting the expanded radiation source individual feature information into the VGG16 convolutional model; the VGG16 network includes 13 convolutional layers, 5 maximum pooling layers and 3 fully connected layers; In the convolution layer, a two-dimensional convolution kernel w∈R is used. 3×3 Extract the feature information of the expanded radiation source individual characteristics to obtain the feature matrix C n ; Among them, n represents the number of convolution operations, m represents the number of convolution kernels, and p i represents the i-th feature matrix obtained, f represents the nonlinear activation function, · represents the corresponding operation of the shared weight of the convolution kernel and the feature matrix, w represents the weight of the convolution kernel, b represents the bias value, R 3×3 represents a 3×3 real matrix; The maximum pooling method is used in the pooling layer, and the feature extraction formula is expressed as follows: p u =Max 2×2 [C n ] Among them, u represents the number of pooling times, Max 2×2 represents the maximum pooling operation method of a 2×2 matrix, p u is the extracted feature; after multiple convolution and pooling operations, the Reshape processed data stream is input into the bidirectional LSTM model to obtain a feature vector with time series attributes, which is expressed as: o t =g(V st +V s ′ T′ ) Among them, s t represents the output of the forward sequence at time t, s t ′ represents the output of the reverse sequence at time t, U Xt Represents the initial input of the forward sequence, U′ Xt Represents the initial input of the reverse sequence, Represents the input at the previous moment in the forward sequence, The input of the next moment in reverse time sequence, o t Represents the output at time t. After feature extraction by the time series model, the spatial + sequence feature vector is input into the RPN network and passes through two feature extraction networks. One of them obtains positive and negative feedback classification through the softmax classification feature vector set, and the other is used to calculate the boundary value regression offset for the feature vector set to obtain accurate prediction; The radiation source small sample data identification module is used to process the radiation source small sample data information to be identified using the optimized radiation source small sample individual identification model to obtain the radiation source small sample data identification result.

5. A radiation source small sample individual identification device, characterized in that: The device comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the method for identifying individual radiation sources in a small sample as described in any one of claims 1-3.

6. A computer storable medium, characterized in that: The computer storable medium stores computer instructions, and when the computer instructions are called, they are used to execute the method for identifying individual radiation source small samples as described in any one of claims 1-3.

Images