A radar emitter identification method and system for unbalanced data

Abstract

The present invention provides a radar emitter identification method and system for imbalanced data, which belongs to the field of signal identification technology, including: obtaining an imbalanced data set of radar emitters, and converting it into a first Wigner-Willie distribution matrix set; determining the maximum Wigner-Willie distribution matrix number and the minimum Wigner-Willie distribution matrix number; oversampling the first Wigner-Willie distribution matrix set and inputting it into a convolutional neural network for the first training; extracting the Wigner-Willie distribution matrix from the first Wigner-Willie distribution matrix set and inputting it into a convolutional neural network after the first training for the second training; obtaining an unknown category of radar emitter data set, and converting it into a second Wigner-Willie distribution matrix set and inputting it into a convolutional neural network after the second training for radar emitter identification. The present invention can accurately realize radar emitter identification.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal recognition, and in particular relates to a radar radiation source recognition method and system for imbalanced data. Background Art

[0002] Radar emitter identification mainly refers to the identification of radar signal carriers through electronic support systems or electronic intelligence reconnaissance systems, providing intelligence support for further inferring the working status, tactical position and threat level of radar emitters. The level of radar emitter identification capability directly determines the level of combat capability of electronic warfare forces.

[0003] The radar radiation source data intercepted by the electronic warfare forces are often unbalanced, that is, some categories have more samples and some categories have fewer samples. If the radiation source samples of various categories cannot be accurately identified based on the unbalanced data, the electronic warfare forces may lose the initiative in electromagnetic space security.

[0004] There are two main methods for the classification and recognition of imbalanced data: resampling and reweighting. Resampling mainly adopts the idea of ​​balancing data to reduce the impact of data imbalance on the recognition algorithm; reweighting mainly adopts the idea of ​​optimizing the recognition algorithm to increase the influence of minority class samples on how to optimize the algorithm parameters. The classification and recognition problem of imbalanced data is essentially a problem caused by data imbalance. Therefore, most technical methods choose resampling methods to solve the classification and recognition problem of imbalanced data. However, most of these methods are applied to the field of image classification and recognition with obvious features, and are still in the exploratory stage in the field of radar emitter recognition.

[0005] At present, radar emitter identification technology has been deeply integrated with deep learning technology. Radar emitter identification methods based on deep learning generally include two stages: radar signal domain transformation and deep learning network training, which can basically solve the radar emitter identification problem in general scenarios. However, when facing the radar emitter identification problem under unbalanced data conditions, more effective and reliable technical means are still needed. Summary of the invention

[0006] One of the purposes of the present invention is to provide a radar emitter identification method for unbalanced data, which can accurately realize radar emitter identification and is suitable for radar emitter identification in various situations such as radar emitter balanced data, radar emitter unbalanced data and a small amount of radar emitter data.

[0007] A second object of the present invention is to provide a radar emitter identification system for unbalanced data.

[0008] In order to achieve one of the above purposes, the present invention adopts the following technical solutions:

[0009] A radar emitter identification method for unbalanced data, the radar emitter identification method comprising:

[0010] Step S1, obtaining an unbalanced data set of radar radiation sources of known categories, and converting it into a first Wigner-Wiley distribution matrix set;

[0011] Step S2, obtaining the Wigner-Willi distribution matrix number corresponding to each category in the first Wigner-Willi distribution matrix set to determine the maximum Wigner-Willi distribution matrix number and the minimum Wigner-Willi distribution matrix number;

[0012] Step S3, oversampling each Wigner-Willi distribution square matrix in the first Wigner-Willi distribution square matrix set according to the maximum Wigner-Willi distribution square matrix number;

[0013] Step S4, inputting each Wigner-Wiley distribution square matrix in the oversampled first Wigner-Wiley distribution square matrix into the convolutional neural network for the first training;

[0014] Step S5, according to the minimum number of Wigner-Wiley distribution matrices, extracting Wigner-Wiley distribution matrices from each category in the first Wigner-Wiley distribution matrix set, and then inputting the extracted Wigner-Wiley distribution matrices into the convolutional neural network after the first training for a second training;

[0015] Step S6: Obtain a radar emitter data set of unknown category, convert it into a second Wigner-Wiley distribution matrix set, and then input it into the convolutional neural network after the second training to perform radar emitter identification.

[0016] Furthermore, in step S1, the specific implementation process of the conversion includes:

[0017] Step S11, performing a Wigner-Wiley domain transformation on each radar radiation source imbalance data in the radar radiation source imbalance data set to obtain a Wigner-Wiley distribution matrix corresponding to each radar radiation source imbalance data;

[0018] Step S12: performing bicubic interpolation and normalization processing on each Wigner-Wiley distribution matrix in turn to obtain a Wigner-Wiley distribution matrix corresponding to the unbalanced data of each radar radiation source.

[0019] Furthermore, in step S3, the specific implementation process of the oversampling includes:

[0020] Step S31, dividing the difference between the maximum Wigner-Willi distribution square matrix number and the Wigner-Willi distribution square matrix number corresponding to each category in the first Wigner-Willi distribution square matrix set by the Wigner-Willi distribution square matrix number corresponding to the corresponding category, so as to determine the number of Wigner-Willi distribution square matrices that need to be supplemented for each Wigner-Willi distribution square matrix corresponding to each category in the first Wigner-Willi distribution square matrix set;

[0021] Step S32: extracting Wigner-Willie distribution square matrices whose number of Wigner-Willie distribution square matrices to be supplemented is less than or equal to the number of Wigner-Willie distribution square matrices of the corresponding category from the first Wigner-Willie distribution square matrix set, and putting them into the first Wigner-Willie distribution square matrix subset, so as to determine all Wigner-Willie distribution square matrices to be supplemented by each Wigner-Willie distribution square matrix corresponding to each category in the first Wigner-Willie distribution square matrix subset;

[0022] Step S33, extracting from the first Wigner-Willie distribution square matrix set the Wigner-Willie distribution square matrix whose number of Wigner-Willie distribution square matrices to be supplemented is greater than the number of Wigner-Willie distribution square matrices of the corresponding category, and putting them into the second Wigner-Willie distribution square matrix subset, so as to determine all the Wigner-Willie distribution square matrices to be supplemented by each Wigner-Willie distribution square matrix corresponding to each category in the second Wigner-Willie distribution square matrix subset;

[0023] Step S34, merge all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset and all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset into the first Wigner-Willi distribution matrix set.

[0024] Furthermore, in the step S32, the specific implementation process of determining all Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset includes:

[0025] Step S321, calculating the first Euclidean distances between each Wigner-Wiley distribution matrix corresponding to each category in the first Wigner-Wiley distribution matrix subset and the remaining Wigner-Wiley distribution matrixes corresponding to the corresponding category, so as to select the Wigner-Wiley distribution matrixes corresponding to the first M smallest first Euclidean distances corresponding to each Wigner-Wiley distribution matrix corresponding to each category;

[0026] Where M is the number of Wigner-Wiley distribution matrices that need to be supplemented for each Wigner-Wiley distribution matrix;

[0027] Step S322: Subtract the corresponding Wigner-Willi distribution matrix from the Wigner-Willi distribution matrices corresponding to the first M smallest first Euclidean distances corresponding to each Wigner-Willi distribution matrix, multiply by a random positive decimal, and then add the corresponding Wigner-Willi distribution matrix to obtain the Wigner-Willi distribution matrix that needs to be supplemented for each Wigner-Willi distribution matrix.

[0028] Furthermore, in the step S33, the specific implementation process of determining all Wigner-Wiley distribution matrices to be supplemented by each Wigner-Wiley distribution matrix corresponding to each category in the second Wigner-Wiley distribution matrix subset includes:

[0029] Step S331, setting the sequence number j of each category of the Wigner-Williams distribution matrix in the second Wigner-Williams distribution matrix subset to 1;

[0030] Step S332, subtract the j-th Wigner-Willi distribution matrix corresponding to the corresponding category from the remaining Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset, multiply by a random positive decimal, and then add the j-th Wigner-Willi distribution matrix corresponding to the corresponding category to obtain the Wigner-Willi distribution matrix that needs to be supplemented for the j-th Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset, and update it to the Wigner-Willi distribution matrix corresponding to the corresponding category in the second Wigner-Willi distribution matrix subset;

[0031] Step S333, let j=j+1, and determine whether j is less than J, if yes, return to step S332; if no, go to step S334;

[0032] Wherein, J is the number of Wigner-Wiley distribution matrices corresponding to each category in the second Wigner-Wiley distribution matrix subset;

[0033] Step S334, calculating the number of Wigner-Willi distribution matrixes that need to be supplemented for each Wigner-Willi distribution matrix corresponding to each category in the updated second Wigner-Willi distribution matrix subset;

[0034] Step S335, determine whether the number of Wigner-Willi distribution matrixes required to be supplemented for each Wigner-Willi distribution matrix corresponding to each category after the update is less than the number of Wigner-Willi distribution matrixes corresponding to the corresponding category after the update; if so, put the updated corresponding category into the first Wigner-Willi distribution matrix subset, and return to step S321; if not, use the updated second Wigner-Willi distribution matrix subset to replace the second Wigner-Willi distribution matrix subset before the update, and return to step S331.

[0035] Furthermore, between step S3 and step S4, the radar emitter identification method further includes:

[0036] Setting one of the Wigner-Wiley distribution square matrices in the oversampled first Wigner-Wiley distribution square matrices as the first Wigner-Wiley distribution square matrices, and setting the remaining Wigner-Wiley distribution square matrices as the second Wigner-Wiley distribution square matrices;

[0037] Calculating the first Euclidean distances between the first Wigner-Wiley distribution matrix and each second Wigner-Wiley distribution matrix to obtain a first Euclidean distance set;

[0038] Selecting second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances from the first Euclidean distance set, and determining whether the categories of the second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances are the same as the categories of the first Wigner-Wiley distribution matrices, if yes, retaining the first Wigner-Wiley distribution matrices; if no, removing the first Wigner-Wiley distribution matrices;

[0039] Wherein, N is a preset value.

[0040] In order to achieve the second objective above, the present invention adopts the following technical solution:

[0041] A radar emitter identification system for imbalanced data, the radar emitter identification system comprising:

[0042] A conversion module, used for acquiring an unbalanced data set of radar radiation sources of known categories and converting it into a first Wigner-Wiley distribution square matrix set;

[0043] An acquisition module is used to acquire the number of Wigner-Willi distribution square matrices corresponding to each category in the first Wigner-Willi distribution square matrix set to determine the maximum number of Wigner-Willi distribution square matrices and the minimum number of Wigner-Willi distribution square matrices;

[0044] an oversampling module, configured to oversample each Wigner-Willi distribution square matrix in the first Wigner-Willi distribution square matrix set according to the maximum number of Wigner-Willi distribution square matrixes;

[0045] A first training module is used to input each Wigner-Wiley distribution square matrix in the oversampled first Wigner-Wiley distribution square matrix set into the convolutional neural network for a first training;

[0046] A second training module is used to extract Wigner-Wiley distribution matrices from each category in the first Wigner-Wiley distribution matrix set according to the minimum number of Wigner-Wiley distribution matrices, and then input the extracted Wigner-Wiley distribution matrices into the convolutional neural network after the first training for a second training;

[0047] The recognition module is used to obtain a radar emitter data set of unknown categories, convert it into a second Wigner-Wiley distribution matrix set, and then input it into a convolutional neural network that has been trained for the second time to perform radar emitter recognition.

[0048] Furthermore, the conversion module includes:

[0049] A Wigner-Wiley domain transformation submodule is used to perform a Wigner-Wiley domain transformation on each radar radiation source imbalance data in the radar radiation source imbalance data set to obtain a Wigner-Wiley distribution matrix corresponding to each radar radiation source imbalance data;

[0050] The normalization processing submodule is used to perform bicubic interpolation and normalization processing on each Wigner-Wiley distribution matrix in turn to obtain the Wigner-Wiley distribution matrix corresponding to the unbalanced data of each radar radiation source.

[0051] Furthermore, the oversampling module includes:

[0052] a processing submodule, configured to divide the difference between the maximum number of Wigner-Willi distribution square matrices and the number of Wigner-Willi distribution square matrices corresponding to each category in the first Wigner-Willi distribution square matrix set by the number of Wigner-Willi distribution square matrices corresponding to the corresponding category, so as to determine the number of Wigner-Willi distribution square matrices that need to be supplemented for each Wigner-Willi distribution square matrix corresponding to each category in the first Wigner-Willi distribution square matrix set;

[0053] A first extraction submodule is used to extract, from the first Wigner-Willi distribution square matrix set, Wigner-Willi distribution square matrices whose number of Wigner-Willi distribution square matrices to be supplemented is less than or equal to the number of Wigner-Willi distribution square matrices of the corresponding category, and put them into the first Wigner-Willi distribution square matrix subset to determine all Wigner-Willi distribution square matrices to be supplemented by each Wigner-Willi distribution square matrix corresponding to each category in the first Wigner-Willi distribution square matrix subset;

[0054] A second extraction submodule is used to extract, from the first Wigner-Wiley distribution square matrix set, the Wigner-Wiley distribution square matrix whose number of Wigner-Wiley distribution square matrixes to be supplemented is greater than the number of Wigner-Wiley distribution square matrixes of the corresponding category, and put them into the second Wigner-Wiley distribution square matrix subset to determine all the Wigner-Wiley distribution square matrices to be supplemented by each Wigner-Wiley distribution square matrix corresponding to each category in the second Wigner-Wiley distribution square matrix subset;

[0055] A merging and extracting submodule is used to merge all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset and all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset into the first Wigner-Willi distribution matrix set.

[0056] Furthermore, the radar emitter identification system further includes:

[0057] A setting module, used for setting one of the Wigner-Wiley distribution square matrices in the oversampled first Wigner-Wiley distribution square matrices as the first Wigner-Wiley distribution square matrices, and setting the remaining Wigner-Wiley distribution square matrices as the second Wigner-Wiley distribution square matrices;

[0058] A calculation module, used for calculating the first Euclidean distances between the first Wigner-Wiley distribution matrix and each second Wigner-Wiley distribution matrix, to obtain a first Euclidean distance set;

[0059] a judgment module, used for selecting second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances from the first Euclidean distance set, and judging whether the categories of the second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances are the same as the categories of the first Wigner-Wiley distribution matrices, if yes, retaining the first Wigner-Wiley distribution matrices; if no, removing the first Wigner-Wiley distribution matrices;

[0060] Wherein, N is a preset value.

[0061] In summary, the technical effects of the technical solution of the present invention are:

[0062] The present invention converts an unbalanced data set of radar radiation sources into a first Wigner-Wiley distribution square matrix set to determine a maximum Wigner-Wiley distribution square matrix number and a minimum Wigner-Wiley distribution square matrix number; uses the maximum Wigner-Wiley distribution square matrix number to obtain a balanced data set of radar radiation sources after oversampling the first Wigner-Wiley distribution square matrix set to implement the first training in a convolutional neural network; uses the minimum Wigner-Wiley distribution square matrix number to obtain a balanced data set of radar radiation sources after extracting Wigner-Wiley distribution square matrices of each category in the first Wigner-Wiley distribution square matrix set to implement the second training in the convolutional neural network; finally, uses a radar radiation source data set of unknown category to implement radar radiation source recognition through the convolutional neural network trained for the second time, and the recognition accuracy is high. The present invention is suitable for radar radiation source recognition of a balanced data set of radar radiation sources, radar radiation source recognition of an unbalanced data set of radar radiation sources, and radar radiation source recognition of a small amount of radar radiation source data. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 It is a schematic diagram of the flow chart of the radar emitter identification method for unbalanced data of the present invention;

[0065] Figure 2 Schematic diagram of the convolutional neural network structure of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 making creative work are within the scope of protection of the present invention.

[0067] This embodiment provides a radar emitter identification method for unbalanced data. Figure 1 , the radar radiation source identification method comprises:

[0068] S1. Obtain an unbalanced data set of radar radiation sources of known categories and convert it into a first Wigner-Wiley distribution matrix set.

[0069] In this embodiment, when the number of radar emitter data of each category differs greatly (i.e., the difference between the maximum number of radar emitter data and the minimum number of radar emitter data is greater than the threshold), it is regarded as an unbalanced radar emitter data set. When the number of radar emitter data of each category is similar (i.e., the difference between the maximum number of radar emitter data and the minimum number of radar emitter data is less than the threshold), it is a balanced radar emitter data set.

[0070] In this embodiment, the unbalanced data of each radar radiation source of known categories are sequentially subjected to Wigner-Wiley domain transformation, bicubic interpolation and normalization processing to obtain a standard Wigner-Wiley distribution matrix with uniform dimension size. The specific implementation process of the conversion includes:

[0071] Step S11, performing a Wigner-Wiley domain transformation on each radar radiation source imbalance data in the radar radiation source imbalance data set to obtain a Wigner-Wiley distribution matrix corresponding to each radar radiation source imbalance data.

[0072] The Wigner-Wiley distribution matrix corresponding to the unbalanced data of each radar radiation source is:

[0073]

[0074] in, is the Wigner-Wiley distribution matrix corresponding to the unbalanced data of the j-th radar radiation source in the i-th category; r ij is the imbalance data of the jth radar radiation source in the i-th category; r ij * For r ij is the conjugate complex number of ; τ is the sampling time corresponding to the radar radiation source imbalance data; t and ω are the time reference independent variable and frequency reference independent variable of the Wigner-Wiley distribution.

[0075] Step S12: performing bicubic interpolation and normalization processing on each Wigner-Wiley distribution matrix in turn to obtain a Wigner-Wiley distribution matrix corresponding to the unbalanced data of each radar radiation source.

[0076] S2. Obtain the Wigner-Willi distribution matrix number corresponding to each category in the first Wigner-Willi distribution matrix set to determine the maximum Wigner-Willi distribution matrix number and the minimum Wigner-Willi distribution matrix number.

[0077] For example, there are 6 categories in the first Wigner-Wiley distribution matrix set, namely quaternary phase shift keying signal, conventional pulse signal, nonlinear frequency modulation signal, linear frequency modulation signal, frequency coded signal and binary phase shift keying signal, and the corresponding Wigner-Wiley distribution matrix numbers are 2000, 400, 200, 200, 50 and 10 respectively. Then the maximum Wigner-Wiley distribution matrix number M and the minimum Wigner-Wiley distribution matrix number N in the first Wigner-Wiley distribution matrix set are 2000 and 10 respectively.

[0078] S3. Oversampling each Wigner-Williams distribution square array in the first Wigner-Williams distribution square array set according to the maximum Wigner-Williams distribution square array number.

[0079] The specific implementation process of oversampling in this embodiment includes:

[0080] Step S31, divide the difference between the maximum Wigner-Willi distribution matrix number and the Wigner-Willi distribution matrix number corresponding to each category in the first Wigner-Willi distribution matrix set by the Wigner-Willi distribution matrix number corresponding to the corresponding category to determine the number of Wigner-Willi distribution matrixes that need to be supplemented for each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix set.

[0081] In this embodiment, the number of Wigner-Willi distribution square matrices that need to be supplemented for each Wigner-Willi distribution square matrices corresponding to each category in the first Wigner-Willi distribution square matrix set is:

[0082] M i,j =floor[(M max -M i ) / M i ];

[0083] Among them, M i,j is the number of Wigner-Willi distribution matrices that need to be supplemented in the jth Wigner-Willi distribution matrix of the i-th category in the first Wigner-Willi distribution matrix set; M max is the maximum Wigner-Wiley distribution matrix number; M i is the number of Wigner-Willi distribution matrices of the i-th category in the first Wigner-Willi distribution matrix set; floor[*] means rounding down.

[0084] For example, the numbers of Wigner-Wiley distribution matrices corresponding to the first, second, third, fourth, fifth and sixth categories in the first Wigner-Wiley distribution matrix set are 2000, 400, 200, 200, 50 and 10 respectively. Then the numbers of Wigner-Wiley distribution matrices required to be generated for the first, second, third, fourth, fifth and sixth categories are 0, 1600, 1800, 1800, 1950 and 1950 respectively, i.e., the maximum number of Wigner-Wiley distribution matrices M max Subtract the number of Wigner-Wiley distribution matrices M of the i-th category i Difference: M max -M i . Then divide the difference by the number of Wigner-Wiley distribution matrices M of the i-th category i Then round down to get the number of Wigner-Wiley distribution matrices required to generate each Wigner-Wiley distribution matrix corresponding to the i-th category. For example, the number of Wigner-Wiley distribution matrices required to generate the j-th Wigner-Wiley distribution matrix corresponding to the 1st category is M. 1,j is 0, the number of Wigner-Wiley distribution matrices M required to generate the j-th Wigner-Wiley distribution matrix corresponding to the second category 2,j=4, the number of Wigner-Wiley distribution matrices M required to generate the j-th Wigner-Wiley distribution matrix corresponding to the 3rd and 4th categories 3,j and M 4,j are all 9. The number of Wigner-Wiley distribution matrices M required to generate the j-th Wigner-Wiley distribution matrix corresponding to the 5th category 5,j =39, the number of Wigner-Wiley distribution matrices M required to generate the jth Wigner-Wiley distribution matrix corresponding to the 6th category 6,j is 199.

[0085] Step S32: extracting Wigner-Willie distribution square matrices whose number of Wigner-Willie distribution square matrices to be supplemented is less than or equal to the number of Wigner-Willie distribution square matrices of the corresponding category from the first Wigner-Willie distribution square matrix set, and putting them into the first Wigner-Willie distribution square matrix subset, so as to determine all Wigner-Willie distribution square matrices to be supplemented by each Wigner-Willie distribution square matrix corresponding to each category in the first Wigner-Willie distribution square matrix subset;

[0086] For example, the number of Wigner-Wiley distribution matrices M required to generate the j-th Wigner-Wiley distribution matrix corresponding to the first category is 1,j is 0, the number of Wigner-Wiley distribution matrices M required to generate the j-th Wigner-Wiley distribution matrix corresponding to the second category 2,j =4, the number of Wigner-Wiley distribution matrices M required to generate the j-th Wigner-Wiley distribution matrix corresponding to the 3rd and 4th categories 3,j and M 4,j are all 9. The number of Wigner-Wiley distribution matrices M required to generate the j-th Wigner-Wiley distribution matrix corresponding to the 5th category 5,j is 39, which are less than 2000, 400, 200, 200 and 50 respectively. Therefore, all Wigner-Wiley distribution matrices corresponding to the 1st, 2nd, 3rd, 4th and 5th categories belong to the first Wigner-Wiley distribution matrix subset.

[0087] In this step, the specific implementation process of determining all Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset includes:

[0088] Step S321, calculating the first Euclidean distances between each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset and the remaining Wigner-Willi distribution matrixes corresponding to the corresponding category, so as to select the Wigner-Willi distribution matrixes corresponding to the first M smallest first Euclidean distances of each Wigner-Willi distribution matrix corresponding to each category.

[0089] In this embodiment, M is the number of Wigner-Wiley distribution matrices that need to be supplemented for each Wigner-Wiley distribution matrix. For example, the first Euclidean distances between the jth Wigner-Wiley distribution matrix corresponding to the second category in the first Wigner-Wiley distribution matrix subset and the remaining 1999 Wigner-Wiley distribution matrices are respectively selected from the 1999 first Euclidean distances to select the Wigner-Wiley distribution matrices corresponding to the first four smallest first Euclidean distances.

[0090] Step S322: Subtract the corresponding Wigner-Willi distribution matrix from the Wigner-Willi distribution matrices corresponding to the first M smallest first Euclidean distances corresponding to each Wigner-Willi distribution matrix, multiply by a random positive decimal, and then add the corresponding Wigner-Willi distribution matrix to obtain the Wigner-Willi distribution matrix that needs to be supplemented for each Wigner-Willi distribution matrix.

[0091] For example, the Wigner-Wiley distribution matrices corresponding to the first four smallest first Euclidean distances are subtracted from the j-th Wigner-Wiley distribution matrix corresponding to the second category, and then multiplied by a random positive decimal (between 0 and 1) and added to the j-th Wigner-Wiley distribution matrix corresponding to the second category to obtain the Wigner-Wiley distribution matrix that needs to be supplemented to the j-th Wigner-Wiley distribution matrix corresponding to the second category.

[0092] Step S33: extracting Wigner-Willie distribution square matrices whose number of Wigner-Willie distribution square matrices to be supplemented is greater than the number of Wigner-Willie distribution square matrices of the corresponding category from the first Wigner-Willie distribution square matrix set, and putting them into the second Wigner-Willie distribution square matrix subset to determine all Wigner-Willie distribution square matrices to be supplemented by each Wigner-Willie distribution square matrix corresponding to each category in the second Wigner-Willie distribution square matrix subset.

[0093] For example, the number of Wigner-Wiley distribution matrices M required to generate the j-th Wigner-Wiley distribution matrix corresponding to the 6th category 6,j It is 199, which is larger than the Wigner-Willi distribution matrix number 10 corresponding to the sixth category and belongs to the second Wigner-Willi distribution matrix subset.

[0094] In this step, the specific implementation process of determining all Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrices corresponding to each category in the second Wigner-Willi distribution matrix subset includes:

[0095] Step S331, setting the sequence number j of each category of the Wigner-Williams distribution matrix in the second Wigner-Williams distribution matrix subset to 1;

[0096] For example, the first Wigner-Wiley distribution matrix corresponding to the sixth category.

[0097] Step S332: subtract the j-th Wigner-Willi distribution matrix corresponding to the corresponding category from the remaining Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset, multiply the result by a random positive decimal, and then add the j-th Wigner-Willi distribution matrix corresponding to the corresponding category to obtain the Wigner-Willi distribution matrix that needs to be supplemented for the j-th Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset, and then update it to the Wigner-Willi distribution matrix corresponding to the corresponding category in the second Wigner-Willi distribution matrix subset.

[0098] The remaining 9 (2nd to 10th) Wigner-Willi distribution matrices corresponding to the 6th category are subtracted from the 1st Wigner-Willi distribution matrix corresponding to the 6th category, multiplied by a random positive decimal and added to the 1st Wigner-Willi distribution matrix corresponding to the 6th category to obtain 9 new Wigner-Willi distribution matrices and put them into the second Wigner-Willi distribution matrix subset.

[0099] Step S333, let j=j+1, and determine whether j is less than J, if yes, return to step S332; if no, go to step S334;

[0100] Wherein, J is the number of Wigner-Willi distribution matrices corresponding to each category in the second Wigner-Willi distribution matrix subset.

[0101] In this embodiment, the number of Wigner-Williams distribution matrices corresponding to the sixth category in the second Wigner-Williams distribution matrix subset is 10.

[0102] Step S334: Calculate the number of Wigner-Williams distribution matrices that need to be supplemented for each Wigner-Williams distribution matrix corresponding to each category in the updated second Wigner-Williams distribution matrix subset.

[0103] The updated number of Wigner-Willi distribution matrices corresponding to the sixth category in the second Wigner-Willi distribution matrix subset is 10+10*9=100. Then according to step S31, the number of Wigner-Willi distribution matrices required to be supplemented for each Wigner-Willi distribution matrix corresponding to the sixth category is calculated to be 19.

[0104] Step S335, determine whether the number of Wigner-Willi distribution matrixes required to be supplemented for each Wigner-Willi distribution matrix corresponding to each category after the update is less than the number of Wigner-Willi distribution matrixes corresponding to the corresponding category after the update; if so, put the updated corresponding category into the first Wigner-Willi distribution matrix subset, and return to step S321; if not, use the updated second Wigner-Willi distribution matrix subset to replace the second Wigner-Willi distribution matrix subset before the update, and return to step S331.

[0105] Since the number of Wigner-Willi distribution matrices 19 required to be supplemented for each Wigner-Willi distribution matrix corresponding to the 6th category is less than the number of Wigner-Willi distribution matrices 100 corresponding to the 6th category, the updated 6th category is placed in the first Wigner-Willi distribution matrix subset.

[0106] Step S34, merge all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset and all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset into the first Wigner-Willi distribution matrix set.

[0107] S4. Input each Wigner-Wiley distribution matrix in the oversampled first Wigner-Wiley distribution matrix set into the convolutional neural network for the first training.

[0108] The convolutional neural network structure of this embodiment refers to Figure 2 , including the first convolutional layer, the first maximum pooling layer, the second convolutional layer, the second maximum pooling layer, the third convolutional layer, the third maximum pooling layer, the fourth convolutional layer, the fourth maximum pooling layer, the fifth convolutional layer, the fifth maximum pooling layer, the flattening layer, the first fully connected layer, the batch normalization layer, the ReLu activation layer, the second fully connected layer, the batch normalization layer and the Softmax activation layer. The number of convolution kernels of the first convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer is 15, the convolution kernel size is 3×3, and the convolution operation can be padded with zeros. The pooling matrix size of the first maximum pooling layer, the second maximum pooling layer, the third maximum pooling layer, the fourth maximum pooling layer and the fifth maximum pooling layer is 3×3, and the pooling operation step size is 2. The number of neurons in the first fully connected layer and the second fully connected layer is 72 and 6 respectively. The activation layer (ReLu) is an activation layer with the activation function of ReLu. The Softmax activation layer is an activation layer with the activation function of Softmax.

[0109] According to the size of the radar radiation source sample, set the input size of the convolutional neural network. Figure 2 The input radar radiation source samples are sequentially passed through the first convolution layer, the first maximum pooling layer, the second convolution layer, the second maximum pooling layer, the third convolution layer, the third maximum pooling layer, the fourth convolution layer, the fourth maximum pooling layer, the fifth convolution layer and the fifth maximum pooling layer to obtain the deep feature I with a size of (4, 4, 20). The deep feature I is then passed through the flattening layer, the first fully connected layer, the batch normalization layer, the ReLu activation layer, the second fully connected layer and the batch normalization layer to obtain the deep feature II with a size of 1×6. The deep feature II is passed through the Softmax activation layer to obtain the output of the convolutional neural network.

[0110] S5. According to the minimum number of Wigner-Wiley distribution matrices, extract the Wigner-Wiley distribution matrix for each category in the first Wigner-Wiley distribution matrix set, and then input it into the convolutional neural network after the first training for the second training.

[0111] For example, the numbers of Wigner-Willi distribution matrices corresponding to the 1st, 2nd, 3rd, 4th, 5th and 6th categories in the first Wigner-Willi distribution matrix set are 2000, 400, 200, 200, 50 and 10 respectively. Ten (i.e., the minimum number of Wigner-Willi distribution matrices) Wigner-Willi distribution matrices are extracted from each category of the first Wigner-Willi distribution matrix set.

[0112] S6. Obtain a radar emitter data set of unknown categories, convert it into a second Wigner-Wiley distribution matrix set, and then input it into the convolutional neural network after the second training to perform radar emitter identification.

[0113] The conversion process in this step is the same as the conversion process in step S1. For the specific implementation process, please refer to the conversion process in step S1, which will not be described in detail here.

[0114] In order to further ensure the accuracy of recognition, between S3 and S4, the radar emitter recognition method of this embodiment further includes:

[0115] One of the Wigner-Willi distribution square matrices in the oversampled first Wigner-Willi distribution square matrices is used as the first Wigner-Willi distribution square matrices, and the remaining Wigner-Willi distribution square matrices are used as the second Wigner-Willi distribution square matrices.

[0116] If the total number of Wigner-Willi distribution matrices in the first Wigner-Willi distribution matrix set after oversampling is 12,000, one of the Wigner-Willi distribution matrices is used as the first Wigner-Willi distribution matrix, and the remaining 11,999 Wigner-Willi distribution matrices are all used as the second Wigner-Willi distribution matrix.

[0117] The first Euclidean distances between the first Wigner-Wiley distribution matrix and each second Wigner-Wiley distribution matrix are calculated to obtain a first Euclidean distance set.

[0118] The first Euclidean distance set of this embodiment includes 11999 first Euclidean distances.

[0119] Selecting second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances from the first Euclidean distance set, and determining whether the categories of the second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances are all the same as the categories of the first Wigner-Wiley distribution matrix; if so, retaining the first Wigner-Wiley distribution matrix; if not, removing the first Wigner-Wiley distribution matrix.

[0120] In this embodiment, N is a preset value.

[0121] This embodiment converts the unbalanced data set of radar radiation sources into the first Wigner-Wiley distribution matrix set to determine the maximum number of Wigner-Wiley distribution matrixes and the minimum number of Wigner-Wiley distribution matrixes; uses the maximum number of Wigner-Wiley distribution matrixes to obtain the balanced data set of radar radiation sources after oversampling the first Wigner-Wiley distribution matrix set, to implement the first training in the convolutional neural network; uses the minimum number of Wigner-Wiley distribution matrixes to obtain the balanced data set of radar radiation sources after extracting the Wigner-Wiley distribution matrix of each category in the first Wigner-Wiley distribution matrix set, to implement the second training in the convolutional neural network; finally, uses the radar radiation source data set of unknown categories to implement radar radiation source identification through the convolutional neural network trained for the second time, and the identification accuracy is high. The invention is suitable for radar radiation source identification of radar radiation source balanced data sets, radar radiation source identification of radar radiation source unbalanced data sets, and radar radiation source identification of a small amount of radar radiation source data.

[0122] The above embodiment is implemented by using the radar radiation source identification system given in the following embodiment:

[0123] Another embodiment provides a radar emitter identification system for imbalanced data, the radar emitter identification system comprising:

[0124] A conversion module, used for acquiring an unbalanced data set of radar radiation sources of known categories and converting it into a first Wigner-Wiley distribution square matrix set;

[0125] An acquisition module is used to acquire the number of Wigner-Willi distribution square matrices corresponding to each category in the first Wigner-Willi distribution square matrix set to determine the maximum number of Wigner-Willi distribution square matrices and the minimum number of Wigner-Willi distribution square matrices;

[0126] an oversampling module, configured to oversample each Wigner-Willi distribution square matrix in the first Wigner-Willi distribution square matrix set according to the maximum number of Wigner-Willi distribution square matrixes;

[0127] A first training module is used to input each Wigner-Wiley distribution square matrix in the oversampled first Wigner-Wiley distribution square matrix set into the convolutional neural network for a first training;

[0128] A second training module is used to extract Wigner-Wiley distribution matrices from each category in the first Wigner-Wiley distribution matrix set according to the minimum number of Wigner-Wiley distribution matrices, and then input the extracted Wigner-Wiley distribution matrices into the convolutional neural network after the first training for a second training;

[0129] The recognition module is used to obtain a radar emitter data set of unknown categories, convert it into a second Wigner-Wiley distribution matrix set, and then input it into a convolutional neural network that has been trained for the second time to perform radar emitter recognition.

[0130] Furthermore, the conversion module includes:

[0131] A Wigner-Wiley domain transformation submodule is used to perform a Wigner-Wiley domain transformation on each radar radiation source imbalance data in the radar radiation source imbalance data set to obtain a Wigner-Wiley distribution matrix corresponding to each radar radiation source imbalance data;

[0132] The normalization processing submodule is used to perform bicubic interpolation and normalization processing on each Wigner-Wiley distribution matrix in turn to obtain the Wigner-Wiley distribution matrix corresponding to the unbalanced data of each radar radiation source.

[0133] Furthermore, the oversampling module includes:

[0134] a processing submodule, configured to divide the difference between the maximum number of Wigner-Willi distribution square matrices and the number of Wigner-Willi distribution square matrices corresponding to each category in the first Wigner-Willi distribution square matrix set by the number of Wigner-Willi distribution square matrices corresponding to the corresponding category, so as to determine the number of Wigner-Willi distribution square matrices that need to be supplemented for each Wigner-Willi distribution square matrix corresponding to each category in the first Wigner-Willi distribution square matrix set;

[0135] A first extraction submodule is used to extract, from the first Wigner-Willi distribution square matrix set, Wigner-Willi distribution square matrices whose number of Wigner-Willi distribution square matrices to be supplemented is less than or equal to the number of Wigner-Willi distribution square matrices of the corresponding category, and put them into the first Wigner-Willi distribution square matrix subset to determine all Wigner-Willi distribution square matrices to be supplemented by each Wigner-Willi distribution square matrix corresponding to each category in the first Wigner-Willi distribution square matrix subset;

[0136] A second extraction submodule is used to extract, from the first Wigner-Wiley distribution square matrix set, the Wigner-Wiley distribution square matrix whose number of Wigner-Wiley distribution square matrixes to be supplemented is greater than the number of Wigner-Wiley distribution square matrixes of the corresponding category, and put them into the second Wigner-Wiley distribution square matrix subset to determine all the Wigner-Wiley distribution square matrices to be supplemented by each Wigner-Wiley distribution square matrix corresponding to each category in the second Wigner-Wiley distribution square matrix subset;

[0137] A merging and extracting submodule is used to merge all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset and all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset into the first Wigner-Willi distribution matrix set.

[0138] Furthermore, the first extraction submodule includes:

[0139] A first calculation subunit is used to calculate the first Euclidean distance between each Wigner-Wiley distribution matrix corresponding to each category in the first Wigner-Wiley distribution matrix subset and the remaining Wigner-Wiley distribution matrix corresponding to the corresponding category, so as to select the Wigner-Wiley distribution matrix corresponding to the first M smallest first Euclidean distances corresponding to each Wigner-Wiley distribution matrix corresponding to each category;

[0140] Where M is the number of Wigner-Wiley distribution matrices that need to be supplemented for each Wigner-Wiley distribution matrix;

[0141] The processing subunit is used to subtract the corresponding Wigner-Wiley distribution matrix from the Wigner-Wiley distribution matrices corresponding to the first M smallest first Euclidean distances corresponding to each Wigner-Wiley distribution matrix, multiply the matrix by a random positive decimal, and then add the corresponding Wigner-Wiley distribution matrix to obtain the Wigner-Wiley distribution matrix that needs to be supplemented for each Wigner-Wiley distribution matrix.

[0142] Furthermore, the second extraction submodule includes:

[0143] A setting subunit, used to set the sequence number j=1 of each category of the Wigner-Wiley distribution matrix in the second Wigner-Wiley distribution matrix subset;

[0144] an updating subunit, for subtracting the j-th Wigner-Wiley distribution matrix corresponding to the corresponding category from the remaining Wigner-Wiley distribution matrix corresponding to each category in the second Wigner-Wiley distribution matrix subset, multiplying the result by a random positive decimal and adding the j-th Wigner-Wiley distribution matrix corresponding to the corresponding category to obtain the Wigner-Wiley distribution matrix that needs to be supplemented for the j-th Wigner-Wiley distribution matrix corresponding to each category in the second Wigner-Wiley distribution matrix subset, and then updating the result to the Wigner-Wiley distribution matrix corresponding to the corresponding category in the second Wigner-Wiley distribution matrix subset;

[0145] A first judging subunit is used to set j=j+1 and judge whether j is less than J. If yes, j is transmitted to the updating subunit; if no, the updated second Wigner-Wiley distribution matrix subset is transmitted to the second calculating subunit;

[0146] Wherein, J is the number of Wigner-Wiley distribution matrices corresponding to each category in the second Wigner-Wiley distribution matrix subset;

[0147] A second calculation subunit is used to calculate the number of Wigner-Wiley distribution matrixes that need to be supplemented for each Wigner-Wiley distribution matrix corresponding to each category in the updated second Wigner-Wiley distribution matrix subset;

[0148] The second judgment subunit is used to judge whether the number of Wigner-Wiley distribution matrixes required to be supplemented for each Wigner-Wiley distribution matrix corresponding to each updated category is less than the number of Wigner-Wiley distribution matrixes corresponding to the updated corresponding category. If so, the updated corresponding category is placed in the first Wigner-Wiley distribution matrix subset and transmitted to the first calculation subunit; if not, the updated second Wigner-Wiley distribution matrix subset is used to replace the second Wigner-Wiley distribution matrix subset before updating, and transmitted to the updating subunit.

[0149] Furthermore, the radar emitter identification system further includes:

[0150] A setting module, used for setting one of the Wigner-Wiley distribution square matrices in the oversampled first Wigner-Wiley distribution square matrices as the first Wigner-Wiley distribution square matrices, and setting the remaining Wigner-Wiley distribution square matrices as the second Wigner-Wiley distribution square matrices;

[0151] A calculation module, used for calculating the first Euclidean distances between the first Wigner-Wiley distribution matrix and each second Wigner-Wiley distribution matrix, to obtain a first Euclidean distance set;

[0152] a judgment module, used for selecting second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances from the first Euclidean distance set, and judging whether the categories of the second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances are the same as the categories of the first Wigner-Wiley distribution matrices, if yes, retaining the first Wigner-Wiley distribution matrices; if no, removing the first Wigner-Wiley distribution matrices;

[0153] Wherein, N is a preset value.

[0154] The principles, formulas and parameter definitions involved in the above embodiments are all applicable and will not be described in detail here.

[0155] Please note that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several variations and improvements can be made, which all belong to the scope of protection of the present application. Therefore, the scope of protection of the patent in this application shall be based on the attached claims.

Claims

1. A radar emitter identification method for unbalanced data, characterized in that: The radar radiation source identification method comprises: Step S1, obtaining an unbalanced data set of radar radiation sources of known categories and converting it into a first Wigner-Wiley distribution matrix set; Step S2, obtaining the Wigner-Willi distribution matrix number corresponding to each category in the first Wigner-Willi distribution matrix set to determine the maximum Wigner-Willi distribution matrix number and the minimum Wigner-Willi distribution matrix number; Step S3, oversampling each Wigner-Willi distribution square matrix in the first Wigner-Willi distribution square matrix set according to the maximum Wigner-Willi distribution square matrix number; Step S4, inputting each Wigner-Wiley distribution square matrix in the oversampled first Wigner-Wiley distribution square matrix into the convolutional neural network for the first training; Step S5, according to the minimum number of Wigner-Wiley distribution matrices, extracting Wigner-Wiley distribution matrices from each category in the first Wigner-Wiley distribution matrix set, and then inputting the extracted Wigner-Wiley distribution matrices into the convolutional neural network after the first training for a second training; Step S6: Obtain a radar emitter data set of unknown category, convert it into a second Wigner-Wiley distribution matrix set, and then input it into the convolutional neural network after the second training to perform radar emitter identification.

2. The radar radiation source identification method according to claim 1, characterized in that: In step S1, the specific implementation process of the conversion includes: Step S11, performing a Wigner-Wiley domain transformation on each radar radiation source imbalance data in the radar radiation source imbalance data set to obtain a Wigner-Wiley distribution matrix corresponding to each radar radiation source imbalance data; Step S12: performing bicubic interpolation and normalization processing on each Wigner-Wiley distribution matrix in turn to obtain a Wigner-Wiley distribution matrix corresponding to the unbalanced data of each radar radiation source.

3. The radar radiation source identification method according to claim 2, characterized in that: In step S3, the specific implementation process of the oversampling includes: Step S31, dividing the difference between the maximum Wigner-Willi distribution square matrix number and the Wigner-Willi distribution square matrix number corresponding to each category in the first Wigner-Willi distribution square matrix set by the Wigner-Willi distribution square matrix number corresponding to the corresponding category, so as to determine the number of Wigner-Willi distribution square matrices that need to be supplemented for each Wigner-Willi distribution square matrix corresponding to each category in the first Wigner-Willi distribution square matrix set; Step S32: extracting Wigner-Willie distribution square matrices whose number of Wigner-Willie distribution square matrices to be supplemented is less than or equal to the number of Wigner-Willie distribution square matrices of the corresponding category from the first Wigner-Willie distribution square matrix set, and putting them into the first Wigner-Willie distribution square matrix subset, so as to determine all Wigner-Willie distribution square matrices to be supplemented by each Wigner-Willie distribution square matrix corresponding to each category in the first Wigner-Willie distribution square matrix subset; Step S33, extracting from the first Wigner-Willie distribution square matrix set the Wigner-Willie distribution square matrix whose number of Wigner-Willie distribution square matrices to be supplemented is greater than the number of Wigner-Willie distribution square matrices of the corresponding category, and putting them into the second Wigner-Willie distribution square matrix subset, so as to determine all the Wigner-Willie distribution square matrices to be supplemented by each Wigner-Willie distribution square matrix corresponding to each category in the second Wigner-Willie distribution square matrix subset; Step S34, merge all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset and all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset into the first Wigner-Willi distribution matrix set.

4. The radar radiation source identification method according to claim 3, characterized in that: In the step S32, the specific implementation process of determining all Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrices corresponding to each category in the first Wigner-Willi distribution matrix subset includes: Step S321, calculating the first Euclidean distances between each Wigner-Wiley distribution matrix corresponding to each category in the first Wigner-Wiley distribution matrix subset and the remaining Wigner-Wiley distribution matrixes corresponding to the corresponding category, so as to select the Wigner-Wiley distribution matrixes corresponding to the first M smallest first Euclidean distances corresponding to each Wigner-Wiley distribution matrix corresponding to each category; Where M is the number of Wigner-Wiley distribution matrices that need to be supplemented for each Wigner-Wiley distribution matrix; Step S322: Subtract the corresponding Wigner-Willi distribution matrix from the Wigner-Willi distribution matrices corresponding to the first M smallest first Euclidean distances corresponding to each Wigner-Willi distribution matrix, multiply by a random positive decimal, and then add the corresponding Wigner-Willi distribution matrix to obtain the Wigner-Willi distribution matrix that needs to be supplemented for each Wigner-Willi distribution matrix.

5. The radar radiation source identification method according to claim 4, characterized in that: In the step S33, the specific implementation process of determining all Wigner-Wiley distribution matrices to be supplemented by each Wigner-Wiley distribution matrix corresponding to each category in the second Wigner-Wiley distribution matrix subset includes: Step S331, setting the sequence number j of each category of the Wigner-Wiley distribution matrix in the second Wigner-Wiley distribution matrix subset to 1; Step S332, subtract the j-th Wigner-Willi distribution matrix corresponding to the corresponding category from the remaining Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset, multiply by a random positive decimal, and then add the j-th Wigner-Willi distribution matrix corresponding to the corresponding category to obtain the Wigner-Willi distribution matrix that needs to be supplemented for the j-th Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset, and update it to the Wigner-Willi distribution matrix corresponding to the corresponding category in the second Wigner-Willi distribution matrix subset; Step S333, let j=j+1, and determine whether j is less than J, if yes, return to step S332; if no, go to step S334; Wherein, J is the number of Wigner-Wiley distribution matrices corresponding to each category in the second Wigner-Wiley distribution matrix subset; Step S334, calculating the number of Wigner-Willi distribution matrixes that need to be supplemented for each Wigner-Willi distribution matrix corresponding to each category in the updated second Wigner-Willi distribution matrix subset; Step S335, determine whether the number of Wigner-Willi distribution matrixes required to be supplemented for each Wigner-Willi distribution matrix corresponding to each category after the update is less than the number of Wigner-Willi distribution matrixes corresponding to the corresponding category after the update; if so, put the updated corresponding category into the first Wigner-Willi distribution matrix subset, and return to step S321; if not, use the updated second Wigner-Willi distribution matrix subset to replace the second Wigner-Willi distribution matrix subset before the update, and return to step S331.

6. The radar radiation source identification method according to claim 5, characterized in that: Between step S3 and step S4, the radar emitter identification method further includes: Setting one of the Wigner-Wiley distribution square matrices in the oversampled first Wigner-Wiley distribution square matrices as the first Wigner-Wiley distribution square matrices, and setting the remaining Wigner-Wiley distribution square matrices as the second Wigner-Wiley distribution square matrices; Calculating the first Euclidean distances between the first Wigner-Wiley distribution matrix and each second Wigner-Wiley distribution matrix to obtain a first Euclidean distance set; Selecting second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances from the first Euclidean distance set, and determining whether the categories of the second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances are the same as the categories of the first Wigner-Wiley distribution matrices, if yes, retaining the first Wigner-Wiley distribution matrices; if no, removing the first Wigner-Wiley distribution matrices; Wherein, N is a preset value.

7. A radar emitter identification system for unbalanced data, characterized in that: The radar emitter identification system comprises: A conversion module, used for acquiring an unbalanced data set of radar radiation sources of known categories and converting it into a first Wigner-Wiley distribution square matrix set; An acquisition module is used to acquire the number of Wigner-Willi distribution square matrices corresponding to each category in the first Wigner-Willi distribution square matrix set to determine the maximum number of Wigner-Willi distribution square matrices and the minimum number of Wigner-Willi distribution square matrices; an oversampling module, configured to oversample each Wigner-Willi distribution square matrix in the first Wigner-Willi distribution square matrix set according to the maximum number of Wigner-Willi distribution square matrixes; A first training module is used to input each Wigner-Wiley distribution square matrix in the oversampled first Wigner-Wiley distribution square matrix set into the convolutional neural network for a first training; A second training module is used to extract Wigner-Wiley distribution matrices from each category in the first Wigner-Wiley distribution matrix set according to the minimum number of Wigner-Wiley distribution matrices, and then input the extracted Wigner-Wiley distribution matrices into the convolutional neural network after the first training for a second training; The recognition module is used to obtain a radar emitter data set of unknown categories, convert it into a second Wigner-Wiley distribution matrix set, and then input it into a convolutional neural network that has been trained for the second time to perform radar emitter recognition.

8. The radar radiation source identification system according to claim 7, characterized in that: The conversion module comprises: A Wigner-Wiley domain transformation submodule is used to perform a Wigner-Wiley domain transformation on each radar radiation source imbalance data in the radar radiation source imbalance data set to obtain a Wigner-Wiley distribution matrix corresponding to each radar radiation source imbalance data; The normalization processing submodule is used to perform bicubic interpolation and normalization processing on each Wigner-Wiley distribution matrix in turn to obtain the Wigner-Wiley distribution matrix corresponding to the unbalanced data of each radar radiation source.

9. The radar emitter identification system according to claim 8, characterized in that: The oversampling module comprises: a processing submodule, configured to divide the difference between the maximum number of Wigner-Willi distribution square matrices and the number of Wigner-Willi distribution square matrices corresponding to each category in the first Wigner-Willi distribution square matrix set by the number of Wigner-Willi distribution square matrices corresponding to the corresponding category, so as to determine the number of Wigner-Willi distribution square matrices that need to be supplemented for each Wigner-Willi distribution square matrix corresponding to each category in the first Wigner-Willi distribution square matrix set; A first extraction submodule is used to extract, from the first Wigner-Willi distribution square matrix set, Wigner-Willi distribution square matrices whose number of Wigner-Willi distribution square matrices to be supplemented is less than or equal to the number of Wigner-Willi distribution square matrices of the corresponding category, and put them into the first Wigner-Willi distribution square matrix subset to determine all Wigner-Willi distribution square matrices to be supplemented by each Wigner-Willi distribution square matrix corresponding to each category in the first Wigner-Willi distribution square matrix subset; A second extraction submodule is used to extract, from the first Wigner-Wiley distribution square matrix set, the Wigner-Wiley distribution square matrix whose number of Wigner-Wiley distribution square matrixes to be supplemented is greater than the number of Wigner-Wiley distribution square matrixes of the corresponding category, and put them into the second Wigner-Wiley distribution square matrix subset to determine all the Wigner-Wiley distribution square matrices to be supplemented by each Wigner-Wiley distribution square matrix corresponding to each category in the second Wigner-Wiley distribution square matrix subset; A merging and extracting submodule is used to merge all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the first Wigner-Willi distribution matrix subset and all the Wigner-Willi distribution matrices to be supplemented by each Wigner-Willi distribution matrix corresponding to each category in the second Wigner-Willi distribution matrix subset into the first Wigner-Willi distribution matrix set.

10. The radar emitter identification system according to claim 9, characterized in that: The radar emitter identification system also includes: A setting module, used for setting one of the Wigner-Wiley distribution square matrices in the oversampled first Wigner-Wiley distribution square matrices as the first Wigner-Wiley distribution square matrices, and setting the remaining Wigner-Wiley distribution square matrices as the second Wigner-Wiley distribution square matrices; A calculation module, used for calculating the first Euclidean distances between the first Wigner-Wiley distribution matrix and each second Wigner-Wiley distribution matrix, to obtain a first Euclidean distance set; a judgment module, used for selecting second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances from the first Euclidean distance set, and judging whether the categories of the second Wigner-Wiley distribution matrices corresponding to the first N smallest first Euclidean distances are the same as the categories of the first Wigner-Wiley distribution matrices, if yes, retaining the first Wigner-Wiley distribution matrices; if no, removing the first Wigner-Wiley distribution matrices; Wherein, N is a preset value.

Images