A radar model automatic identification method and system combining intra-pulse and inter-pulse features

Abstract

This invention belongs to the field of radar signal recognition technology and discloses an automatic radar model identification method and system that combines intra-pulse and inter-pulse features. It employs a three-level classification method combining feature parameter matching, deep learning convolutional neural networks, and intra-pulse parameter analysis. The first level is a tolerance range classification based on feature parameter matching; the second level is image classification using convolutional neural networks, achieving secondary classification of radar signal modulation modes based on image recognition; the third level is intra-pulse modulation parameter classification for conventional pulses (no intra-pulse modulation), frequency-coded signals, composite modulation, nonlinear frequency modulation, linear frequency modulation signals, and phase-coded signals, achieving three-level radar signal classification using a time-frequency domain multi-parameter joint classification method. This invention obtains the intra-pulse modulation characteristics of radar signals by analyzing intra-pulse parameters and improves the accuracy of radar signal recognition by combining inter-pulse and intra-pulse radar signal parameters.

Description

Technical Field

[0001] This invention belongs to the field of radar signal recognition technology, and particularly relates to an automatic radar model identification method and system that combines intra-pulse and inter-pulse features. Background Technology

[0002] Radar emitter identification (EID) is a crucial core of electronic intelligence reconnaissance. By analyzing and identifying intercepted enemy radar signals, it provides commanders with battlefield situational information and tactical decision-making information. With the rapid development of electronic science and technology, traditional radar signal identification methods based on pulse description words (PDWs), including three-dimensional features such as pulse width (PW), carrier frequency (RF), and repetition period (PRI), are no longer sufficient to cope with today's complex battlefield electromagnetic environment. Furthermore, existing technologies suffer from problems such as the overlapping and blurred boundaries of different radar signal parameters in complex electromagnetic environments, and the unavoidable pulse splitting. Therefore, there is an urgent need to design an automatic radar signal identification method that combines intra-pulse and inter-pulse features.

[0003] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0004] Traditional radar signal reconnaissance methods based on pulse descriptors, including pulse width, carrier frequency, and repetition period (three-dimensional features), are no longer sufficient to cope with today's complex battlefield electromagnetic environment. Existing radar signal reconnaissance equipment can only perform template matching and identification of signals based on traditional three-dimensional features. This leads to an impact on the accuracy of target identification when two or more radars have overlapping parameters and blurred boundaries in the three-dimensional features. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides an automatic radar model identification method and system that combines intra-pulse and inter-pulse features, aiming to solve the problems of low accuracy and difficulty in avoiding pulse splitting in the traditional template matching identification method for radar signals in complex electromagnetic environments.

[0006] This invention is implemented as follows: an automatic radar model identification method combining intra-pulse and inter-pulse features includes:

[0007] A step-by-step identification method is adopted, which combines traditional feature parameter template matching, deep convolutional neural networks, and intrapulse parameter analysis. The first level is based on EDW (Emitter Discrepancy)... The system uses a feature parameter matching method to identify radar signals. When the identification result is only a Type 1 radar signal, the result is directly pushed. When the identification result is a Type 2 or higher radar signal, it enters the second-level identification stage. The second level uses a convolutional neural network to classify radar signal modulation methods (including linear frequency modulation, phase coding, conventional pulse (no modulation within the pulse), frequency coding, composite modulation, nonlinear frequency modulation, etc.) based on time-frequency images. This aims to distinguish radar platforms with similar carrier frequency, pulse width, and repetition period but different intra-pulse modulation methods in the first-level identification. When the classification result is still a Type 2 or higher radar signal, it enters the third-level identification stage. In the third level, if the classification results of the first-level identification based on feature parameter matching and the second-level identification based on intra-pulse modulation method are still multiple types of radar, the system performs intra-pulse parameter analysis of the radar intermediate frequency signal for the linear frequency modulation signal and phase coding signal classified in the second-level identification, and performs bandwidth calculation for other signals such as conventional pulse (no modulation within the pulse) and composite modulation. The calculated parameter values ​​are matched with the radar radiation source model database with tolerance. The final identification result of the radar signal is pushed through a multi-parameter joint classification method.

[0008] Furthermore, the automatic radar model identification method based on the combined intra-pulse and inter-pulse features includes the following steps:

[0009] Step 1: Construct a radar signal classification model;

[0010] Step two: Conduct radar signal classification training and identification;

[0011] Step 3: Update the radar signal classification model.

[0012] Furthermore, the construction of the radar signal classification model in step one includes:

[0013] The classification method is divided into three levels. The core idea is to gradually narrow the classification and recognition range. The first level is classification based on feature parameter matching. The second level is classification of radar signal intra-pulse modulation mode based on convolutional neural network of time-frequency image under the condition of classification of the first level. (Under this classification model, the intra-pulse modulation mode of radar signal is classified, specifically including: linear frequency modulation signal, phase-coded signal, conventional pulse (no intra-pulse modulation) signal, frequency-coded signal, nonlinear frequency modulation signal, and composite modulation signal). The third level is recognition based on intra-pulse parameter analysis under the condition of classification of the first level and classification of the second level.

[0014] Furthermore, the construction of the radar signal classification model in step one includes:

[0015] Part 1, the construction of the radar parameter model database, includes:

[0016] (1) Store the clarified radar intermediate frequency signal and sorted parameters, including radar time domain waveform, intra-pulse parameters and inter-pulse parameters. The inter-pulse parameters include the sorted pulse repetition period PRI, carrier frequency RF, and pulse width PW. The intra-pulse parameters include signal bandwidth (B), frequency modulation slope (y) of linear frequency modulation signal, symbol width (μ) of phase coded signal, and encoding method (z).

[0017] (2) Based on the detected and clarified radar parameter data, establish a corresponding radar radiation source model database S (hereinafter referred to as database S). The fields in database S include: radar signal carrier frequency (RF), pulse width (PW), pulse repetition period (PRI), time-frequency image (PIC) after time-frequency transformation of a single pulse, intra-pulse modulation type (Modu), signal bandwidth (B), frequency modulation slope (y) of linear frequency modulation signal, symbol width (μ) of phase coded signal, encoding method (z), and radar model (Target).

[0018] Part 2 involves the construction of a convolutional neural network model based on time-frequency images. This model establishes a convolutional neural network model for the intra-pulse modulation of radar signals.

[0019] (1) After the radar time-domain signal obtained in step (1) of Part 1 is transformed by time-frequency conversion, signals with the same modulation mode and different signal-to-noise ratios are selected to obtain several time-frequency image sets λ1.

[0020] (2) Use the time-frequency image set λ1 obtained in step (1) of Part 2 to establish radar signal intra-pulse modulation mode classification dataset λ2 and test set λ3 respectively;

[0021] (3) Construct a convolutional neural network, input the time-frequency image dataset λ2 into the convolutional neural network for training, form a primary classification neural network model, and use the test set λ3 to verify the accuracy of the model.

[0022] Furthermore, in step (1), time-frequency transformation methods such as short-time Fourier transform (STFT) are used to perform time-frequency transformation on the radar intermediate frequency signal to obtain several time-frequency images.

[0023] The convolutional neural network structure in step (3) of Part 2 includes a network structure with N convolutional layers, M pooling layers, and K fully connected layers, where N, M, and K are integers greater than or equal to 2; the activation function of the N convolutional layers is the ReLU activation function, and softmax is used for multi-target classification. The convolutional kernel size is m*m, where m is an odd integer greater than 3 and less than 31. The output is the direct classification and recognition of the corresponding radar signal intra-pulse modulation type or a certain type of special intra-pulse modulation method.

[0024] Part 3 involves analyzing and inputting the intra-pulse modulation parameters of signals such as linear frequency modulation, phase coding, conventional pulse (no intra-pulse modulation), and composite modulation into the database S. The frequency modulation slope y is calculated for linear frequency modulation signals, the symbol width μ and coding method z are calculated for phase coding signals, and the signal bandwidth B is calculated for other modulation methods such as conventional pulse (no intra-pulse modulation) and composite modulation.

[0025] Furthermore, the radar signal classification training and identification in step two includes:

[0026] (1) When a new radar signal is detected, the sorted radar signal includes three parameters: carrier frequency (CF), repetition period (PRI) and pulse width. The three parameters of the signal are compared and queried with tolerance with the three template parameters in the database S. The target is identified at the first level within the set tolerance range.

[0027] (2) Under the first-level recognition condition, when there is one and only one target result that matches the template data in the database S, the recognition result is directly pushed to the user;

[0028] (3) Under the first-level recognition condition, when there are two or more target results matching the template data in the database S, the second-level signal recognition process is carried out.

[0029] (4) Among the multiple radar signal boundaries that appeared in the previous step, the pulse signal r to be identified is... x After performing time-frequency conversion to form a time-frequency image, it is input into the second part of the model for secondary recognition, and the classification result of the modulation mode is output.

[0030] (5) Input the output modulation method identification result into the model of Part 3, which is a signal with modulation method such as conventional pulse (no modulation within the pulse), frequency coding, nonlinear frequency modulation, and composite modulation. Output the signal bandwidth calculation result under the same modulation method. Establish template data with radar radiation source signal model database S, perform database statement comparison query with tolerance, and push the identification result.

[0031] (6) Analyze the intra-pulse modulation parameters of the output modulation method identification results as linear frequency modulation signal and phase-coded signal, calculate the frequency modulation slope of the signal to form data parameter y, calculate the encoding method z and symbol width μ of the phase-coded signal, establish template data with radar radiation source signal model database S, perform database statement comparison query with tolerance, and push the identification results.

[0032] Furthermore, the radar signal classification model update in step three includes:

[0033] (1) In step two (1), when there is no result in matching the sorted signal with the template data in the database S, the user is prompted that there is no radar signal template matching. The user can update the parameters of the database S and further filter unknown signals in the feature parameter matching.

[0034] (2) Label the newly detected radar signals with unknown modulation methods, import them into the training dataset for retraining, and update the convolutional neural network model.

[0035] (3) Analyze the intra-pulse and inter-pulse parameters of the newly detected radar signals and update the database S.

[0036] Another object of the present invention is to provide an automatic radar model identification system based on the combined intra-pulse and inter-pulse features of the aforementioned automatic radar model identification method, wherein the automatic radar model identification system based on the combined intra-pulse and inter-pulse features includes:

[0037] The model building module is used to build radar signal classification models;

[0038] The training and recognition module is used for classification training and recognition of radar signals;

[0039] The model update module is used to update the radar signal classification model.

[0040] Another object of the present invention is to provide a computer device comprising a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the following steps:

[0041] A three-level classification method is adopted, which combines feature parameter matching, deep learning convolutional neural networks, and intra-pulse parameter estimation. The first level is the inter-pulse classification based on carrier frequency, repetition period, and pulse width, which is based on the traditional feature parameter matching method. The second level is the image classification using convolutional neural networks, which realizes secondary classification of radar signal modulation modes based on image recognition. The third level is the classification after radar intra-pulse parameter estimation, which realizes the three-level classification and recognition of radar signals using a time-frequency domain multi-parameter joint classification method.

[0042] Another objective of this invention is to provide an information data processing terminal for implementing the aforementioned automatic radar model identification system based on combined intra-pulse and inter-pulse characteristics.

[0043] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0044] The automatic radar signal identification method based on the combined intra-pulse and inter-pulse features provided by this invention combines multi-level classification methods, utilizes multi-parameter joint verification, and introduces a deep learning framework to improve accuracy while achieving rapid radar signal identification.

[0045] This invention analyzes radar intra-pulse parameters to obtain the intra-pulse modulation characteristics of radar signals, and improves the accuracy of radar signal recognition by combining inter-pulse and intra-pulse radar signal parameters.

[0046] This invention utilizes the traditional feature parameter matching method to achieve primary radar model identification based on feature parameters. In cases of ambiguity, a convolutional neural network is used to achieve secondary classification of radar signal modulation methods based on time-frequency images. In cases of ambiguity in secondary classification, intra-pulse parameter analysis is used to achieve tertiary identification of radar signals such as intra-pulse linear frequency modulation, phase coding, and conventional pulses (without intra-pulse modulation), which can effectively improve the accuracy of radar signal identification.

[0047] In the field of engineering applications of radar signal recognition, only the traditional feature parameter matching method is used. The present invention uses a method that combines feature parameter matching, convolutional neural network time-frequency image classification, and intra-pulse parameter feature combination for recognition, which increases the dimensional information of signal recognition and improves the reliability and accuracy of the signal recognition method.

[0048] This invention differs greatly from patent number CN 113156391 A. The patent and this invention are fundamentally different in their application. The patent is applied to radar signal sorting, while this invention is applied to the accurate identification of radar models. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of the automatic radar model identification method based on the combination of intra-pulse and inter-pulse features provided in this embodiment of the invention;

[0051] Figure 2 This is a block diagram of the radar model automatic identification system that combines intra-pulse and inter-pulse features provided in an embodiment of the present invention.

[0052] Figure 3 This is a specific network structure diagram provided in the embodiments of the present invention;

[0053] In the diagram: 1. Model building module; 2. Training and recognition module; 3. Model update module. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] To address the problems existing in the prior art, the present invention provides a method and system for automatic radar model identification that combines intra-pulse and inter-pulse features. The present invention will be described in detail below with reference to the accompanying drawings.

[0056] To enable those skilled in the art to fully understand how the present invention is specifically implemented, this section provides an explanatory description of the embodiments that expand upon the technical solutions of the claims.

[0057] The automatic radar model identification method based on combined intra-pulse and inter-pulse features provided in this invention includes a three-level classification method that integrates feature parameter matching, deep learning convolutional neural networks, and intra-pulse parameter estimation. The first level is a tolerance range classification based on feature parameter matching; the second level is image classification using convolutional neural networks, achieving secondary classification of radar signal modulation modes based on image recognition; the third level is classification of intra-pulse modulation parameters for different modulation modes, including linear frequency modulated signals and phase-coded signals, and bandwidth parameters for conventional pulses (no intra-pulse modulation) and composite modulation signals, achieving three-level radar signal classification using a time-frequency domain multi-parameter joint classification method. This invention's automatic radar model identification method combines multi-level classification, utilizes multi-parameter joint judgment, and introduces a deep learning framework, improving accuracy while achieving rapid radar signal identification. This invention obtains the intra-pulse modulation features of radar signals by analyzing intra-pulse parameters; and improves radar signal identification accuracy by combining inter-pulse and intra-pulse radar signal parameters.

[0058] like Figure 1 As shown, the automatic radar model identification method based on the combined intra-pulse and inter-pulse features provided in this embodiment of the invention includes the following steps:

[0059] S101, Construct a radar signal classification model;

[0060] SA1. Detecting and storing radar signals and parameters: Detecting radar signals means that the equipment collects radar time-domain signals and stores the sorted parameters. The parameters include intra-pulse parameters and inter-pulse parameters. Inter-pulse parameters include the sorted pulse repetition period PRI, carrier frequency RF, and pulse width PW. Intra-pulse parameters include signal bandwidth (B), frequency modulation slope (y) of the linear frequency modulated signal, symbol width (μ) of the phase-coded signal, and encoding method (z), etc.

[0061] SA2. The original radar signal detected in step SA1 is transformed into time and frequency. Based on the characteristics of different time and frequency images, signals with the same modulation method and different signal-to-noise ratios are manually selected to obtain several time and frequency image sets λ1.

[0062] SA3. Form a radar radiation source parameter model database S. The database fields include radar signal carrier frequency (RF), pulse width (PW), pulse repetition period (PRI), pulse time-frequency image (PIC), intra-pulse modulation type (Modu), signal bandwidth (B) (conventional pulse (no intra-pulse modulation), frequency coded, composite modulation signal, etc.), radar model (Target), frequency modulation slope (y) (only linear frequency modulation signal), symbol width (μ) and encoding method (z) (only phase coded signal).

[0063] SA4. Use the time-frequency image set λ1 obtained in step SA2 to establish a primary classification dataset λ2 and a test set λ3 for the intra-pulse modulation mode of radar signals.

[0064] SA5. Construct a convolutional neural network for modulation scheme classification:

[0065] SA5-1, specifically, employs a network structure with N convolutional layers, M pooling layers, and K fully connected layers, where N, M, and K are integers greater than or equal to 2; the N convolutional layers use the ReLU activation function, employ softmax for classification, and the output is the corresponding modulation type. The convolutional kernel size is m*m, where m is an odd integer greater than 3 and less than 31. Figure 3 The diagram shows the specific network structure of this invention when N=2 and m=3, where n is the number of target categories to be classified, and the convolutional layers are... Figure 3 The representation in the code is "convolutional layer (kernel)@(kernel size)"; pooling layers are... Figure 3 The representation in Figure * is "pooling layer@(pooling window size)"; the fully connected layer is represented as "fully connected layer@(number of neurons)". The kernel sizes for each convolutional layer are m*m*3 and 16*m*m, respectively, where m is an odd integer greater than 3 and less than 31. Pooling layers are used to reduce the number of training parameters after the convolutional layer output.

[0066] SA5-2. Classify the time-frequency images obtained in step SA3 to establish a classification dataset λ2 for radar signal modulation methods. The radar signal classification dataset is divided into six categories according to the modulation method within the radar signal pulse: linear frequency modulation signal, nonlinear frequency modulation signal, phase-coded signal, conventional pulse (no modulation within the pulse) signal, frequency-coded signal, and composite modulation signal. The data is input into a neural network to finally form a classification model of modulation methods based on the convolutional neural network of time-frequency images, and the accuracy of the model is verified using the test set λ3.

[0067] S102, for radar signal classification training and identification;

[0068] SB1. When a new radar signal is detected, the target is identified by comparing and querying the template data in the database S within a set tolerance range.

[0069] SB2. After calculation by SB1, if there is one and only one target result that matches the template data in database S within the tolerance range of the newly detected signal, the recognition result is directly pushed to the user.

[0070] SB3. After calculation by SB1, if there are two or more target results when the newly detected signal matches the template data in database S, it proves that there are multiple radar signal boundaries that are blurred in the sorted signal, and the next step of the signal identification process is carried out.

[0071] SB4. In SB3, among the radar signal aliasing signals, the ambiguous signal pulse signal r... x Perform time-frequency conversion to form a time-frequency image input to the second part of the model, and output the classification result of the modulation mode;

[0072] SB5-1: The output modulation method identification result is a signal with modulation methods such as conventional pulse (no modulation within the pulse), frequency coding, nonlinear frequency modulation, and composite modulation. Input it into the model in Part 3 to obtain the signal bandwidth parameters. Perform a database comparison query with the database S with tolerance and push the identification result.

[0073] SB5-2. Analyze the intra-pulse modulation parameters of the output modulation mode identification result as linear frequency modulation signal and phase-coded signal, input them into the model of Part 3, obtain the intra-pulse modulation parameters, and push the identification result by performing a database comparison query with the database S with tolerance.

[0074] SB5-3. In step (6), the radar single pulse time domain signal with linear frequency modulation obtained in step (4) is classified, the frequency modulation slope of the signal is calculated to form data parameter y, and the encoding method z and symbol width μ of the signal with phase coding obtained are calculated.

[0075] S103, update the radar signal classification model:

[0076] SC1, after calculation by SB1, if there is no result matching the template data in database S within the tolerance range, the user will be prompted that there is no radar signal template matching. The user can update the template database and parameter feature library, and further filter unknown signals in the feature parameter matching.

[0077] SC2. Label the newly detected radar signals with unknown modulation methods, import them into the training dataset for retraining, and update the convolutional neural network model.

[0078] SC3. Analyze the intra-pulse and inter-pulse parameters of the newly acquired radar signals and update the database S.

[0079] like Figure 2 As shown, the radar model automatic identification system combining intra-pulse and inter-pulse features provided in this embodiment of the invention includes:

[0080] Model building module 1 is used to build radar signal classification models;

[0081] Training and recognition module 2 is used for classification training and recognition of radar signals;

[0082] Model update module 3 is used to update the radar signal classification model.

[0083] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0084] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for automatic radar model identification based on combined intra-pulse and inter-pulse features, characterized in that, The automatic radar model identification method that combines intra-pulse and inter-pulse features includes: A step-by-step identification method using feature parameter template matching, deep convolutional neural networks, and intra-pulse parameter analysis is employed to push the final identification result of radar signals. This step-by-step identification method includes: The first level is feature parameter matching based on EDW (Emitter Discreption Word) parameters; when identified as a Type 1 radar radiation source, the identification result is directly pushed. The second level uses a convolutional neural network to classify the radar signal modulation method based on images, and performs a secondary classification based on the results of the first level identification. The third level performs intra-pulse parameter analysis of radar intermediate frequency signals for unmodulated conventional pulses, composite modulation, linear frequency modulation, and phase-coded signals, and performs a tertiary classification based on the results of the second level identification. The automatic radar model identification method that combines intra-pulse and inter-pulse features specifically includes the following steps: Step 1: Construct a radar signal classification model; Step two: Conduct radar signal classification training and identification; Step 3: Update the radar signal classification model; The classification model in step one includes: Model based on feature parameter matching; A convolutional neural network-based classification model for intra-pulse modulation of radar signals based on time-frequency images; Analysis and identification of intrapulse modulation parameters for conventional pulses, composite modulation, linear frequency modulation, and phase-coded signals without intrapulse modulation.

2. The automatic radar model identification method based on combined intra-pulse and inter-pulse features as described in claim 1, characterized in that, Radar signal classification models include: (1) Store the radar intermediate frequency signal and the sorted parameters, including the radar time-domain waveform, intra-pulse parameters and inter-pulse parameters. The inter-pulse parameters include the sorted pulse repetition period PRI, carrier frequency RF, and pulse width PW. The intra-pulse parameters include the signal bandwidth (B) and the frequency modulation slope of the linear frequency modulated signal (B). ), symbol width of phase-coded signal ( ), encoding method ( ); (2) Establish a corresponding radar parameter database based on the detected and identified radar parameter data. The database fields include: radar signal carrier frequency (RF), pulse width (PW), pulse repetition period (PRI), time-frequency image (PIC) of a single pulse after short-time Fourier transform, intra-pulse modulation type (Modu), signal bandwidth (B), and frequency modulation slope of the linear frequency modulated signal obtained in step (1). ), phase-coded signal symbol width ( ), encoding method ( Radar platform model (Target); Modulation mode identification: Under the condition of fuzzy matching of characteristic parameters of sorted signals, image recognition based on convolutional neural networks is used to form an identification model of the intra-pulse modulation mode of radar signals. 1) After the radar time-domain signal obtained in step (1) is transformed by time-frequency, signals with the same modulation method but different signal-to-noise ratios are selected to obtain several time-frequency image sets. The radar intermediate frequency signal was transformed using the short-time Fourier transform (STFT) method to obtain several time-frequency images. 2) The time-frequency image set obtained in step 1) Establish separate classification datasets for radar signal intra-pulse modulation methods. With test set ; 3) Construct a convolutional neural network to process the time-frequency image dataset. The input is used to train a convolutional neural network, forming a primary classification neural network model, and then tested using a test set. Verify the accuracy of the model; The convolutional neural network structure includes N convolutional layers, M pooling layers, and M fully connected layers, where N, M, and K are integers greater than or equal to 2. The activation function of the N convolutional layers is the ReLU activation function, and softmax is used for multi-target classification. The output is the direct classification and recognition of the corresponding radar signal intra-pulse modulation type or a certain type of special intra-pulse modulation method.

3. The automatic radar model identification method based on combined intra-pulse and inter-pulse features as described in claim 2, characterized in that, The intrapulse modulation parameters of unmodulated conventional pulses, composite modulation, frequency-coded, linear frequency modulation, and phase-coded signals were analyzed and entered into a database. Calculate the signal bandwidth B for unmodulated conventional pulses, composite modulated signals, and frequency-coded signals; calculate the frequency modulation slope for linear frequency modulated signals. Phase-coded signal calculation symbol width and encoding method .

4. The automatic radar model identification method based on combined intra-pulse and inter-pulse features as described in claim 1, characterized in that, The radar signal classification training and identification in step two includes: (1) When a new radar signal is detected, the sorted radar signal is compared with the database. The template data in the template is used for comparison queries with tolerance, and the target is first-level identified within the set tolerance range; (2) Within the tolerance range and with the database When there is one and only one target result in the template data matching, the recognition result is pushed directly to the user; (3) Within the tolerance range and with the database If the template data matching results for two or more targets, it proves that there are multiple radar signals with blurred boundaries or incorrect sorting parameters in the sorted signals, and the next step of the signal identification process is carried out. (4) The pulse signal of the first-level identification and sorting Perform time-frequency conversion to form a time-frequency image input to the second part of the model, and output the classification result of the modulation mode; (5) Calculate the frequency modulation slope to form data parameters for the radar single-pulse time-domain signals obtained in step (4) with linear frequency modulation. The encoding method of the signal is calculated from the phase-coded signal obtained by classification. and symbol width For the unmodulated conventional pulses obtained from classification, frequency coding, nonlinear frequency modulation, and composite modulation signals are used to calculate the signal bandwidth B, which is then processed through a tolerance-based database. Compare the queries and output the recognition results.

5. The automatic radar model identification method based on combined intra-pulse and inter-pulse features as described in claim 4, characterized in that, Step 3, updating the radar signal classification model, includes: (1) In step two (1), a problem occurs within the tolerance range and with the database. When no results are found in the template data matching, the user is prompted that there is no radar signal template matching. The user can update the template database and parameter feature library, and further filter unknown signals in the feature parameter matching. (2) Label the newly detected radar signals with unknown modulation methods, import them into the training dataset for retraining, and update the convolutional neural network model; (3) Analyze the intra-pulse and inter-pulse parameters of the newly acquired radar signals and update the database. .

6. A radar model automatic identification system that implements the radar model automatic identification method based on combined intra-pulse and inter-pulse features as described in any one of claims 1 to 5, characterized in that, The radar model automatic identification system that combines intra-pulse and inter-pulse features includes: The model building module is used to build radar signal classification models; The training and recognition module is used for classification training and recognition of radar signals; The model update module is used to update the radar signal classification model; Classification models include: Model based on feature parameter matching; A convolutional neural network-based classification model for intra-pulse modulation of radar signals based on time-frequency images; Analysis and identification of intrapulse modulation parameters for conventional pulses, composite modulation, linear frequency modulation, and phase-coded signals without intrapulse modulation.

7. A computer device comprising a memory and a processor, the memory storing a computer program, wherein when the computer program is executed by the processor, the processor performs the automatic radar model identification method based on the combined intra-pulse and inter-pulse characteristics as described in any one of claims 1 to 5.

Images