Radar signal sorting method based on deep learning BERT model

Abstract

The invention provides a radar signal sorting method based on a deep learning BERT model. The method comprises the following steps: processing parameter information of pulse description words and corresponding labels to form a data set; encoding the data set, and dividing the data set into a training set, a verification set and a test set; then building a deep learning BERT model, training and verifying the model by using the training set and the verification set, and storing model parameters; applying the trained model parameters to the model, and reasoning on a test set; and finally, calculating the sorting accuracy of the model, retaining the model parameter with the highest sorting accuracy, and completing the task model. According to the method, more parameters of the PDW are fully utilized, so that the sorting accuracy is improved; due to the end-to-end mode, the participation degree of manually setting the sorting parameters is reduced; the method has strong adaptability to high-density aliasing signal flow and complex modulation signals, and is less affected by complex electromagnetic environment.

Description

Technical Field

[0001] The present application relates to the field of radar reconnaissance system signal and information processing technology, and in particular to a radar signal sorting method based on a deep learning BERT model. Background Art

[0002] As one of the key technologies of radar reconnaissance system, the goal of radar signal sorting technology is to separate the pulse sequence generated by each radar from the received random aliased pulse stream. The accuracy of the sorting result has a significant impact on situational awareness, intelligence acquisition and decision-making.

[0003] The pulse description word (PDW) is a radar signal parameter that can be directly obtained through measurement. It is mainly composed of five instantaneous parameters, namely, radio frequency (RF), pulse width (PW), pulse amplitude (PA), time of arrival (TOA), and direction of arrival (DOA). The pulse repetition interval (PRI) can be obtained by calculating TOA. Most traditional sorting methods use PRI as the entry point for radar signal sorting. The traditional radar signal sorting method based on PRI is actually a deep mining of TOA information, aiming to find radar signals with similar PRI in the aliased pulse stream.

[0004] However, with the diversification of radar modulation, random modulation and agile modulation have been maturely applied, resulting in the task of radar signal sorting becoming increasingly difficult with the single parameter PRI in the context of increasing pulse density. This is because pulse-to-pulse parameters such as RF, PW, PA and DOA are less utilized. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that the traditional radar signal sorting method based on PRI has poor adaptability to complex modulated radar signals and poor sorting effect in the complex electromagnetic environment with increasing pulse density. An adaptive signal comprehensive processing technology based on deep learning BERT model is proposed. The technical solution of the present invention is:

[0006] Step 1: Multiple radars are turned on at the same time, and the receiver is used to obtain the aliased radar signals within a period of time to form pulse description words, and then structured storage is performed.

[0007] Step 2: normalize and preprocess RF, PW, PA and DOA;

[0008] Step 3: Use known radar information to automatically label the corresponding pulse description words to form a structured sample set file with column information of RF, PW, PA, TOA, DOA and label;

[0009] Step 4, segment the pulse description words in the text data to facilitate the model to learn the parameter characteristics of the pulse description words.

[0010] Step 5: Encode the labels in the dataset to facilitate parameter updates during model training.

[0011] Step 6: Divide the data set into training set, validation set, and test set in a ratio of 6:2:2.

[0012] Step 7: Build a deep learning BERT model, define the optimizer and loss function in the model, set the learning rate and training rounds, etc.

[0013] Step 8: Train the model, save the model parameters in each round of training, and draw the loss curve on the training set and validation set to observe whether the model converges.

[0014] Step 9: If the model converges, apply the model parameters saved in all rounds to the test set, calculate the sorting accuracy, draw the accuracy curve, retain the model parameters of the round with the highest sorting accuracy, and complete the task model.

[0015] Compared with the existing traditional methods, the present invention has the following significant advantages: (1) compared with radar signal sorting from the perspective of a single PRI parameter, it makes full use of more parameters of PDW, thereby improving sorting accuracy; (2) the end-to-end mode can reduce the manual participation in setting the parameters in the sorting process; (3) it has strong adaptability to high-density aliasing signal streams and complex modulated signals, and is less affected by complex electromagnetic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A sample data processing flow chart provided for an embodiment of the present application;

[0017] Figure 2 Flowchart of a radar signal sorting method based on a deep learning BERT model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0019] In order to improve the poor adaptability of the traditional radar signal sorting method based on PRI processing, the other four parameters in PDW are fully utilized to adapt to the radar signal sorting task requirements in the current complex electromagnetic environment. Deep learning, as a machine learning method based on artificial neural networks, has considerable advantages in processing complex problems and data. Deep learning can explore the intrinsic characteristics of sample data as much as possible, has strong learning and generalization capabilities, and can process high-dimensional data. In view of the characteristics of deep learning, the multi-dimensional parameter data of PDW is input into the network model, and the learning ability of the model is used to solve the radar signal sorting problem.

[0020] Step 1: Multiple radars are turned on at the same time, and the receiver is used to obtain the aliased radar signals within a period of time to form a pulse description word, and the RF, PW, PA, TOA and DOA in the pulse description word are structured and stored.

[0021] Step 2: In order to improve the model convergence speed and accuracy to a certain extent, RF, PW, PA and DOA are normalized and preprocessed.

[0022] Specifically, the preprocessing includes: In order to reflect the possible pulse loss phenomenon in a complex electromagnetic environment, the PDW data of each category in the data set is randomly lost in the same proportion. At the same time, in order to further improve the anti-interference ability of the model, a preset number of interference pulses are randomly generated in the sample data and the same data processing is performed; finally, in order to balance the categories, reduce the model tendency and improve the generalization ability, the number of PDWs in each category is kept equal.

[0023] Step 3: Write pulse description word batch labeling software, use known radar information to automatically label the corresponding pulse description words, and form a structured sample set file with column information of RF, PW, PA, TOA, DOA and label. The file format is CSV.

[0024] Step 4: Load a sample set, where the sample set contains pulse description words and label information.

[0025] Step 5, segment the pulse description words in the text data according to the dimensions including RF, PW, PA, TOA, and DOA, so as to facilitate the model to learn the parameter characteristics of the pulse description words.

[0026] Step 6: Encode the labels in the dataset to facilitate parameter updates during model training.

[0027] Step 7, divide the data set into training set, validation set and test set according to a preset ratio. A feasible preset ratio is 6:2:2.

[0028] Step 8: Build a deep learning BERT model, define the optimizer and loss function in the model, and set the learning rate and training rounds.

[0029] Specifically, the deep learning BERT model uses the standard BERT model, with 12 encoding layers, the Adam optimizer, the learning rate of 5e-5, the loss function of cross entropy, and the number of iterations of 100. The above parameters are changed according to different actual conditions.

[0030] Step 9: Train the model, save the model parameters in each round of training, add the total loss value of the training set and the total loss value of the validation set calculated in each round to two lists respectively, and draw the loss curves on the training set and validation set at the end of training to observe whether the model converges.

[0031] Step 10: If the model converges, apply the model parameters saved in all rounds to the test set and calculate the sorting accuracy. The accuracy calculation formula is:

[0032]

[0033] In formula (1), n pp Indicates the number of predicted results and labels of category p, t p Represents the true number of category p in the test set; the accuracy calculated in each round is added to the list, and the accuracy curve is drawn at the end of inference. The model parameters of the round with the highest sorting accuracy are retained to complete the task model.

[0034] The above-described embodiments of the present application do not constitute a limitation on the protection scope of the present application.

Claims

1. A radar signal sorting method based on a deep learning BERT model, characterized in that: The method comprises: Step 1: Multiple radars are turned on at the same time, and the receiver is used to obtain the aliased radar signals within a period of time to form a pulse description word, and the RF, PW, PA, TOA and DOA in the pulse description word are structured and stored; Step 2: normalize and preprocess RF, PW, PA and DOA; Step 3: Use known radar information to automatically label the corresponding pulse description words to form a structured sample set file with column information of RF, PW, PA, TOA, DOA and label; Step 4, loading a sample set, wherein the sample set includes pulse description words and label information; Step 5, segmenting the pulse description words in the text data according to the dimensions including RF, PW, PA, TOA, and DOA; Step 6: Encode the labels in the data set; Step 7, divide the data set into training set, validation set and test set according to the preset ratio; Step 8: Build a deep learning BERT model, define the optimizer and loss function in the model, and set the learning rate and training rounds; Step 9: Train the model, save the model parameters in each round of training, add the total loss value of the training set and the total loss value of the validation set calculated in each round to two lists respectively, and draw the loss curves on the training set and validation set at the end of training to observe whether the model converges; Step 10: If the model converges, the model parameters saved in all rounds are applied to the test set to calculate the sorting accuracy.

2. The method according to claim 1, characterized in that In step 2, preprocessing includes: The same proportion of random missing data is processed for each category of PDW in the data set; A preset number of interference pulses are randomly generated in the sample data, and the same data processing is performed; The number of PDWs in each category remains equal.

3. The method according to claim 1, characterized in that In step 8, the deep learning BERT model adopts the standard BERT model with 12 encoding layers, the Adam optimizer is selected as the optimizer, the learning rate is set to 5e-5, the loss function is cross entropy, and the iteration round is set to 100.

4. The method according to claim 1, characterized in that: The accuracy is determined by: In formula (1), n pp Indicates the number of predicted results and labels of category p, t p Represents the true number of category p in the test set; the accuracy calculated in each round is added to the list, and the accuracy curve is drawn at the end of inference. The model parameters of the round with the highest sorting accuracy are retained to complete the task model.

Images