Informer-based monopulse secondary radar angle measurement method

Abstract

The invention discloses a monopulse secondary radar angle measurement method based on Informer, and the method comprises the steps: decoding video data obtained by a secondary radar, and constructing an original decoding matrix according to the decoded data; respectively obtaining three target azimuth angles by using the original decoding matrix through a weighted average algorithm, a weighted average algorithm after low-pass filtering and a weighted average algorithm after Kalman filtering, and fusing the three target azimuth angles to obtain a target angle vector; fusing the original decoding matrix and the target angle vector to obtain a first splicing matrix; when determining that the secondary radar detection target and the ADS-B target are the same target, associating secondary radar data and ADS-B data of the target to form a first sample matrix and a second sample matrix; taking a coding matrix obtained by splicing the first sample matrix and the second sample matrix as the input of an Informer model, and outputting a prediction angle sequence; and performing performance evaluation on the Informer model through the evaluation index. According to the invention, accurate prediction of the target angle is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of angle measurement, in particular to a single pulse secondary radar angle measurement method based on Informer. BACKGROUND

[0002] The secondary radar system is a widely used radio positioning system in the field of air traffic control and surveillance, which realizes the identification, positioning and information exchange of targets by transmitting inquiry signals from ground stations and receiving response signals from aircraft or other targets. Unlike primary radar which relies on target echo, secondary radar response signals carry coded information (such as height, identity code), thus providing higher data reliability and anti-jamming capability. However, in terms of angle measurement, secondary radar systems face precision and robustness challenges. The current mainstream angle measurement technology is the single pulse method, which uses the sum and difference channels of the secondary radar antenna to calculate the target's deviation angle relative to the beam axis through amplitude ratio or phase difference, achieving high-precision target bearing estimation within a single pulse.

[0003] Single pulse angle measurement technology has been the mainstream solution in active radar and secondary radar since the mid-20th century, widely used in warning tracking, precision guidance and air traffic surveillance. Existing technologies have further improved their performance through various optimization methods, such as using wideband amplitude single pulse visualization methods, processing response signal information through pre-defined format, and using visualization tools to assist angle calculation, improving the operator's intuitive understanding of the angle measurement process and debugging efficiency. Similarly, some self-calibration mechanisms optimize amplitude / phase ratio through filtering and calibration algorithms, breaking through the limitations of traditional amplitude comparison single pulse angle measurement, achieving high-precision multi-target bearing estimation in complex backgrounds. Other improvements include single pulse angle measurement optimization for mechanical scanning microwave radars, solving the problem of low precision through subarray division and echo processing, and receiver improvement in low signal environments, using improved receivers to maintain angle measurement capability in low SNR. Although these existing technologies optimize precision through calibration, filtering and visualization, secondary radar single pulse angle measurement still has significant problems, especially in real complex electromagnetic environments, such as interference sensitivity, poor dynamic adaptability, limited multi-target resolution, etc. These limitations are increasingly prominent in modern aviation surveillance, and there is an urgent need for a new single pulse secondary radar angle measurement method.

[0004] The main problems of existing single pulse secondary radar angle measurement methods include: (1) Interference sensitivity: under conditions of multipath reflection, electronic jamming or low SNR, nonlinear noise (such as main lobe multi-point source interference) can cause amplitude / phase distortion, increase angle measurement error, especially in rain, fog or multi-target scenarios; (2) Poor dynamic adaptability: traditional methods rely on linear approximation or fixed threshold statistical models, which are difficult to capture the nonlinear dependence relationship changing over time in sequence signals. For example, when the number of responses is insufficient, the average calculation of the traditional model cannot accurately determine the signal center, resulting in increased angle error. In addition, changes in pulse repetition frequency or target maneuvers will cause dynamic fluctuations in the signal sequence, and fixed parameters cannot be adjusted in real time, causing angle lag or ambiguity, which is particularly evident in complex electromagnetic environments; (3) Limited multi-target resolution capability: in dense airspace or multi-source conditions, angle ambiguity or false targets are easily generated, and two close-range targets cannot be effectively distinguished, leading to tracking loss or low monitoring efficiency. SUMMARY

[0005] To solve the problems of insufficient precision and poor robustness of traditional monopulse secondary radar angle measurement methods in complex electromagnetic environments, the application provides a monopulse secondary radar angle measurement method based on Informer.

[0006] The application discloses a monopulse secondary radar angle measurement method based on Informer, which comprises: Step 1: decode the video data obtained by the secondary radar, and construct an original decoding matrix according to the decoded data; the rows of the original decoding matrix are Σ, Δ, Ω, S, and B, and the columns are R1 to R N , N represents that the target has responded N times, and each response is a column; Step 2: obtain three target azimuth angles by using the weighted average algorithm, the low-pass filtering and weighted average algorithm, and the Kalman filtering and weighted average algorithm in the fusion encoding module, respectively, and fuse the three target azimuth angles to obtain a target angle vector; fuse the original decoding matrix and the target angle vector to obtain a first splicing matrix; Step 3: according to the target code, time, azimuth, and distance, when it is determined that the secondary radar detected target and the ADS-B target are the same target, associate the secondary radar data and the ADS-B data of the target to form a first sample matrix and a second sample matrix; Step 4: use the encoding matrix obtained by splicing the first sample matrix and the second sample matrix as the input of the Informer model, and the Informer model is composed of an encoder Encoder, a decoder Decoder, a full connection layer, and an output layer; the encoder uses a sparse attention mechanism to extract the target angle features in the encoding matrix and sends them to the decoder, the input matrix of the decoder is obtained by performing mask processing on the encoding matrix, the matrix after the decoder is sent to the full connection layer, and the predicted angle sequence is output after mapping by the full connection layer, and the Adam optimizer is used to update the parameters of the Informer model; Step 5: performance evaluation of the Informer model by evaluating indicators; the evaluation indicators include correlation coefficient, root mean square error, mean absolute error, and mean absolute percentage error.

[0007] Further, the step 1 comprises: The video data of different types of targets collected by the secondary radar is coded to obtain original coding data. The original coding data is organized into an original coding matrix in the form of a matrix; a single original coding matrix contains all data of a single round of response of a target received by the secondary radar; all data includes sum channel amplitude Σ, difference channel amplitude Δ, control channel amplitude Ω, symbol bit S, and wave bit number B; all data of a single round of response of a target is a column of a single original coding matrix, and there are N columns, N being the total number of rounds of response of the target.

[0008] Further, the process of obtaining the target angle vector comprises: The weighted average, low-pass filter weighted average, and Kalman filter weighted average are used to obtain the target azimuth, and all obtained target azimuths form the target angle vector.

[0009] Further, the target azimuth obtained by the weighted average comprises: Weighted average: the target azimuth is obtained by calculating the values of sum channel amplitude Σ and difference channel amplitude Δ, querying the off-boresight angle (OBA) table generated by the antenna pattern by the difference, and performing weighted average processing on the azimuth angles calculated for N times of response to obtain the final target azimuth angle. The target azimuth obtained by the low-pass filter weighted average comprises: Low-pass filter weighted average: the sum channel amplitude Σ and the difference channel amplitude Δ are respectively subjected to low-pass filter processing, and the value of the sum channel amplitude Σ minus the difference channel amplitude Δ is calculated, and the target azimuth is obtained by querying the OBA table and weighted average calculation. The target azimuth obtained by the Kalman filter weighted average comprises: Kalman filter weighted average: the values of the sum channel amplitude Σ and the difference channel amplitude Δ are respectively subjected to Kalman filter processing, and the target azimuth is obtained by querying the OBA table and weighted average calculation.

[0010] Further, the weighted average processing of the azimuth angles calculated for N times of response to obtain the final target azimuth angle comprises: The point group azimuth is obtained by the mass center method weighted average through the following formula:

[0011] wherein, is the azimuth of the i-th point group.i The sum of the response and the channel amplitude value. For the first i The azimuth angle obtained from the OBA query. The azimuth angle of the target point trace group obtained by weighted averaging using the center of mass method. The number of responses.

[0012] Further, the step of performing low-pass filtering on the reach amplitude Σ and the difference channel amplitude Δ respectively includes: The sum and difference channel amplitudes Σ and Δ are subjected to low-pass filtering using the following formula:

[0013] in, This represents the amplitude value of the sum or difference channel acquired at the current moment. This represents the current output value of the filter. This is the output value of the filter at the previous moment. The sampling time of the low-pass filter. This is the sampling frequency of the low-pass filter.

[0014] Further, the step of performing Kalman filtering on the values ​​of the access amplitude Σ and the difference channel amplitude Δ includes: The state prediction equation, covariance prediction equation, Kalman gain calculation equation, state update equation, and state update equation of the Kalman filter are used to filter the access amplitude Σ and the difference channel amplitude Δ, respectively. The state prediction equation is:

[0015] in, For the sake of the previous The time predicted by the first Theoretical prediction value at time, For the first The final predicted value at time [time]. The interval time, For the first Estimated velocity at any given time; The covariance prediction equation is:

[0016] in, For the sake of the previous The time predicted by the first Covariance at time, For process noise covariance; The Kalman gain calculation equation is as follows:

[0017] in, For Kalman gain, To measure the noise covariance, its value is the measurement accuracy of the secondary radar and the channel and difference channels; The state update equation is

[0018] in, These are the measured values, i.e., the measured amplitude values ​​of the sum or difference channel; The covariance update equation is:

[0019] in, For the sake of the previous The time predicted by the first Covariance at time.

[0020] Furthermore, the correlation coefficient, root mean square error, mean absolute error, and mean absolute percentage error are all determined based on the actual target angle and the target angle predicted by the Informer model.

[0021] Due to the adoption of the above technical solutions, this application has the following advantages: This application integrates the mass center method, low-pass filtering method and Kalman filtering method, and uses the Informer sparse attention mechanism to extract features, efficiently capturing nonlinear dependencies and dynamic patterns in the signal sequence, so that the model can make selection and correction of the three orientations according to the implicit features of the original data, thereby solving the problems of insufficient accuracy and poor robustness of the traditional single-pulse secondary radar angle measurement method in complex electromagnetic environments, and improving the angle measurement accuracy. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0023] Figure 1 This is a flowchart illustrating an Informer-based monopulse secondary radar angle measurement method according to an embodiment of this application.

[0024] Figure 2 This is a schematic diagram of the channel low-pass filtering and Kalman filtering results in an embodiment of this application.

[0025] Figure 3 This is a schematic diagram of the difference channel low-pass filtering and Kalman filtering results in an embodiment of this application.

[0026] Figure 4 is an ADS-B target azimuth angle and Informer predicted azimuth angle diagram of an embodiment of the present application.

[0027] Figure 5 is an Informer model effect diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0028] The present application is further illustrated in conjunction with the accompanying drawings and embodiments, and the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art shall fall within the scope of protection of the embodiments of the present application.

[0029] The traditional monopulse technique calculates the target angle through static amplitude ratio or phase difference, and has the problems of poor anti-interference and dynamic adaptability, and limited multi-target resolution capability. The present application introduces the Informer model (the Informer model is a deep learning model based on the Transformer architecture, specially designed for efficient processing of long sequence time series prediction tasks), encodes the original and Σ, difference Δ, control (Ω), sign bit (S) and wave bit (B) data decoded from the secondary radar output through a fusion encoding module, associates the encoded target data with the ADS-B target through a target association module, and finally uses the associated target data as the training input of the Informer model, uses the sparse attention mechanism of the Informer for feature extraction, and then efficiently captures the nonlinear dependence and dynamic pattern in the signal sequence, to realize accurate prediction of the angle.

[0030] Referring to Figure 1 , the present application provides an embodiment of a monopulse secondary radar angle measurement method based on Informer, which comprises: Step 1: decode the video data obtained by the secondary radar, and construct an original decoding matrix according to the decoded data; the rows of the original decoding matrix are Σ, Δ, Ω, S and B, and the columns are R1 to R N , N represents that the target has responded N times, and each response is a column; Step 2: obtain three target azimuth angles in the fusion encoding module by using the original decoding matrix with weighted average algorithm, low-pass filtering and weighted average algorithm, and Kalman filtering and weighted average algorithm respectively, and fuse the three target azimuth angles to obtain a target azimuth angle vector; fuse the original decoding matrix and the target azimuth angle vector to obtain a first splicing matrix; Step 3: according to the target code, time, azimuth and distance, when it is determined that the secondary radar detected target and the ADS-B target are the same target, associate the secondary radar data and the ADS-B data of the target to form a first sample matrix and a second sample matrix; Step 4: The encoding matrix obtained by splicing the first sample matrix and the second sample matrix is taken as the input of the Informer model, and the Informer model is composed of an encoder Encoder, a decoder Decoder, a full connection layer and an output layer; the encoder extracts the target angle feature in the encoding matrix by using a sparse attention mechanism, and sends the target angle feature to the decoder; the input matrix of the decoder is obtained by performing mask processing on the encoding matrix; the matrix after the decoder is sent to the full connection layer, and the predicted angle sequence is output after mapping by the full connection layer; the Adam optimizer is used to update the parameters of the Informer model; Step 5: The performance of the Informer model is evaluated by evaluation indexes; the evaluation indexes include correlation coefficient, root mean square error, mean absolute error and mean absolute percentage error.

[0031] In the embodiment of the application, step 1 comprises: The video data of different types of targets collected by the secondary radar is decoded to obtain original decoded data; The original decoded data is organized into an original decoded matrix in the form of a matrix; a single original decoded matrix contains all data of a single round of response of a target received by the secondary radar; all data includes sum channel amplitude Σ, difference channel amplitude Δ, control channel amplitude Ω, symbol bit S and wave bit number B; all data of a single round of response of a target is taken as a column of a single original decoded matrix, and there are N columns, and N is the total number of rounds of response of the target.

[0032] In the embodiment of the application, the process of obtaining the target angle vector comprises: The weighted average, low-pass filter weighted average and Kalman filter weighted average are used to obtain the target azimuth, and all the obtained target azimuths form the target angle vector.

[0033] In the embodiment of the application, the target azimuth obtained by using the weighted average comprises: Weighted average: the target azimuth is obtained by calculating the values of the sum channel amplitude Σ and the difference channel amplitude Δ, querying the off-boresight angle (OBA) table generated by the antenna pattern by the difference, and performing weighted average processing on the azimuth angles obtained by calculating N times of response to obtain the final target azimuth; The target azimuth obtained by using the low-pass filter weighted average comprises: Low-pass filter weighted average: the sum channel amplitude Σ and the difference channel amplitude Δ are respectively subjected to low-pass filter processing, and the value of the sum channel amplitude Σ minus the difference channel amplitude Δ is calculated, and the target azimuth is obtained by querying the OBA table and weighted average calculation; The target azimuth obtained by using the Kalman filter weighted average comprises: KALMAN filter weighted average: respectively, and the value of the sum of the amplitude of the access channel Σ and the difference channel amplitude Δ, through the OBA table query and weighted average calculation, the target azimuth is obtained.

[0034] In the embodiment of the application, the azimuth calculated by N times of response is weighted and averaged to obtain the final target azimuth, including: The point group azimuth is obtained by weighted average through the following formula:

[0035] Among them, is the sum of the amplitude of the access channel Σ of the first i response, is the azimuth obtained by OBA query of the first i time, is the azimuth of the target point group obtained by weighted average through the center of mass method, is the number of responses.

[0036] In the embodiment of the application, the sum of the amplitude of the access channel Σ and the difference channel amplitude Δ are respectively low-pass filtered, including: The sum of the amplitude of the access channel Σ and the difference channel amplitude Δ are low-pass filtered through the following formula:

[0037] Among them, is the amplitude value of the sum of the amplitude of the access channel Σ or the difference channel collected at the current time, is the output value of the filter at the current time, is the output value of the filter at the last time, is the sampling time of the low-pass filter, is the sampling frequency of the low-pass filter.

[0038] In the embodiment of the application, the values of the sum of the amplitude of the access channel Σ and the difference channel amplitude Δ are respectively Kalman filtered, including: The state prediction equation, the covariance prediction equation, the Kalman gain calculation equation, the state update equation and the state update equation of the Kalman filter are used to filter the sum of the amplitude of the access channel Σ and the difference channel amplitude Δ respectively: The state prediction equation is:

[0039] Among them, is the theoretical prediction value at the first time obtained according to the prediction of the previous time, is the final prediction value at the first time, is the interval time, the first moment speed estimation value; The covariance prediction equation is:

[0040] wherein, the first moment covariance predicted according to the previous moment, the process noise covariance; The Kalman gain calculation equation is:

[0041] wherein, the Kalman gain, the measurement noise covariance, the value of which is the measurement accuracy of the secondary radar and the sum and difference channels; The state update equation is

[0042] wherein, the actual measured value, that is, the actual amplitude value of the sum and difference channels; The covariance update equation is:

[0043] wherein, the first moment covariance predicted according to the previous moment.

[0044] In the embodiments of the present application, the correlation coefficient, the root mean square error, the mean absolute error and the mean absolute percentage error are all determined according to the actual target angle and the target angle predicted by the Informer model.

[0045] In order to facilitate understanding, a more specific embodiment of the present application is given: As Figure 1 shown, the present application provides an embodiment of a single-pulse secondary radar angle measurement method based on Informer, which specifically includes the following processes: S1: Training data acquisition. Collect secondary radar original decoding data (target code, target distance, Σ channel amplitude, Δ channel amplitude, Ω channel amplitude, symbol bit S, wave bit number B) and ADS-B data (target code, time, longitude, latitude) of different types of targets respectively. Organize the original decoding data into an original decoding matrix in the form of a matrix , the rows of the matrix are Σ, Δ, Ω, S and B, and the columns are R1 to RN, indicating that N times of responses are responded, and each response is a column.

[0046] S2: Data fusion encoding. Fuse the original decoding matrix and the prior calculated angle vector . The original decoding matrix contains all the data of one target single round of the secondary radar received, including the sum channel amplitude Σ, the difference channel amplitude Δ, the control channel amplitude (Ω), the sign bit (S) and the wave bit number (B), which provides the target with the most original information. The angle vector contains the azimuth angle of the target calculated by three traditional methods, respectively, direct weighted average, low-pass filter weighted average and Kalman filter weighted average.

[0047] S3: Target association. According to time, target code, azimuth and distance, associate the secondary radar target and ADS-B target to form sample data and .

[0048] S4: Model training. The sample data and are spliced into the encoding matrix as the input of the Informer model. The Informer model is composed of an encoder, a decoder, a fully connected layer and an output layer. The encoder extracts the target angle features in the encoding matrix using sparse attention mechanism and sends them to the decoder. The input matrix of the decoder is obtained after the encoding matrix is processed by the mask. The matrix after the decoder is sent to the fully connected layer, and the predicted angle sequence is output after the mapping of the fully connected layer. The Adam optimizer is used to update the parameters of the Informer model.

[0049] S5: Model evaluation. The evaluation indicators of the model are correlation coefficient (R2), root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE).

[0050] Optionally, the specific process behavior of S1 described is: S11: To ensure the diversity of training data, a large amount of secondary radar and ADS-B data in different regions and different time periods is collected, and data screening and processing are performed to eliminate target data missing for a long time and other obvious abnormal data.

[0051] Optionally, the specific process behavior of S2 described is: S21: The weighted average directly calculates the value of "sum minus difference", and then queries the OBA table generated by the difference value through the antenna diagram to obtain the target azimuth.N The azimuth angles obtained from the first response calculation are weighted and averaged to obtain the azimuth angles of the target point trace group; the low-pass filter weighted average first performs low-pass filtering on the "sum" and "difference" channels respectively and calculates the "sum" minus the "difference" value, and then obtains the target azimuth angle by looking up the OBA table and calculating the weighted average; the Kalman filter weighted average first performs Kalman filtering on the "sum" and "difference" values ​​respectively, and then obtains the target azimuth angle by looking up the OBA table and calculating the weighted average.

[0052] S22: The azimuth angle of the point group is obtained by weighted averaging using the center of mass method. The specific calculation formula is as follows:

[0053] in, For the first i The sum of the response and the channel amplitude value. For the first i The azimuth angle obtained from the OBA query. The azimuth angle of the target point trace group obtained by weighted averaging using the center of mass method. Number of responses S23: The low-pass filter primarily filters the amplitude values ​​of the sum and difference channels. The complete formula for calculating the filter is:

[0054] in, This represents the amplitude value of the sum or difference channel acquired at the current moment. This represents the current output value of the filter. This represents the output value of the filter at the previous moment. According to MATLAB simulation results, the sampling time of the low-pass filter ( The sampling frequency is set to 0.0001 seconds. The frequency is 300Hz.

[0055] S24: The sum and difference channel amplitude values ​​are filtered separately using a Kalman filter. The formula for calculating the Kalman filter is as follows: ①The state prediction equation is:

[0056] in, For the sake of the previous The time predicted by the first Theoretical prediction value at time, For the first The final predicted value at time. The interval time, For the first Estimated velocity at any given time.

[0057] ②Covariance prediction equation is:

[0058] Wherein, is the covariance of the first moment predicted according to the previous moment, is the covariance of the first moment predicted according to the previous moment, is the process noise covariance. ③Kalman gain calculation equation is:

[0059] Wherein,

[0060] is the Kalman gain, is the measurement noise covariance, and its value is the measurement accuracy of the secondary radar and the sum and difference channels.

[0061] ④State update equation is

[0062] Wherein, is the measured value, that is, the measured amplitude value of the sum and difference channels.

[0063] ⑤Covariance update equation is:

[0064] Optionally, the specific process behavior of S3 described is: S31: Use the target association module to associate the same secondary radar and ADS-B targets to generate and , specifically, determine that the secondary radar detection target and the ADS-B target are the same target according to the target code, time, azimuth and distance, and associate the secondary radar data and the ADS-B data of the target to form the corresponding and matrix row by row.

[0065] Optionally, the specific process behavior of S4 described is: S41: Divide 80% of the encoding matrix into a training set and 20% into a test set. Normalize all data, and the normalized data range is [0, 1]. The Informer model batch size (Batchsize) is set to 4, the input sequence length (Sequencelength) is set to 8, the label length (Labellength) is set to 7, and the prediction length (Predictionlength) is set to 1.

[0066] ​Optionally, the specific process behavior of S5 described is as follows: S51: The formula for calculating the root mean square error is:

[0067] S52: The formula for calculating the correlation coefficient is:

[0068] S53: The formula for calculating the mean absolute error is:

[0069] S54: The formula for calculating the mean absolute percentage error is:

[0070] Example 2 The specific process of S1 in Example 1 is as follows: S11: The secondary radar raw decoded data includes the target code, target range, Σ channel amplitude, Δ channel amplitude, Ω channel amplitude, symbol bit S, and wave number B. ADS-B target data includes the target code, time, longitude, and latitude. The ADS-B target azimuth is calculated using spherical trigonometry.

[0071] S12: To ensure the diversity of training data, a large amount of secondary radar and ADS-B data from different regions and time periods are collected, and the data is filtered and processed to remove target data that has been missing for a long time and other obviously abnormal data.

[0072] Example 3 The specific process of S2 in Example 1 is as follows: S21: The weighted average is obtained by directly calculating and subtracting the difference, and then querying the OBA table generated from the antenna diagram using the difference to obtain the target azimuth. N The azimuth angles obtained from the first response calculation are weighted and averaged to obtain the azimuth angles of the target point trace group; the low-pass filter weighted average first performs low-pass filtering on the sum and difference channels respectively and calculates the sum minus the difference value, and then obtains the target azimuth angle by looking up the OBA table and calculating the weighted average; the Kalman filter weighted average first performs Kalman filtering on the sum and difference values ​​respectively, and then obtains the target azimuth angle by looking up the OBA table and calculating the weighted average.

[0073] S22: Figure 2 The figure shows the results of low-pass filtering and Kalman filtering of the two channels. It illustrates the filtering effects of the low-pass filter and Kalman filter on the original decoded data of the two channels. It demonstrates that the low-pass filter and Kalman filter can effectively filter outomas in the amplitude of the two channels while maintaining tracking accuracy.

[0074] S23: Figure 3The figure shows the results of low-pass filtering and Kalman filtering of the difference channel. It illustrates the filtering effects of the low-pass filter and Kalman filter on the original difference channel decoded data. It demonstrates that the low-pass filter and Kalman filter can effectively filter outomas in the difference channel amplitude while maintaining tracking accuracy.

[0075] Example 4 The specific process of S4 in Example 1 is as follows: S41: Encoding Matrix 80% of the data is used for training, and 20% for testing. All data is normalized, with the normalized data range being [0,1]. The Informer model has a batch size of 4, an input sequence length of 8, a label length of 7, and a prediction length of 1.

[0076] Example 5 The specific process of S5 in Example 1 is as follows: S51: Figure 4 Single-round query for a single target ( N A schematic diagram showing the target azimuth calculated by the spherical trigonometric formula when the target value is 100, the target azimuth predicted by Informer, the target azimuth obtained by direct OBA query of the measured value, the target azimuth obtained by OBA query after low-pass filtering, and the target azimuth obtained by query after Kalman filtering. Figure 4 The effects of low-pass and Kalman filters on the original data are shown, as well as the deviation between the Informer prediction and the ADS-B target orientation, demonstrating the better prediction performance of the Informer.

[0077] S52: Figure 5 This section compares the performance of low-pass filtering, Kalman filtering, and the Informer model. (Informer model) The RMSE=0.17, MAE=0.17, and MAPE=0.0118 scores show significant improvements in all four metrics compared to low-pass filtering and Kalman filtering algorithms.

[0078] The application proposes a single pulse secondary radar angle measurement method based on Informer. The quality center method, low-pass filtering method and Kalman filtering method are fused, and finally the Informer sparse attention mechanism is used to extract features, efficiently capture the nonlinear dependence and dynamic pattern in the signal sequence, so that the model can select and correct the three azimuths according to the implicit characteristics of the original data, thereby solving the problem of insufficient precision and poor robustness of the traditional single pulse secondary radar angle measurement method in complex electromagnetic environment, and improving the angle measurement precision. Test results show that the RMSE of the model on the test set is 0.17, 94.61%, which fully shows that the model error is small, the generalization ability is strong, and the target prediction ability is very excellent.

[0079] From the implementation process, using the processing idea in the application, the informer model can be replaced by machine learning models such as LSTM or Transformer to realize single pulse secondary radar target angle prediction.

[0080] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the application and not to limit them, although the application has been described in detail with reference to the above examples, those skilled in the art should understand that: the specific embodiments of the application can still be modified or replaced by the same, without departing from the spirit and scope of the application. Any modification or equivalent replacement should be covered within the protection scope of the claims of the application.

Claims

1. A method for angle measurement using a single-pulse secondary radar based on Informer, characterized in that, include: Step 1: Decode the video data acquired by the secondary radar, and construct the original decoding matrix based on the decoded data; The rows of the original decoding matrix are Σ, Δ, Ω, S, B, and the columns are R1 to R2. N N represents the number of times the target responded, with each response being a separate column. Step 2: In the fusion coding module, the original decoding matrix is ​​used to obtain three target azimuth angles by weighted averaging, weighted averaging after low-pass filtering, and weighted averaging after Kalman filtering, respectively. The three target azimuth angles are then fused to obtain the target angle vector. The original decoding matrix and the target angle vector are then fused to obtain the first concatenation matrix. Step 3: Based on the target code, time, azimuth, and distance, when it is determined that the secondary radar detection target and the ADS-B target are the same target, associate the secondary radar data and ADS-B data of the target to form a first sample matrix and a second sample matrix. Step 4: The encoding matrix obtained by concatenating the first sample matrix and the second sample matrix is ​​used as the input of the Informer model. The Informer model consists of an encoder, a decoder, a fully connected layer, and an output layer. The encoder uses a sparse attention mechanism to extract the target angle features in the encoding matrix and sends them to the decoder. The input matrix of the decoder is obtained by masking the encoding matrix. The matrix after decoding is sent to the fully connected layer. After mapping by the fully connected layer, the predicted angle sequence is output. The Adam optimizer is used to update the parameters of the Informer model. Step 5: Evaluate the performance of the Informer model using evaluation metrics, including correlation coefficient, root mean square error, mean absolute error, and mean absolute percentage error.

2. The monopulse secondary radar angle measurement method based on Informer according to claim 1, characterized in that, Step 1 includes: The video data of different types of targets acquired by the secondary radar are decoded to obtain the raw decoded data. The raw decoded data is organized into a raw decoding matrix in the form of a matrix; a single raw decoding matrix contains all the data of a single round of response from a target received by the secondary radar; all the data includes the reach amplitude Σ, the difference channel amplitude Δ, the control channel amplitude Ω, the symbol bit S and the wave bit number B; all the data of a single round of response from a target are used as a column of a single raw decoding matrix, with a total of N columns, where N is the total number of rounds of response from the target.

3. The monopulse secondary radar angle measurement method based on Informer according to claim 1, characterized in that, The process of obtaining the target angle vector includes: The target azimuth angles are obtained by weighted averaging, low-pass filter weighted averaging, and Kalman filter weighted averaging, respectively, and all the obtained target azimuth angles are used to form a target angle vector.

4. The Informer-based single-pulse secondary radar angle measurement method according to claim 3, characterized in that, The target azimuth angle is obtained by using the weighted average, including: Weighted average: By calculating the values ​​of the reach amplitude Σ and the difference channel amplitude Δ, the target azimuth is obtained by querying the deviation angle OBA table generated by the antenna diagram through the difference. The azimuth angles obtained from the N responses are weighted and averaged to obtain the final target azimuth angle. The target azimuth angle is obtained by filtering and weighting the values ​​using the low-pass filter, including: Low-pass filter weighted average: The reach amplitude Σ and the difference channel amplitude Δ are low-pass filtered respectively, and the value of the reach amplitude Σ minus the difference channel amplitude Δ is calculated. The target azimuth angle is obtained by querying the OBA table and calculating the weighted average. The target azimuth angle is obtained by weighted averaging using the Kalman filter, including: Kalman filter weighted average: The values ​​of the access amplitude Σ and the difference channel amplitude Δ are processed by Kalman filtering, and the target azimuth angle is obtained by querying the OBA table and calculating the weighted average.

5. The Informer-based single-pulse secondary radar angle measurement method according to claim 4, characterized in that, The weighted average of the azimuth angles calculated from N responses is used to obtain the final target azimuth angle, including: The azimuth angle of the point group is obtained by using the weighted average of the center of mass method and the following formula: in, For the first i The sum of the response and the channel amplitude value. For the first i The azimuth angle obtained from the OBA query. The azimuth angle of the target point trace group obtained by weighted averaging using the center of mass method. The number of responses.

6. The Informer-based single-pulse secondary radar angle measurement method according to claim 4, characterized in that, The step of performing low-pass filtering on the reach amplitude Σ and the difference channel amplitude Δ includes: The sum and difference channel amplitudes Σ and Δ are subjected to low-pass filtering using the following formula: in, This represents the amplitude value of the sum or difference channel acquired at the current moment. This represents the current output value of the filter. This is the output value of the filter at the previous moment. The sampling time of the low-pass filter. This is the sampling frequency of the low-pass filter.

7. The Informer-based single-pulse secondary radar angle measurement method according to claim 4, characterized in that, The step of performing Kalman filtering on the values ​​of the access amplitude Σ and the difference channel amplitude Δ includes: The state prediction equation, covariance prediction equation, Kalman gain calculation equation, state update equation, and state update equation of the Kalman filter are used to filter the access amplitude Σ and the difference channel amplitude Δ, respectively. The state prediction equation is: in, For the sake of the previous The time predicted by the first Theoretical prediction value at time, For the first The final predicted value at time [time]. The interval time, For the first Estimated velocity at any given time; The covariance prediction equation is: in, For the sake of the previous The time predicted by the first Covariance at time, For process noise covariance; The Kalman gain calculation equation is as follows: in, For Kalman gain, To measure the noise covariance, its value is the measurement accuracy of the secondary radar and the channel and difference channels; The state update equation is in, These are the measured values, i.e., the measured amplitude values ​​of the sum or difference channel; The covariance update equation is: in, For the sake of the previous The time predicted by the first Covariance at time.

8. The monopulse secondary radar angle measurement method based on Informer according to claim 1, characterized in that, The correlation coefficient, root mean square error, mean absolute error, and mean absolute percentage error are all determined based on the actual target angle and the target angle predicted by the Informer model.

Images