Distributed radar target positioning method based on high-resolution range profile joint recognition

Abstract

The invention discloses a distributed radar target positioning method based on high-resolution range profile combined identification, and belongs to the technical field of signal processing, and the method comprises the steps: S1, a radar A and a radar B respectively emit linear frequency modulation pulses to a detection area, the signals are received by the radars after being scattered by a scene, and the received echo signal is # imgabs0 #; s2, carrying out dechirp receiving, and carrying out Fourier transform on a result to obtain a scene HRRP; s3, enabling the obtained scene HRRP data to pass through a moving target indication filter to obtain a moving target HRRP; s4, performing target detection on the moving target HRRP data by using CA-CFAR, and intercepting the HRRP data of the moving target # imgabs1 # and the moving target # imgabs2 #; s5, fusing the moving targets respectively determined by the radars A and B in a distance direction space to obtain a target point, and identifying the target point HRRP through a neural network to judge whether the target is true or false; according to the distributed radar target positioning method based on high-resolution range profile combined identification, the false target condition existing in distributed radar target positioning is solved, and target positioning is accurately and efficiently completed.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a distributed radar target positioning method based on high-resolution range profile joint recognition. Background Art

[0002] Synthetic aperture radar has strong target perception ability on a moving platform, but it is difficult to achieve effective target positioning when the platform is stationary, and the application scenarios are limited. Taking the intelligent driving platform as an example, when the radar moves with the vehicle, a synthetic aperture can be formed to obtain a two-dimensional high-resolution image, which has significant capabilities in environmental representation. However, after the vehicle is stationary, due to the lack of angle information of the single-channel radar, a single radar can only determine the distance unit where the target exists but cannot accurately locate the specific location of the target, that is, the target is on the ring corresponding to the distance unit. Therefore, distributed radar can be introduced for target positioning. If both radars can see all targets, the target must be at the intersection of the two radar target distance units. However, in a multi-target scenario, the intersection of the distance units where different targets are located will form a "false target" similar to the real target, thereby affecting the target positioning task.

[0003] In the existing technology, centralized processing is used to send the observation echoes of each radar to the fusion center for multi-target detection based on the spatial grid, and the generalized likelihood ratio detector and its corresponding constant false alarm rate (CFAR) detection threshold are derived to achieve signal-level collaborative detection of spatial matching data. At the same time, a false target elimination method based on a geometric joint decision algorithm is proposed for "false targets". The algorithm judges the real target and the false target by the strength of the target response, and there is a probability of misjudgment. If the first strong target is misjudged, subsequent judgments will be affected.

[0004] The existing literature "A distributed radar target positioning method" introduces the idea of ​​neural network, transforms the nonlinear unconstrained optimization problem into a nonlinear constrained optimization problem, uses the gradient differential system equation expansion as the neural network system, and uses the target position variable, the first relaxation variable, and the second relaxation variable as neurons in the neural network system to obtain the estimated value of the target position. This algorithm is suitable for high-precision target positioning problems, but not for multi-target positioning in complex scenarios.

[0005] Another existing document, "Distributed radar target detection method based on Doppler channel maximum value quantization", quantizes the Doppler channel maximum value into an actual quantized value according to the least squares quantization method and sends it to the fusion center through a communication link; the fusion center obtains the global optimal statistics through the quantized value, and calculates the global decision threshold through the global optimal statistics and the false alarm rate, and finally obtains the target detection result in the spatial resolution unit according to the threshold. The final target detection result of this method depends on the setting of the false alarm rate, and has no obvious advantage in resolving false targets. Summary of the invention

[0006] The purpose of the present invention is to provide a distributed radar target positioning method based on high-resolution range profile joint recognition to solve the problems existing in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides a distributed radar target positioning method based on high-resolution range profile joint recognition, comprising the following steps: S1, radar A and radar B respectively transmit linear frequency modulation pulses to the detection area. The signals are received by the radar after being scattered by the scene. The received echo signals are ; S2, perform de-slant reception, which is to directly mix the radar echo signal and the reference signal, and then perform Fourier transform on the mixed result to obtain the scene HRRP; S3. In order to obtain high-threat moving targets, the obtained scene HRRP data is passed through a moving target indication (MTI) filter to filter out a large number of static targets and ground clutter, making the moving targets more obvious and obtaining the moving target HRRP; S4. Use CA-CFAR to perform target detection on moving target HRRP data to determine the moving target and After the distance image position, intercept the moving target and HRRP data; S5. The moving targets respectively determined by radar A and radar B are fused in the distance space to obtain the target point, and the target point HRRP is identified by the neural network to determine whether the target is real or fake; when the HRRPs are judged to be the same target, the target point is a real target; when the HRRPs are judged to be different targets, the target point is a false target.

[0008] Preferably, the echo signal in step S1 It is expressed as: ; in, is the backscatter coefficient of the target, is the echo delay of the target point, Indicates fast time, that is, distance time, represents the pulse duration, To indicate a plural unit, represents the carrier frequency of the signal, Indicates the modulation frequency. Represents a rectangular window function.

[0009] Preferably, the reference signal constructed in step S2 is: ; in, Indicates the echo delay of the reference signal; The echo signal is mixed with the reference signal to obtain the processed echo signal: ; in, ; Then perform Fourier transform on the mixed result to obtain the scene HRRP. The scene HRRP expression is: ; in, are the target scattering points, and represent its real and imaginary parts respectively.

[0010] Preferably, the principle of step S3 is: since the stationary target and ground clutter have no velocity and no relative displacement with the radar, their Doppler frequency is zero, the phase of their echo signal does not change with time, and the amplitude in the spectrum does not change with time, while the phase of the moving target is constantly changing. Therefore, by subtracting the pulse echo with the same clutter component, the clutter and stationary target in the echo can be eliminated, and only the moving target is retained. Common MTI filters include two-pulse cancellation and three-pulse cancellation. The three-pulse cancellation is used to obtain the HRRP of the moving target, and its output signal for: ; Its impulse response is: ; in, Represents the input signal, that is, the current processing echo, Represents the echo of the adjacent pulse duration, represents the echo of the duration of two adjacent pulses, represents the impulse signal, Indicates delay The impulse signal, Indicates delay impulse signal.

[0011] Preferably, the input of the CA-CFAR detector in step S4 is a unit to be tested D, protection units, and The protection units are located on both sides of the unit under test. The reference unit is located on both sides of the protection unit. The interference power estimated by CA-CFAR is: ; Set threshold The interference power is multiplied by the scale factor Get, the expression is: ; Scale Factor The expected false alarm probability is given by Get, the expression is: ; Using CA-CAFR to identify moving targets and After reaching the position at the distance, intercept the moving target and HRRP data.

[0012] Preferably, step S5 comprises: S51, construct a one-dimensional convolutional neural network to train HRRP data; S52, inputting the obtained target HRRP data detected by the target point radar A and the radar B into the trained one-dimensional convolutional neural network to identify the envelope features of the HRRP data.

[0013] Preferably, step S51 specifically includes: capturing the moving target captured by radar A and radar B and The HRRP data are normalized and aligned to overcome the amplitude sensitivity and translation sensitivity of the HRRP data and collect moving target and moving target Measured data to build a moving target and HRRP data set. Since HRRP data is one-dimensional data and has translation sensitivity, and the global receptive field of convolutional neural network can solve the translation sensitivity problem, a one-dimensional convolutional neural network is constructed to train the HRRP data. Batch normalization layer and random dropout method are introduced into the one-dimensional convolutional neural network to improve the generalization of the model and avoid the network falling into overfitting state.

[0014] Preferably, in step S52, the obtained target HRRP data detected by the target point radar A and the radar B are input into the trained one-dimensional convolutional neural network, and the envelope characteristics of the HRRP data are identified to obtain the basic probability distribution of the HRRP of the target detected by the radar A and the radar B by the one-dimensional convolutional neural network, which is and , the target point is the category and categories The probability value is calculated by the softmax function, and the formula is: ; in, represents the number of categories of the neural network, is the output vector, for Middle The value of the category, Indicates the category that needs to be calculated currently; set category is 0, category If it is 1, then the one-dimensional convolutional neural network outputs 00 or 11 for a true target, and 01 or 10 for a false target.

[0015] Therefore, the present invention adopts the above-mentioned distributed radar target positioning method based on high-resolution range profile joint recognition, which has the following beneficial effects: (1) Recognize HRRP through neural network to solve the false target problem in distributed radar target positioning and complete target positioning accurately and efficiently; (2) Highly timely HRRP data makes it possible for the system to work in real time. Rapid and effective target positioning effectively solves the lack of environmental detection capability of vehicle-mounted radar when the vehicle is stationary, fills the gap in the working of vehicle-mounted radar, ensures its working ability in various situations, and provides an important basis for the next decision.

[0016] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a distributed radar target positioning method based on high-resolution range profile joint recognition of the present invention; Figure 2 This is a schematic diagram of a target point in an embodiment of the present invention; Figure 3 This is a schematic diagram of a false target point HRRP data pair according to an embodiment of the present invention; Figure 4 Schematic diagram of HRRP data pair of true target points according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] See also Figure 1 , a distributed radar target positioning method based on high-resolution range profile joint recognition includes the following steps: S1, radar A and radar B respectively transmit linear frequency modulation pulses to the detection area. The signals are received by the radar after being scattered by the scene. The received echo signals are , expressed as: ; in, is the backscatter coefficient of the target, is the echo delay of the target point, Indicates fast time, that is, distance time, represents the pulse duration, To indicate a plural unit, represents the carrier frequency of the signal, Indicates the modulation frequency. Represents a rectangular window function.

[0020] S2. Then the data is de-slanted and received, so as to directly mix the radar echo signal and the reference signal, and then Fourier transform the mixed result to obtain the scene HRRP; the constructed reference signal is: ; in, Indicates the echo delay of the reference signal; The echo signal is mixed with the reference signal to obtain the processed echo signal: ; in, ; Then perform Fourier transform on the mixed result to obtain the scene HRRP. The scene HRRP expression is: ; in, are the target scattering points, and represent its real and imaginary parts respectively.

[0021] S3. The scenes are often very complex. The echo signals often contain stationary targets, ground clutter, moving targets, etc. In order to obtain high-threat moving targets, the obtained scene HRRP data is passed through a moving target indication (MTI) filter to filter out a large number of stationary targets and ground clutter, making the moving targets more obvious. The principle is: since stationary targets and ground clutter have no speed and no relative displacement with the radar, their Doppler frequency is zero, and the phase of their echo signals does not change with time, and the amplitude in the spectrum does not change with time, while the phase of the moving target is constantly changing. Therefore, the clutter and stationary targets in the echo can be eliminated by subtracting the pulse echo with the same clutter component, leaving only the moving target. Common MTI filters include two-pulse cancellation and three-pulse cancellation. Three-pulse cancellation is used to obtain the moving target HRRP, and its output signal for: ; Its impulse response is: ; in, Represents the input signal, that is, the current processing echo, Represents the echo of the adjacent pulse duration, represents the echo of the duration of two adjacent pulses, represents the impulse signal, Indicates delay The impulse signal, Indicates delay impulse signal.

[0022] S4. Although the scene HRRP removes static targets and some ground clutter in the scene after MTI filtering, ground clutter cannot be completely filtered out. Due to the influence of atmospheric factors such as airflow, clutter is not always stationary. Therefore, it is necessary to detect and locate the target in the signal for subsequent feature extraction. The commonly used target detection method is CFAR detection. The main principle of CFAR detection is to continuously update the threshold value according to the estimated value of the noise variance, estimate the variance of the environmental noise through the units near the unit to be tested, and then calculate the threshold value according to the false alarm rate. Finally, the threshold value is compared with the unit to be tested to determine whether the unit to be tested is a target. Use CA-CFAR to perform target detection on the moving target HRRP data. The input of the CA-CFAR detector is a unit to be tested D. protection units, and The protection units are located on both sides of the unit under test. The reference unit is located on both sides of the protection unit. The interference power estimated by CA-CFAR is: ; Set threshold The interference power is multiplied by the scale factor Get, the expression is: ; Scale Factor The expected false alarm probability is given by Get, the expression is: ; Using CA-CAFR to identify moving targets and After reaching the position at the distance, intercept the moving target and HRRP data.

[0023] S5. The moving targets determined by radar A and radar B are fused in the distance space to obtain a target point schematic diagram, such as Figure 2 As shown, Figure 2 Contains four target points, including two true targets and two false targets, namely aa, bb, ab, and ba. Since the peaks and valleys in the target's HRRP reflect the relative geometric structure details of the target, they can be used as important information for target identification. The false target point is inconsistent with the targets detected by the two radars. Therefore, the target point HRRP can be identified by a neural network to determine whether the target is true or false. When the HRRPs are judged to be the same target, the target point is a true target; when the HRRPs are judged to be different targets, the target point is a false target. The false target point data is as follows Figure 3 As shown in Figure 2, the target HRRP data structures detected by the two radars are significantly different. Figure 4 As shown in the figure, the HRRP data structures of the targets detected by the two radars are basically the same, including: S51, intercept the moving target captured by radar A and radar B and The HRRP data are normalized and aligned to overcome the amplitude sensitivity and translation sensitivity of the HRRP data and collect moving target and moving target Measured data to build a moving target and HRRP data set. Since HRRP data is one-dimensional data and has translation sensitivity, and the global receptive field of convolutional neural network can solve the translation sensitivity problem, a one-dimensional convolutional neural network is constructed to train the HRRP data. Batch normalization layer and random dropout method are introduced into the one-dimensional convolutional neural network to improve the generalization of the model and avoid the network falling into overfitting state.

[0024] S52, input the obtained target HRRP data detected by radar A and radar B into the trained one-dimensional convolutional neural network, identify the envelope characteristics of the HRRP data, and obtain the basic probability distribution of the HRRP of the target detected by radar A and radar B by the one-dimensional convolutional neural network, which is and , the target point is the category and categories The probability value is calculated by the softmax function, and the formula is: ; in, represents the number of categories of the neural network, is the output vector, for Middle The value of the category, Indicates the category that needs to be calculated currently; set category is 0, category If it is 1, then the one-dimensional convolutional neural network outputs 00 or 11 for a true target, and 01 or 10 for a false target.

[0025] The method of this embodiment is verified by measured data, which is collected by millimeter wave radar and processed by MATLAB software. By identifying pedestrians and bicycles, the recognition accuracy of the neural network reaches 99%. The high accuracy of the neural network recognition and the high timeliness of the HRRP data ensure that the present invention can efficiently and accurately complete the distributed radar target positioning.

[0026] Therefore, the present invention adopts the above-mentioned distributed radar target positioning method based on high-resolution range image joint recognition. In view of the situation that the single-channel radar lacks angle information and cannot accurately locate the target, the high-resolution distributed radar is used to detect the target for target distance positioning, and then the target geometric structure information contained in the target HRRP is used to perform neural network recognition and distinguish between true targets and false targets, so as to complete the task of accurate target positioning. It effectively solves the disadvantage of insufficient target positioning capability of radar when the platform is stationary, and makes up for the environmental detection function of synthetic aperture radar anytime and anywhere; at the same time, the high timeliness of HRRP data makes it possible for the system to work in real time.

[0027] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A distributed radar target positioning method based on high-resolution range profile joint recognition, characterized in that: The following steps are involved: S1, radar A and radar B respectively transmit linear frequency modulation pulses to the detection area. The signals are received by the radar after being scattered by the scene. The received echo signals are ; S2, perform de-slant reception, which is to directly mix the radar echo signal and the reference signal, and then perform Fourier transform on the mixed result to obtain the scene HRRP; S3, passing the obtained scene HRRP data through a moving target indication filter to obtain a moving target HRRP; S4. Use CA-CFAR to perform target detection on moving target HRRP data to determine the moving target and After the distance image position, intercept the moving target and HRRP data; S5. The moving targets respectively determined by radar A and radar B are fused in the distance space to obtain the target point, and the target point HRRP is identified by the neural network to determine whether the target is real or fake; when the HRRPs are judged to be the same target, the target point is a real target; when the HRRPs are judged to be different targets, the target point is a false target.

2. The distributed radar target positioning method based on high-resolution range profile joint recognition according to claim 1 is characterized in that: The echo signal in step S1 It is expressed as: ； in, represents the backscattering coefficient of the target, Indicates the echo delay of the target point, Indicates fast time, represents the pulse duration, To indicate a plural unit, represents the carrier frequency of the signal, Indicates the modulation frequency. Represents a rectangular window function.

3. The distributed radar target positioning method based on high-resolution range profile joint recognition according to claim 2 is characterized in that: The reference signal constructed in step S2 is: ； in, Indicates the echo delay of the reference signal; The echo signal is mixed with the reference signal to obtain the processed echo signal: ； in, ; Then perform Fourier transform on the mixed result to obtain the scene HRRP, which is expressed as: ； in, are the target scattering points, and represent its real and imaginary parts respectively.

4. The distributed radar target positioning method based on high-resolution range profile joint recognition according to claim 1 is characterized in that: In step S3, by subtracting the pulse echo with the same clutter component, the clutter and stationary targets in the echo are eliminated, and the moving targets are retained. The moving target indication filter is a three-pulse cancellation, and its output signal for: ； Its impulse response is: ； in, represents the input signal, Represents the echo of the adjacent pulse duration, represents the echo of the duration of two adjacent pulses, represents the impulse signal, Indicates delay The impulse signal, Indicates delay impulse signal.

5. The distributed radar target positioning method based on high-resolution range profile joint recognition according to claim 3 is characterized in that: The input of the CA-CFAR detector in step S4 is a unit to be tested D, protection units, and The protection units are located on both sides of the unit under test. The reference unit is located on both sides of the protection unit. The interference power estimated by CA-CFAR is: ； Set threshold The interference power is multiplied by the scaling factor Get, the expression is: ； Scale Factor The expected false alarm probability is given by Get, the expression is: ； Using CA-CAFR to identify moving targets and After reaching the position at the distance, intercept the moving target and HRRP data.

6. The distributed radar target positioning method based on high-resolution range profile joint recognition according to claim 1 is characterized in that: Step S5 includes: S51, construct a one-dimensional convolutional neural network to train HRRP data; S52, inputting the obtained target HRRP data detected by the target point radar A and the radar B into the trained one-dimensional convolutional neural network to identify the envelope features of the HRRP data.

7. The distributed radar target positioning method based on high-resolution range profile joint recognition according to claim 6 is characterized in that: Step S51 specifically includes: capturing the moving target captured by radar A and radar B and The HRRP data of the moving target are normalized and aligned. and moving target Measured data to build a moving target and A one-dimensional convolutional neural network is constructed to train the HRRP data set, and batch normalization layers and random dropout methods are introduced into the one-dimensional convolutional neural network.

8. The distributed radar target positioning method based on high-resolution range profile joint recognition according to claim 7 is characterized in that: Step S52 inputs the obtained target HRRP data detected by radar A and radar B into the trained one-dimensional convolutional neural network, identifies the envelope features of the HRRP data, and obtains the basic probability distribution of the HRRP of the target detected by radar A and radar B by the one-dimensional convolutional neural network, which is and , the target point is the category and categories The probability value is calculated by the softmax function, and the formula is: ； in, represents the number of categories of the neural network, is the output vector, for Middle The value of the category, Indicates the category that needs to be calculated currently; set category is 0, category If it is 1, then the one-dimensional convolutional neural network outputs 00 or 11 for a true target, and 01 or 10 for a false target.

Images