Method and system for correcting amplitude and phase fluctuation of range profile of dechirp receiving broadband imaging radar

Abstract

The invention discloses a dechirp receiving broadband imaging radar range profile amplitude and phase fluctuation correction method and system, and the method comprises the steps: obtaining actual measurement radar echo data containing amplitude and phase errors and corresponding simulated ideal undistorted echo data through actual measurement and simulation, and taking the data as a group of radar echo data; performing dechirp processing on each group of radar echo data, and then performing Fourier transform to obtain a one-dimensional range profile in a complex form; real number processing is carried out on the one-dimensional distance image in the complex number form, a real number vector maintaining complex number information is constructed, and labeling is carried out to form a training set; taking the training set as the input of a deep learning network, carrying out supervised training, and outputting to obtain a real number vector; and reconstructing a real number vector output by the deep learning network to obtain a corrected complex range profile. According to the invention, the sidelobe distortion error of the range profile with the shift change characteristic is compensated.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, specifically to a method and system for correcting phase fluctuations in range image amplitude of a deslant receiving broadband imaging radar. Background Technology

[0002] Range resolution of imaging radar for: ; in, At the speed of light, This refers to the bandwidth of the imaging radar pulse. To achieve higher precision imaging, the instantaneous operating bandwidth of imaging radar systems is becoming increasingly wider. In the millimeter-wave and terahertz frequency bands, the operating bandwidth of imaging radar can reach several GHz or even tens of GHz.

[0003] However, it's not just the bandwidth of the imaging radar system that affects image quality; amplitude and phase nonlinear fluctuations in the radar system's radio frequency channel can also lead to a decrease in image quality. For example... Figure 1 As shown, a one-dimensional range pulse compression imaging simulation is performed using a linear frequency modulated (LFM) signal. Figure 1 The left figure shows the pulse compression result of an ideal LFM signal. Figure 1 The right figure shows the pulse compression result of a distorted LFM signal affected by nonlinear fluctuations in channel amplitude and phase. Nonlinear fluctuations in channel amplitude and phase lead to amplitude and phase distortion of the LFM signal, resulting in problems such as main lobe widening, sidelobe elevation, and stray echoes in the one-dimensional range image after pulse compression.

[0004] Currently, amplitude and phase correction of RF channels is mainly achieved through broadband digital acquisition to digitally correct channel amplitude and phase fluctuations. Taking the least squares-based channel amplitude and phase fluctuation correction as an example, the main implementation process is as follows: S1, inject the ideal LFM signal into the digital transmission channel to obtain a distorted LFM signal with channel amplitude and phase fluctuations.

[0005] S2, use a broadband digital acquisition board to digitally acquire the distorted LFM signal to obtain a digitized LFM signal.

[0006] S3. Divide the frequency response of the transmit channel by the frequency domain of the ideal LFM signal and the digitized distorted LFM signal.

[0007] S4. Construct the frequency factor matrix and fit it using the frequency domain least squares method to obtain the FIR filter coefficients used for channel amplitude-phase equalization.

[0008] S5 uses an amplitude-phase equalization FIR filter to pre-distort the transmitted LFM signal, resulting in an amplitude-phase corrected LFM signal output.

[0009] The amplitude and phase fluctuations in the broadband RF transceiver channel and the local oscillator signal channel cause the de-skewed output signal (radar acquisition signal) to exhibit nonlinear shift characteristics. Current RF transceiver systems are mostly based on digital technology for signal generation and acquisition. Therefore, methods using broadband digital acquisition and digital predistortion techniques to correct in-band amplitude and phase fluctuations are highly versatile. However, in ultra-wideband imaging radar systems, as mentioned above, it is necessary to correct the RF transceiver channel and the local oscillator signal channel separately before de-skewing the radar acquisition signal. This usually requires modifications to the radar hardware structure for reception, and is costly and complex. Methods exist for directly correcting the nonlinear shift characteristics of the de-skewed output signal, but because the error has shift characteristics, these methods employ piecewise amplitude and phase error compensation.

[0010] To achieve better imaging results, it is necessary to correct the in-band amplitude and phase fluctuations of the radio frequency (RF) channel. However, in an imaging radar system with analog de-skewing reception, if both the RF channel signal and the reference (mixing) signal have in-band amplitude and phase fluctuations, the output signal after de-skewing will exhibit nonlinear shift characteristics. Traditional methods use segmented compensation for different distances to compensate the signal to a certain extent, but the segment size affects the restoration performance, system cost, and complexity, requiring a trade-off and presenting drawbacks.

[0011] In recent years, deep learning technology has demonstrated powerful capabilities in image restoration and denoising, and shows great potential for application in radar signal processing. However, most neural networks currently only support real-valued inputs and outputs. In imaging radar systems that require deskewing, the original received radar signal is a real number, but its meaning for target features is not obvious, making it difficult for networks to learn these features. Therefore, it is necessary to perform a Fourier transform on the original received radar signal to obtain a one-dimensional range image before feeding it to the neural network for processing. The target feature meaning of the one-dimensional range image is obvious and suitable for network learning. For radar imaging algorithms, such as back projection (BP) and range-Doppler (RD) algorithms, a complex one-dimensional range image is required, which neural networks that only support real-valued inputs and outputs cannot process. Furthermore, whether deep learning methods can learn the channel impairments of deskewing radar, whether they have generalization compensation capabilities, and whether they can compensate for nonlinear shift characteristics of the signal still present certain challenges. Currently, most neural networks support real-valued signal inputs and outputs, and neural networks that support complex-valued signal inputs and outputs often require special modifications, resulting in high costs and complexity.

[0012] The invention patent with patent application publication number CN119846622A discloses a non-orthogonal waveform MIMO radar imaging method based on deep learning. This method obtains the target echo signal across the entire imaging scene; performs waveform separation on the target echo signal of the entire imaging scene to obtain waveform vectors corresponding to different transmitted waveforms, and performs azimuth compression to obtain an ideal target image for the MIMO radar; uses the frequency domain echo signals at different receiving elements in the MIMO radar echo signal model and the ideal target image as a training set to train the deep network model; and uses the trained deep network model to perform target imaging on the echo signals at different receiving elements. Through waveform separation and azimuth compression, this method helps improve the imaging resolution of the MIMO radar, thereby depicting target details more precisely. This patent primarily addresses coupling interference between multiple channels in MIMO radar. The training set requires ideal target images from the MIMO radar, necessitating the construction of complex scattering point models. However, this is often difficult to obtain in practice because the RCS (radar cross section) scattering characteristics of real targets are typically very complex. It usually requires data obtained through a combination of theoretical calculations, anechoic chamber static measurements, field static measurements, and field dynamic measurements to obtain a detailed description and feature analysis of the target's RCS. Furthermore, this patent modifies the neural network, directly inputting the real and imaginary parts of the echo transform as dual-channel inputs, resulting in high cost and complexity. Summary of the Invention

[0013] The technical problem to be solved by this invention is to solve the problem of nonlinear shift characteristics distortion of range image caused by amplitude and phase fluctuations in a single radar channel, and to propose a real-to-digital conversion method for complex signals of imaging radar range image. This method preserves the characteristics of echo signals while allowing real neural networks to support complex signal input and output, and is low in cost and complexity.

[0014] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for correcting phase fluctuations in range image amplitude of a deslanting receiving broadband imaging radar includes: Through actual measurement and simulation, the measured radar echo data containing amplitude and phase errors and its corresponding simulated ideal distortion-free echo data are obtained and used as a set of radar echo data. After deskewing each set of radar echo data, a Fourier transform is performed to obtain a one-dimensional range image in complex form. The one-dimensional distance image in complex form is converted to real number form to construct a real vector that preserves the complex information. They are then labeled to form a training set; The training set is used as input to a deep learning network for supervised training, and the output is a real-valued vector. ; The real-number vector output by the deep learning network Reconstruct the image to obtain the corrected complex distance image.

[0015] In this embodiment, the one-dimensional distance image in complex form is converted to real number to construct a real vector that preserves the complex information. ,include: Based on the fact that the spectrum of a one-dimensional range profile in complex form satisfies conjugate symmetry, and that effective information is fully contained in the points within the first N / 2 echo signals, the real and imaginary parts of the semispectral sequence of the one-dimensional range profile in complex form are sequentially concatenated to form a real vector. .

[0016] In this embodiment, real number vector Expressed using the following formula: ; In the formula, It is a sequence of the real parts of the semispectral data. It is a semi-spectral imaginary part sequence. For points in the echo signal, This represents the total number of points in the echo signal.

[0017] In this embodiment, the real-number vector output by the deep learning network is... Reconstruction is performed to obtain the corrected complex range image, including: From real number vectors The real part numerical sequence needs to be extracted from each part. and imaginary part numerical sequence ;in, For semi-spectral indexing, ; The sequence of real part values and imaginary part numerical sequence Combined into a corrected half-spectral complex signal; The full spectrum is recovered from the corrected half-spectrum complex signal using conjugate symmetry, thus obtaining the corrected complex range image.

[0018] In this embodiment, the corrected complex range image is represented by the following formula: ; In the formula, This is the corrected complex range image. For frequency units, This is the corrected half-spectral complex range image. This is a complex conjugate operation. This is a mirror image of the corrected half-spectral complex range image. This represents the total number of points in the echo signal.

[0019] In this embodiment, for real number vectors that retain complex information Labeling is defined as: the real vector corresponding to the measured radar echo data. Marked as measured labels, the real vector corresponding to the ideal distortion-free echo data is... , marked as the target label.

[0020] In this embodiment, the training set is used as input to the deep learning network for supervised training, including: By minimizing the loss function between the deep learning network output and the target label, the network parameters of the deep learning network are optimized, enabling the deep learning network to learn to recover an undistorted ideal one-dimensional distance image from distorted input.

[0021] In this embodiment, the deep learning network is the U-Net deep learning network.

[0022] In this embodiment, the deskewing process for each set of radar echo data is performed by mixing each set of radar echo data with the local oscillator reference.

[0023] The present invention also provides a system for correcting range image amplitude phase fluctuations according to the above-described deslant receiving broadband imaging radar range image amplitude phase fluctuation correction method, comprising: The echo data module is used to obtain measured radar echo data containing amplitude and phase errors and their corresponding simulated ideal distortion-free echo data through actual measurement and simulation, and to form a set of radar echo data. The Fourier transform module is used to perform Fourier transform on each set of radar echo data after deskewing to obtain a one-dimensional range image in complex form. The realization module is used to convert a one-dimensional distance image in complex form into a real vector that preserves the complex information. They are then labeled to form a training set; The training module is used to take the training set as input to the deep learning network, perform supervised training, and output real-valued vectors. ; The spectrum reconstruction module is used to convert the real-valued vectors output by the deep learning network. Reconstruct the image to obtain the corrected complex distance image.

[0024] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a radar one-dimensional range image amplitude and phase error correction method based on U-Net deep learning network. This method can learn distorted range images from a small number of scattering points and then generalize to compensate for amplitude and phase fluctuations in range images of complex models, effectively suppressing sidelobes of one-dimensional range images and significantly improving the quality of range Doppler two-dimensional images.

[0025] This invention addresses the distortion of range profiles caused by nonlinear shifts in amplitude and phase fluctuations within a single radar channel. The training set contains only echo data from a small number (1-3) of scattering points, enabling the repair and further two-dimensional imaging of range profiles from targets with complex scattering points. The training set construction method is simple, leveraging the generalization ability of neural networks.

[0026] This invention applies the U-Net deep learning network to amplitude and phase fluctuation correction in broadband deslant receiving imaging radar. By utilizing the network to learn channel impairments, it avoids the complex extraction of channel impairment correction features required by traditional algorithms and solves the problem of traditional algorithms struggling to compensate for errors with shift-dependent characteristics. Furthermore, this invention leverages the generalization performance advantage of the U-Net deep learning network; the U-Net deep learning network can repair echoes with a large number of scattering targets by learning only echoes with a small number of scattering points.

[0027] This invention utilizes the advantages of deskewing receivers to directly process the output signal after deskewing, eliminating the need for high-speed sampling to correct amplitude and phase fluctuations in broadband signals.

[0028] This invention addresses the sidelobe distortion of range images caused by channel amplitude-phase fluctuations, a problem that traditional methods struggle to compensate for, by applying the U-Net deep learning network to the error compensation of the output signal of a deslanting imaging radar. Through real-valued preprocessing of the training dataset using the Fast Fourier Transform, data with a smaller number of scattering points is used as the training set to compensate for amplitude-phase fluctuations in the range images and range-Doppler images of targets with simple or complex scattering points. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of a one-dimensional distance pulse compression imaging simulation of an ideal LFM signal and a distorted LFM signal.

[0030] Figure 2 This is a flowchart of a method for correcting phase fluctuations in range image amplitude of a broadband imaging radar according to an embodiment of the present invention.

[0031] Figure 3 This is a schematic diagram of the amplitude and phase fluctuations of the deslant receiving broadband imaging radar echo in an embodiment of the present invention.

[0032] Figure 4 This is a flowchart illustrating the process of extracting and realizing the effective semi-spectrum according to an embodiment of the present invention.

[0033] Figure 5 This is a flowchart of the complex spectrum reconstruction process according to an embodiment of the present invention.

[0034] Figure 6 This is a schematic diagram of the U-Net network structure according to an embodiment of the present invention.

[0035] Figure 7This is a single scattering point error compensation diagram according to an embodiment of the present invention.

[0036] Figure 8 This is a multi-scattering point error compensation diagram according to an embodiment of the present invention. Detailed Implementation

[0037] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0038] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0039] Please see Figure 2 As shown, the present invention provides a method for correcting phase fluctuations in range image amplitude of a deslanting receiving broadband imaging radar, comprising: S10, through actual measurement and simulation, obtains the measured radar echo data containing amplitude and phase errors and its corresponding simulated ideal distortion-free echo data, and uses them as a set of radar echo data.

[0040] Please see Figure 2 , 3 As shown, in one embodiment of the present invention, an imaging radar sends an LFM signal toward a target, which reflects the signal back, acquiring measured radar echo data containing amplitude and phase errors. The measured radar echo data may include a scene containing a small number of scattering points to control the complexity of data generation and ensure the network can learn the core error features.

[0041] In this embodiment, the scattering point is defined as follows: if the target is a "point-like" strong reflector (such as a corner reflector, a small metal object, etc.), the echo appears as a signal returning from a "point," which is a point target. In real-world scenarios, complex targets (such as aircraft, vehicles, and buildings) can be considered as being composed of many strong scattering centers. Each scattering center corresponds to a local structure on the target with strong electromagnetic wave reflection (such as edges, corners, vertical surfaces of flat plates, etc.). The scattering point is an idealized model of these scattering centers, representing that the echo mainly originates from these discrete spatial locations. This can be understood as radar echo data originating from an observation scenario with a relatively small number of scattering points. In this embodiment, radar system simulation is applied to compare the measured radar echo data with the corresponding simulated ideal distortion-free echo data.

[0042] S20: After deskewing each set of radar echo data, a Fourier transform is performed to obtain a one-dimensional range image in complex form.

[0043] Please see Figure 2 , 3 As shown, in one embodiment of the present invention, the deskewing process, i.e., the demodulation frequency modulation, specifically, mixes each group of radar echo data with the local oscillator reference.

[0044] In this embodiment, for the de-skewed real-time signal Perform a Fast Fourier Transform (FFT) to obtain a one-dimensional range image in complex form. The complex spectrum can be represented by the real part. and the virtual part Combinations: ; in, The imaginary unit, For time. And, f =0,1,2,…, N -1 represents a frequency unit. and All are real functions. Due to the input signal... For a real signal, its spectrum satisfies conjugate symmetry: In the formula, for . conjugate.

[0045] S30, the one-dimensional distance image in complex form is converted to real number form to construct a real number vector that preserves the complex information. They are then labeled to form a training set.

[0046] Please see Figures 2 to 4 As shown, in one embodiment of the present invention, since the radar directly acquires real signals and does not display their range profile features, preprocessing is required before inputting them into the neural network for processing, so that the neural network can learn the features. The output of the neural network also needs to be post-processed to recover the radar signal. Specifically, the one-dimensional range profile in complex form is converted into a real number, constructing a real vector that retains the complex information. ,include: Based on the fact that the spectrum of a one-dimensional range profile in complex form satisfies conjugate symmetry, and that effective information is fully contained in the points within the first N / 2 echo signals, the real and imaginary parts of the semispectral sequence of the one-dimensional range profile in complex form are sequentially concatenated to form a real vector. .

[0047] In this embodiment, during the process of extracting the effective semispectrum and converting it to a real number, the following is defined: n=0,1,…,N / 2-1, the real part sequence of the half-spectrum and imaginary part sequence Sequentially pieced together to form a length of N A real vector, used as the subsequent Network input real vector Expressed using the following formula: ; In the formula, It is a sequence of the real parts of the semispectral data. It is a semi-spectral imaginary part sequence. For points in the echo signal, This represents the total number of points in the echo signal.

[0048] In this embodiment, for real number vectors that retain complex information Labeling is defined as: the real vector corresponding to the measured radar echo data. Marked as measured labels, the real vector corresponding to the ideal distortion-free echo data is... , marked as the target label.

[0049] In this embodiment, the above can also be understood as assigning the first half of the FFT result to the real part of the complex spectrum and the second half of the result to the imaginary part of the complex spectrum, thereby constructing a real vector that retains all complex information, which serves as the input and output labels of the deep learning network.

[0050] S40 uses the training set as input to the deep learning network for supervised training, and outputs a real-valued vector. .

[0051] Please see Figures 2 to 6 As shown, in one embodiment of the present invention, the deep learning network is specifically an L-layer U-Net deep learning network. Real vectors based on measured radar echo data in the training set are used as inputs to the U-Net deep learning network, and real vectors corresponding to ideal, distortion-free echo data are used as training targets to perform supervised training on the U-Net deep learning network. This invention does not improve upon the existing U-Net deep learning network; it simply applies the existing U-Net deep learning network.

[0052] In this embodiment, during the training process, the network parameters of the deep learning network are optimized by minimizing the loss function between the output of the deep learning network and the target label, so that the deep learning network can learn to recover the distortion-free ideal one-dimensional distance image from the distorted input.

[0053] S50 represents the real-valued vector output by the deep learning network. Reconstruct the image to obtain the corrected complex distance image.

[0054] Please see Figures 2 to 6 As shown, in one embodiment of the present invention, the real-number vector output by the deep learning network is... Reconstruction is performed to obtain the corrected complex range image, including: From real number vectors The real part numerical sequence needs to be extracted from each part. and imaginary part numerical sequence : ; In the formula, For semi-spectral indexing, .

[0055] The sequence of real part values and imaginary part numerical sequence Combined into a corrected half-spectral complex signal: ; In the formula, This is the corrected half-spectral complex signal.

[0056] The full spectrum is recovered by utilizing conjugate symmetry of the corrected half-spectrum complex signal, i.e., the corrected complex range image is obtained: ; In the formula, This is the corrected complex range image. This is the corrected half-spectral complex range image. This is a complex conjugate operation. This is the mirror image of the corrected half-spectral complex distance image.

[0057] In this embodiment, the real vector output by the U-Net deep learning network is recombined into a complex signal according to the rule that the first half is the real part and the second half is the imaginary part, so that a complex range image with significantly suppressed sidelobes and distortion can be obtained after correction.

[0058] Please see Figure 7 and Figure 8 As shown, in one embodiment of the present invention, by simulating the amplitude and phase fluctuations of the receiving channel and the reference (mixing) channel, a certain amount of measured radar distortion data of 1 to 3 scattering point targets is simulated. Simultaneously, ideal data without amplitude and phase fluctuations is simulated as labels to train the U-Net deep learning network. After the U-Net deep learning network is trained, its performance is tested. The U-Net deep learning network is used to repair distorted ISAR (Inverse Synthetic Aperture Radar) echo sequences of single and multiple scattering point targets, and the distortion correction of the range image and the corresponding range Doppler image correction results are observed, as shown in the figure. Figure 7 This is a single-scattering point error compensation diagram. Figure 7 In the middle, (a)~(c) represent the distortion, U-Net deep learning network inpainting, and ideal distance image, respectively, and (d)~(f) represent the corresponding distance Doppler images. Figure 8 This is a multi-scattering point error compensation diagram. Figure 8 In the image, (a) to (c) show the distorted image, the image repaired by the U-Net deep learning network, and the idealized image, respectively; (d) to (f) show the corresponding range-Doppler images. And, Figure 7 , 8 Figure (a) shows the distorted one-dimensional range image matrix, where the horizontal axis represents distance and the vertical axis represents the echo sequence number. Figure (d) is the distorted two-dimensional range-Doppler image obtained by performing an FFT on the vertical axis of Figure (a), where the horizontal axis represents distance and the vertical axis represents Doppler. Figures (b) and (e) show the one-dimensional range image matrix and two-dimensional range-Doppler image after restoration according to the present invention. Figures (c) and (f) show the ideal one-dimensional range image matrix and range-Doppler image, respectively.

[0059] from Figure 7 , Figure 8 As can be seen from the results, after the repair by the present invention, for cases where the training set is similar (single scattering point error compensation) and cases where the training set is not present (5 scattering point error compensation), the one-dimensional range image matrix and the two-dimensional range-Doppler image are close to the ideal results. This shows that the U-Net deep learning network not only learns the channel impairment and compensation method, but also has a certain generalization ability.

[0060] The present invention also provides a system for the aforementioned method for correcting phase fluctuations in range image amplitude of a broadband imaging radar, comprising: The echo data module is used to acquire measured radar echo data containing amplitude and phase errors and their corresponding simulated ideal distortion-free echo data through actual measurement and simulation, and to form a set of radar echo data.

[0061] The Fourier transform module is used to perform Fourier transform on each set of radar echo data after deskewing to obtain a one-dimensional range image in complex form.

[0062] The realization module is used to convert a one-dimensional distance image in complex form into a real vector that preserves the complex information. They are then labeled to form a training set.

[0063] The training module is used to take the training set as input to the deep learning network, perform supervised training, and output real-valued vectors. .

[0064] The spectrum reconstruction module is used to convert the real-valued vectors output by the deep learning network. Reconstruct the image to obtain the corrected complex distance image.

[0065] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0066] The above-described embodiments are merely examples of implementation methods of the invention. The scope of protection of the present invention is not limited to the above-described embodiments. For those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for correcting phase fluctuations in range image amplitude of a deslanting receiving broadband imaging radar, characterized in that, include: Through actual measurement and simulation, the measured radar echo data containing amplitude and phase errors and its corresponding simulated ideal distortion-free echo data are obtained and used as a set of radar echo data. After deskewing each set of radar echo data, a Fourier transform is performed to obtain a one-dimensional range image in complex form. The one-dimensional distance image in complex form is converted to real number form to construct a real vector that preserves the complex information. They are then labeled to form a training set; The training set is used as input to a deep learning network for supervised training, and the output is a real-valued vector. ; The real-number vector output by the deep learning network Reconstruct the image to obtain the corrected complex distance image.

2. The method for correcting phase fluctuations in range image amplitude of a broadband imaging radar according to claim 1, characterized in that, The one-dimensional distance image in complex form is converted to real number form to construct a real vector that preserves the complex information. ,include: Based on the fact that the spectrum of a one-dimensional range profile in complex form satisfies conjugate symmetry, and that effective information is fully contained in the points within the first N / 2 echo signals, the real and imaginary parts of the semispectral sequence of the one-dimensional range profile in complex form are sequentially concatenated to form a real vector. .

3. The method for correcting range image amplitude phase fluctuations in broadband imaging radar according to claim 2, characterized in that, real vector Expressed using the following formula: ； In the formula, It is a sequence of the real parts of the semispectral data. It is a semi-spectral imaginary part sequence. For points in the echo signal, This represents the total number of points in the echo signal.

4. The method for correcting range image amplitude phase fluctuations in a broadband imaging radar according to claim 1, characterized in that, The real-number vector output by the deep learning network Reconstruction is performed to obtain the corrected complex range image, including: From real number vectors The real part numerical sequence needs to be extracted from each part. and imaginary part numerical sequence ;in, For semi-spectral indexing; The sequence of real part values and imaginary part numerical sequence Combined into a corrected half-spectral complex signal; The full spectrum is recovered from the corrected half-spectrum complex signal using conjugate symmetry, thus obtaining the corrected complex range image.

5. The method for correcting phase fluctuations in range image amplitude of a broadband imaging radar according to claim 4, characterized in that, The corrected complex range image is represented by the following formula: ； In the formula, This is the corrected complex distance image. For frequency units, This is the corrected half-spectral complex range image. This is a complex conjugate operation. This is a mirror image of the corrected half-spectral complex range image. This represents the total number of points in the echo signal.

6. The method for correcting range image amplitude phase fluctuations in a broadband imaging radar according to claim 1, characterized in that, For real vectors that preserve complex information Labeling is defined as: the real vector corresponding to the measured radar echo data. Marked as measured labels, the real vector corresponding to the ideal distortion-free echo data is... , marked as the target label.

7. The method for correcting range image amplitude phase fluctuations in a broadband imaging radar according to claim 6, characterized in that, Supervised training is performed by using the training set as input to a deep learning network, including: By minimizing the loss function between the deep learning network output and the target label, the network parameters of the deep learning network are optimized, enabling the deep learning network to learn to recover an undistorted ideal one-dimensional distance image from distorted input.

8. The method for correcting phase fluctuations in range image amplitude of a broadband imaging radar according to claim 7, characterized in that, The deep learning network is the U-Net deep learning network.

9. The method for correcting phase fluctuations in range image amplitude of a broadband imaging radar according to claim 7, characterized in that, The deskewing process for each set of radar echo data is as follows: each set of radar echo data is mixed with the local oscillator reference.

10. A system for correcting phase fluctuations in range image amplitude of a broadband imaging radar according to any one of claims 1-9, characterized in that, include: The echo data module is used to obtain measured radar echo data containing amplitude and phase errors and their corresponding simulated ideal distortion-free echo data through actual measurement and simulation, and to form a set of radar echo data. The Fourier transform module is used to perform Fourier transform on each set of radar echo data after deskewing to obtain a one-dimensional range image in complex form. The realization module is used to convert a one-dimensional distance image in complex form into a real vector that preserves the complex information. They are then labeled to form a training set; The training module is used to take the training set as input to the deep learning network, perform supervised training, and output real-valued vectors. ; The spectrum reconstruction module is used to convert the real-valued vectors output by the deep learning network. Reconstruct the image to obtain the corrected complex distance image.

Images