Ultra-wideband radar range profile enhancement method for target-oriented amplitude flicker loss scene
By combining stepped-frequency ultra-wideband radar and deep networks, the problems of target amplitude flicker and data loss in complex scenarios were solved, achieving high-quality range image reconstruction and ensuring the stability and accuracy of imaging and positioning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- QIANYUAN NATIONAL LABORATORY
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-05
AI Technical Summary
In complex scenarios, existing technologies based on DFT or IDF for range image reconstruction are affected by changes in target pose, scattering center occlusion, and azimuth angle, resulting in significant target amplitude flickering. This makes it impossible to provide stable, high-fidelity range image input, which affects subsequent imaging and positioning results.
Frequency domain echo data is acquired by step-frequency ultra-wideband radar. Near-end iterative solution is performed using an iterative expansion reconstruction network. Compensation enhancement is performed by combining an enhancement network. An imaging objective function with joint low-rank-sparse constraints is established to suppress sidelobe lift and residual static clutter, thereby achieving high-quality range image reconstruction.
Under complex conditions such as target amplitude flicker, data loss, and strong static background, the continuity of the target main lobe morphology and range-time trajectory is maintained, achieving high-resolution range image reconstruction and providing stable, high-fidelity input for subsequent imaging and positioning.
Smart Images
Figure CN122151071A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of radar signal processing technology, and in particular to an ultra-wideband radar range image enhancement method for scenarios with missing target amplitude scintillation. Background Technology
[0002] Ultra-wideband radar can obtain high-resolution range profiles (HRRPs) of detected targets by emitting broadband electromagnetic waves and performing high-resolution time delay estimation on the echoes. It is now widely used in scenarios such as vital sign detection, concealed target localization, and behavior recognition of targets behind walls. High-quality range profiles are not only the foundation for subsequent imaging and localization processing, but also a key input for higher-level tasks such as target detection, trajectory estimation, and behavior analysis.
[0003] Currently, related technologies are usually based on Discrete Fourier Transform (DFT) or its inverse transform (IDFT), that is, applying DFT / IDFT to frequency domain samples with uniform intervals to obtain HRRP.
[0004] However, in complex scenarios, the range image reconstruction methods based on DFT or IDF often suffer from significant target amplitude flickering due to factors such as target pose changes, scattering center occlusion, and azimuth angle changes. This makes it impossible to provide stable and high-fidelity range image input for subsequent imaging and positioning applications. Summary of the Invention
[0005] In view of this, this application provides an ultra-wideband radar range image enhancement method for scenarios with missing target amplitude flicker. The main purpose is to improve the range image reconstruction methods based on DFT or IDF in related technologies. In complex scenarios, the range images actually measured are affected by factors such as target attitude changes, scattering center obstruction, and azimuth angle changes, and usually have obvious target amplitude flicker, which cannot provide stable and high-fidelity range image input for subsequent imaging and positioning applications.
[0006] Firstly, this application provides an ultra-wideband radar range image enhancement method for scenarios with missing target amplitude scintillation, the method comprising: The step-frequency ultra-wideband radar is used to detect targets in the detection scene and acquire frequency domain echo data corresponding to the targets. Based on the iterative unfolding reconstruction network, the imaging objective function corresponding to the frequency domain echo data is solved by near-end iterative solution to obtain the coarse reconstructed range image corresponding to the detected target; the imaging objective function includes a function established under the constraints of the imaging model, which combines the low-rank characteristics of strong static echo between frames and the sparsity characteristics of the target range image in the range-time domain. The coarse reconstructed range image is enhanced by an enhancement network to generate an enhanced range image corresponding to the detected target. The enhancement network is used to compensate for the trajectory hole of the detected target under the conditions of amplitude flicker and sampling loss, and to suppress the sidelobe lift of the detected target and the residual static clutter of the detected scene.
[0007] Secondly, this application provides an ultra-wideband radar range image enhancement device for scenarios with missing target amplitude scintillation, the device comprising: The acquisition module is configured to detect targets in the detection scene using a stepped-frequency ultra-wideband radar and acquire frequency domain echo data corresponding to the targets. The solution module is configured to perform near-end iterative solution of the imaging objective function corresponding to the frequency domain echo data based on the iterative unfolding reconstruction network, and obtain the coarse reconstructed range image corresponding to the detected target; the imaging objective function includes a function established under the constraints of the imaging model, which combines the low-rank characteristics of strong static echo between frames and the sparsity characteristics of the target range image in the range-time domain; An enhancement module is configured to compensate and enhance the coarse reconstructed range image through an enhancement network to generate an enhanced range image corresponding to the detected target. The enhancement network is used to compensate for trajectory holes of the detected target under the conditions of amplitude flicker and sampling loss of the detected target, and to suppress sidelobe lift of the detected target and residual static clutter of the detected scene.
[0008] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of the first aspect.
[0009] Fourthly, this application provides an electronic device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the method of the first aspect.
[0010] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method of the first aspect.
[0011] By employing the above technical solution, this application provides an ultra-wideband radar range image enhancement method for scenarios with missing target amplitude scintillation. Compared with related technologies, this application uses a stepped-frequency ultra-wideband radar to detect targets in a detection scenario and acquires frequency domain echo data corresponding to the targets. Based on an iterative unfolding reconstruction network, the imaging objective function corresponding to the frequency domain echo data is solved iteratively at the near end to obtain a coarse reconstructed range image corresponding to the targets. The imaging objective function includes a function established under the constraints of the imaging model, combining the low-rank characteristics of strong static echoes between frames and the sparsity characteristics of the target range image in the range-time domain. The coarse reconstructed range image is compensated and enhanced by an enhancement network to generate an enhanced range image corresponding to the targets. The enhancement network is used to compensate for trajectory holes of the targets under the conditions of target amplitude scintillation and sampling loss, and to suppress sidelobe rise of the targets and residual static clutter in the detection scenario.
[0012] In this way, this application can introduce the low-rank characteristics of strong static echoes between frames and the sparsity characteristics of target range images in the range-time domain on the basis of the imaging model, and establish an imaging objective function with joint low-rank-sparse constraints. Based on the frequency domain echo data of the target in the detection scene, the range image is solved by using a proximal gradient iterative algorithm through an iterative unfolding reconstruction network to obtain the coarse reconstructed range image corresponding to the target. Then, the enhancement network is used for enhancement processing to compensate for the trajectory holes of the target and suppress the sidelobe lift of the target and the residual static clutter of the detection scene. Thus, through iterative unfolding of the reconstruction network and the enhancement network, even under complex conditions such as target amplitude flicker, data loss and strong static background, the continuity of the target main lobe morphology and range-time trajectory can still be maintained, realizing high-quality high-resolution range image reconstruction and enhancement, providing stable and high-fidelity range image input for subsequent imaging and positioning applications.
[0013] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description
[0014] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 A flowchart illustrating an ultra-wideband radar range image enhancement method for scenarios with missing target amplitude scintillation, provided in an embodiment of this application, is shown. Figure 2 This illustration shows a schematic diagram of an example detection scenario provided in an embodiment of this application; Figure 3 This illustration shows an example of an iterative unfolding reconstruction network structure provided in an embodiment of this application; Figure 4 This illustration shows a schematic diagram of an example of an enhanced network structure provided in an embodiment of this application; Figure 5 This invention provides a schematic diagram of an example overall network architecture. Figure 6 This paper presents a comparison image of distance image reconstruction results, as provided in an embodiment of this application. Figure 6 (a) is a simulation scene diagram. Figure 6 (b) is the theoretical distance image. Figure 6 (c) Displaying the range image reconstruction results corresponding to the IDFT transform method for moving targets. Figure 6 (d) shows the distance image reconstruction result corresponding to the singular value decomposition plus IDFT method. Figure 6 (e) shows the distance image reconstruction results corresponding to the robust principal component analysis method. Figure 6 (f) shows the distance image reconstruction result corresponding to the weighted nuclear norm minimization method. Figure 6 (g) represents the distance image reconstruction results corresponding to the low-rank prior and the joint total variation method. Figure 6 (h) represents the distance image reconstruction result corresponding to the method in the embodiments of this application; Figure 7 This paper shows a comparison diagram of the 100th frame imaging result of an example provided in an embodiment of this application; Figure 7 (a) Imaging results for moving target display using the IDFT transform method Figure 7 (b) shows the imaging results corresponding to the singular value decomposition plus IDFT method. Figure 7 (c) shows the imaging results corresponding to the robust principal component analysis method. Figure 7 (d) shows the imaging results corresponding to the weighted nuclear norm minimization method. Figure 7 (e) shows the imaging results corresponding to the low-rank prior and the joint total variation method. Figure 7(f) shows the imaging results corresponding to the method in the embodiments of this application; Figure 8 This paper presents a comparison chart of positioning results for an example provided in an embodiment of this application. Figure 8 (a) Displays the localization result corresponding to the IDFT transform method for moving targets. Figure 8 (b) shows the localization results corresponding to the singular value decomposition plus IDFT method. Figure 8 (c) shows the localization results corresponding to the robust principal component analysis method. Figure 8 (d) shows the localization result corresponding to the weighted nuclear norm minimization method. Figure 8 (e) shows the localization results corresponding to the low-rank prior and the joint total variation method. Figure 8 (f) is the positioning result corresponding to the method in the embodiment of this application; Figure 9 This paper presents a schematic diagram of the structure of an ultra-wideband radar range image enhancement device for scenarios with missing target amplitude scintillation, according to an embodiment of this application. In the figure, X (m) represents the target's horizontal coordinate, and Y (m) represents the target's vertical coordinate. Detailed Implementation
[0017] To better understand the above-mentioned objectives, features, and advantages of this application, the solution of this application will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.
[0018] Theoretically, the time delay introduced by the target will manifest as a phase shift in the echo, and HRRP reconstruction can be achieved by analyzing the spectral characteristics of the received echo. Based on this principle, the most commonly used HRRP reconstruction method applies DFT / IDFT to frequency domain sampling with uniform intervals to obtain HRRP. This type of method has advantages such as simple implementation and high computational efficiency. To achieve "ghost-free" imaging and localization, clutter suppression and IDFT processing can be performed on the echo first, candidate echoes can be detected from the HRRP map, and then the iterative adaptive method (IAA) can be used to jointly estimate the direction of departure (DoD) and direction of arrival (DoA) to identify and remove ghost echoes at the signal level. Finally, back projection is used to reconstruct the ghost-free scene. For human activity recognition, the HRRP map can be obtained by performing IDFT on multi-period echo sequences. Combined with a region vectorized convolutional gated recurrent network, and a Keystone transform can be applied to the radar echo to compensate for range cell migration and improve Doppler resolution, achieving relatively robust recognition performance in wall-penetrating scenarios. However, in complex scenarios, the range images obtained by actual measurement often have the following typical problems: (1) The target amplitude flicker is obvious. Affected by factors such as target attitude change, scattering center blockage and azimuth angle change, the target echo intensity of the same distance unit will fluctuate drastically between consecutive frames, which is manifested as the main lobe peak randomly rising or falling in the time dimension, i.e. the so-called "amplitude flicker" phenomenon, making it difficult to stably represent the target trajectory; (2) Data is missing or incomplete. Due to factors such as the dynamic range limitation of the transmit / receive link, loss of some pulses or subcarrier frequencies, accidental deletion of threshold detection, and manual subsampling, the effective echoes of some time-range cells are not recorded or are suppressed, resulting in holes, breaks, or even complete local loss in the range image, which seriously affects the reliability of subsequent imaging and positioning; (3) Strong static background and wall clutter interference are prominent. In the detection environment through walls or indoors, the strong echoes generated by static walls, furniture and other fixed scattering objects occupy a large amount of energy in the range image, masking the weak target echoes, causing the main lobe shape of the target to be distorted, and the side lobes and artifacts to be significantly raised, further aggravating the adverse effects of target amplitude flicker and missing.
[0019] To address the technical problem that range image reconstruction methods based solely on DFT or IDF struggle to simultaneously maintain main lobe morphology, trajectory continuity, and background suppression in complex scenarios such as target amplitude flicker, data gaps, and strong static backgrounds, thus failing to provide stable, high-fidelity range image input for subsequent imaging and positioning applications, this embodiment provides an ultra-wideband radar range image enhancement method for scenarios with missing target amplitude flicker. Figure 1 As shown, the method includes: Step 101: Detect the target in the detection scene using a stepped-frequency ultra-wideband radar and obtain the frequency domain echo data corresponding to the target.
[0020] In some embodiments, the stepped-frequency ultra-wideband radar is an ultra-wideband radar deployed for specific detection scenarios, employing stepped-frequency signal modulation. It possesses characteristics such as high range resolution and strong wall penetration, enabling multi-moment dynamic detection of targets. Correspondingly, the detection scenarios may include non-line-of-sight detection scenarios with strong static backgrounds, such as wall penetration detection, indoor detection, and life search and rescue in rubble. The detection targets may include objects to be detected, such as point targets behind walls / obstacles, human targets, and concealed objects.
[0021] For example, a stepped-frequency ultra-wideband radar can be used to transmit a stepped-frequency signal of 1.6~2.2GHz to collect frequency domain echo data of the target at multiple moments. The frequency domain echo data can be the time domain echo signal collected by the receiving antenna after the electromagnetic wave transmitted by the stepped-frequency ultra-wideband radar is reflected by the target and scattered by scene clutter. It is then converted into frequency dimension signal data by Fourier transform and used as the core raw data for reconstructing a high-resolution range image, corresponding to the transmission frequency characteristics of the stepped-frequency radar, such as amplitude characteristics and phase characteristics.
[0022] Among them, the amplitude characteristic corresponds to the energy intensity of the radar received echo, reflecting the scattering ability of the target and static background on electromagnetic waves of different frequencies. The stronger the target scattering ability / the more the electromagnetic wave frequency matches the target resonant frequency, the higher the amplitude value of the corresponding frequency point. The phase characteristic corresponds to the phase shift of the electromagnetic wave after reflection by the target / background, which is directly related to the spatial distance between the target and the radar (the phase shift is positively correlated with the propagation distance). It is the core basis for ultra-wideband radar to realize range calculation through IDFT / DFT.
[0023] Step 102: Based on the iterative unfolding reconstruction network, perform near-end iterative solution of the imaging objective function corresponding to the frequency domain echo data to obtain the coarse reconstructed range image corresponding to the detected target.
[0024] The imaging objective function is a function established under the constraints of the imaging model, combining the low-rank characteristics of strong static echoes between frames and the sparsity characteristics of the target range image in the range-time domain.
[0025] Specifically, the iterative unfolding reconstruction network can include multiple network layers, each of which may contain components such as momentum acceleration, strong clutter removal, and range image reconstruction. Through end-to-end training, the network can directly output a coarsely reconstructed range image based on frequency domain echo data, while maintaining good physical interpretability.
[0026] In specific application scenarios, due to complex detection environments, inherent characteristics of radar hardware, and dynamic changes in targets, missing frequency dimension sampling leads to incomplete target frequency domain features. Strong static clutter superimposed on the target signal further amplifies inter-frame amplitude fluctuations, exacerbating target amplitude flicker. To address this, an imaging model can be constructed by fusing the frequency dimension sampling matrix, DFT operator, and low-rank background term. Considering the significant low-rank characteristics of strong static objects (such as walls) and fixed scatterers between frames, and the sparse distribution of targets in the range-time plane, low-rank and sparsity constraints are introduced into the imaging model to construct a joint low-rank-sparse imaging objective function. This function is used to extract the target range image from distorted and incomplete frequency domain echoes. Specifically, an iterative expansion reconstruction network is used for multiple iterations. The preliminary high-resolution range image obtained through near-end iteration is used as the coarse reconstructed range image corresponding to the detected target. This achieves effective separation of the detected target from the strong static background, preliminary compensation for missing frequency dimension sampling, and suppression of most noise interference.
[0027] Step 103: The coarsely reconstructed range image is enhanced by an augmentation network to generate an enhanced range image corresponding to the detected target.
[0028] Among them, the enhancement network is used to compensate for the trajectory hole of the target under the conditions of target amplitude flicker and sampling loss, and to suppress the sidelobe rise of the target and the residual static clutter of the detection scene.
[0029] In some embodiments, the coarse reconstructed range image output by the iterative unfolding reconstruction network can be input into the enhancement network to perform nonlinear compensation for trajectory holes, side lobes, and residual clutter caused by amplitude flicker and data loss, while maintaining the main lobe structure. The two-stage deep network, composed of the iterative unfolding reconstruction network and the enhancement network, achieves nonlinear compensation enhancement of the coarse reconstructed range image. The obtained enhanced range image can serve as a high-fidelity input for subsequent through-wall imaging, target localization, and behavior analysis modules, enabling stable perception and applications in scenarios with missing target amplitude flicker.
[0030] In this way, this embodiment can introduce the low-rank characteristics of strong static echoes between frames and the sparsity characteristics of target range images in the range-time domain on the basis of the imaging model, and establish an imaging objective function with joint low-rank-sparse constraints. Based on the frequency domain echo data of the target in the detection scene, the range image is solved by using a proximal gradient iterative algorithm through an iterative unfolding reconstruction network to obtain the coarse reconstructed range image corresponding to the target. Then, the enhancement network is used to enhance the target to compensate for the trajectory holes of the target and suppress the sidelobe lift of the target and the residual static clutter of the detection scene. Thus, through iterative unfolding of the reconstruction network and the enhancement network, even under complex conditions such as target amplitude flicker, data loss and strong static background, the continuity of the target main lobe morphology and range-time trajectory can still be maintained, achieving high-quality high-resolution range image reconstruction and enhancement, providing stable and high-fidelity range image input for subsequent imaging and positioning applications.
[0031] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the implementation of this embodiment, step 101 may optionally include: detecting the target in the detection scene using a stepped-frequency ultra-wideband radar, and acquiring multi-frame frequency domain echo data corresponding to the target at multiple acquisition times; preprocessing the multi-frame frequency domain echo data to obtain frequency domain echo data.
[0032] In some embodiments, a MIMO ultra-wideband radar imaging system can be used to detect unknown detection scenarios. Radar detection data corresponding to a stepped-frequency ultra-wideband radar can be set, such as frequency data, acquisition data (e.g., frame rate), antenna array, signal characteristics, etc., and frequency domain echo data corresponding to multiple time points can be obtained based on the set radar detection data. As one possible implementation, simulation can be performed using Matlab, and a stepped-frequency radar can be employed. =1.6 GHz-2.2 GHz, with 301 frequency points, center frequency =1.9 GHz, frame rate 50 Hz, sampling time 4 s ( =200 frames); the antenna center is located at (5 m, 0 m). If the array center is taken as the origin, the transmitting antennas are located at (-0.15 m, 0 m) and (0.15 m, 0 m) respectively, and the receiving antennas are located at (-0.1125 m, 0 m), (-0.0375 m, 0 m), (0.0375 m, 0 m) and (0.1125 m, 0 m) respectively.
[0033] For example, such as Figure 2The diagram illustrates a detection scenario where the radar reaches the wall at a distance of 0.5 m, and the wall thickness is 0.2 m. Several point targets are randomly placed behind the wall, with their radar cross-sections uniformly distributed within a given range to simulate scatterers of varying intensities. The sample echoes are superimposed with complex Gaussian white noise with a signal-to-noise ratio (SNR) of 20 dB, and amplitude gaps and amplitude flicker are randomly introduced into the time-range plane. Correspondingly, a detection coordinate system can be constructed by arranging the walls corresponding to the radar array, where x represents the horizontal axis corresponding to the wall, and y represents the vertical axis perpendicular to the wall.
[0034] Optionally, preprocessing the multi-frame frequency domain echo data to obtain the frequency domain echo data may specifically include: performing DC component removal processing on the multi-frame frequency domain echo data to obtain frequency domain echo data after removing DC bias; performing amplitude and phase consistency calibration on the multi-channel data corresponding to the multi-frame frequency domain echo data to obtain frequency domain echo data after eliminating channel errors; and performing amplitude normalization processing on the multi-frame frequency domain echo data to obtain frequency domain echo data after eliminating inter-frame amplitude flicker.
[0035] For example, echo acquisition and preprocessing can be performed, and preprocessing may include DC removal and channel calibration; specifically, preprocessing can be performed according to the following steps: S11. Perform DC component removal processing on each frame of echo data; S12. Perform amplitude and phase consistency calibration on multi-channel data; S13. Normalize the echo data to reduce the impact of target amplitude flicker on subsequent reconstruction and enhancement.
[0036] Optionally, before step 102, the method of this embodiment may further include: fusing the frequency dimension sampling matrix, DFT transform operator and low-rank background term in the detection scenario, and constructing an imaging model between the zero-padding frequency domain echo data and the target range image for the frequency domain echo data with missing frequency dimensions.
[0037] In some embodiments, the transmitted signal of the stepped-frequency ultra-wideband radar is a stepped-frequency signal. Therefore, for a stepped-frequency single-channel single-frame echo, the original observed frequency domain echo data can be represented as a... The frequency domain vector. Before performing IDFT, zero-padding can be performed before and after the step frequency signal, respectively, according to physical meaning. After zero-padding, the length of the frequency domain vector is extended to The extended frequency domain vector can be denoted as .right By performing IDFT operations, the range profile corresponding to the detected target can be obtained. .
[0038] Furthermore, the imaging model, considering missing samples and strong static echo characteristics, can be specifically constructed as follows: ; In the formula, F (·) can represent the DFT transform operator, used to avoid explicitly constructing and multiplying large-scale perceptron matrices (dictionary matrices). This is the frequency domain vector (echo measurement matrix) corresponding to the multi-frame echo data after zero-padding. It can represent the sampling matrix (frequency dimension) corresponding to frequency domain echo data; It can represent the single-channel multi-frame target range profile (range profile matrix) corresponding to frequency domain echo data; It can represent the low-rank background term corresponding to the detection scene.
[0039] In some embodiments, based on the imaging model, considering the significant low-rank characteristics of strongly static objects (such as walls) and fixed scatterers between frames, and the sparse distribution of the detected target in the range-time plane, low-rank constraints and sparsity constraints are introduced to construct an imaging objective function that combines low-rank and sparsity characteristics, which can be expressed as: ; In the formula, This represents a sparsity constraint on the target range profile, which can be used to enhance the sparse distribution characteristics of the target scattered energy in the range-time plane; This represents the nuclear norm constraint on the low-rank background term, which can be used to enhance the low-rank characteristics between static clutter frames and achieve separation of the target from the clutter. , These are the regularization parameters for balancing sparse prior constraints and low-rank prior constraints, respectively. It can be used to balance the weights of target sparsity constraints and control the sparsity intensity of target range images. It can be used to balance the weights of low-rank background constraints and control the low-rank intensity of static background clutter.
[0040] Correspondingly, range image inversion and reconstruction can be performed based on the imaging objective function, and the first image corresponding to the imaging objective function... The low-rank background term and target distance image obtained in the next iteration can be represented as follows: ; Furthermore, to address the above problem, a proximal gradient iterative algorithm can be used for solution. For example, regarding... The proximal gradient iteration is introduced for solving the problem. The low-rank background term obtained in the next iteration can be expressed as: ; Among them, the proximal gradient descent step corresponding to the low-rank background term , The step size is used for proximal gradient updates. It can be obtained by the singular value thresholding operator: For any input matrix Singular value threshold operator ,in , This is a soft threshold operator; Similarly, the proximal gradient iteration is introduced to solve the problem. The range image of the next stacked target can be represented as: ; Among them, the near-end gradient descent step corresponding to the target distance image. ,but It can be calculated using the soft threshold operator: .
[0041] The preliminary range image obtained by near-end iterative solution can achieve preliminary separation of the detected target from the background and preliminary compensation for missing frequency dimension sampling.
[0042] Optionally, during the proximal gradient iteration solution of the imaging objective function, an extrapolation update can be performed on the iterative variables corresponding to the imaging objective function based on a momentum acceleration strategy to obtain single-channel multi-frame range images and low-rank background terms. The imaging objective function can be used to perform residual calculation, gradient step size update, and soft thresholding and singular value thresholding operators through the proximal gradient iteration algorithm to alternately update single-channel multi-frame range images and low-rank background terms.
[0043] For example, to accelerate convergence, the Nesterov momentum parameter and extrapolation step can be introduced, with the update formula as follows: (a) Momentum: ; in, For momentum acceleration parameters, For the reason The calculated external impulsive force coefficient.
[0044] (b) Extrapolation: ; in, , To , The extrapolation acceleration point was created.
[0045] Accordingly, the above can be... , Replace with and This achieves accelerated convergence.
[0046] This solution method, which combines Nesterov momentum-accelerated proximal gradient iterative algorithm, integrates the variable change trend of the previous iteration, predicts the current iteration direction, and gradually approximates the true target distance image, avoiding the slow convergence and oscillation problems of traditional proximal gradient.
[0047] Optionally, step 102 may specifically include: unfolding the proximal iterative solution process corresponding to the imaging objective function layer by layer to construct an iterative unfolding reconstruction network. The iterative unfolding reconstruction network is composed of multiple iterative modules connected in series. The iterative modules include at least an observation residual calculation unit, a single-channel multi-frame range image gradient update unit, a low-rank background term gradient update unit, a sparse proximal unit, and a low-rank proximal unit. Using the iterative unfolding reconstruction network, based on the single-channel multi-frame range image and the low-rank background term obtained from the proximal iterative solution of the imaging objective function, a coarse reconstructed range image is obtained.
[0048] Optionally, the iterative unfolding reconstruction network can set the step size and threshold parameters as learnable parameters, while maintaining consistency between the sampling matrix and the transformation operator and the imaging model.
[0049] Specifically, in the construction process of the iterative unfolding reconstruction network, the multiple iterations of the imaging objective function can be mapped one-to-one to multiple network layers. Each layer may include momentum acceleration, strong clutter removal, range image reconstruction, etc. Figure 3 As shown, the input parameter Y, through K layers of iteration, outputs a coarse reconstructed range image X. Taking the k-th layer as an example, momentum acceleration, strong clutter removal, and range image reconstruction can be used to obtain the single-channel multi-frame target range image and low-rank background term output by the k-th layer. This is then iteratively updated through the (k+1)-th layer until the K-th layer update is complete. The step size and threshold parameters are set as learnable parameters for the iterative unfolding reconstruction network, and the sampling matrix and transformation operator are kept consistent with the imaging model. Through end-to-end joint training, the iterative unfolding reconstruction network is obtained, enabling it to directly output a coarse reconstructed range image based on the observed data. At the same time, it retains good physical interpretability.
[0050] Optionally, step 103 may specifically include: inputting the complex information corresponding to the coarse reconstructed range image into different channels of the enhancement network according to the real and imaginary parts respectively; using the enhancement network to perform amplitude normalization and dynamic range recalibration processing on the coarse reconstructed range image to generate an enhanced range image corresponding to the detected target; the enhancement network fuses features of different scales through multi-scale skip connections to compensate and enhance the trajectory holes, side lobes and residual clutter of the detected target while maintaining the main lobe structure of the detected target.
[0051] The enhancement network can be a convolutional neural network with an encoder-decoder structure, such as the U-Net network, to coarsely reconstruct the distance image of the interval of interest. As input to the augmentation network, the real and imaginary parts of the multi-frame range-time graphs are organized into a two-channel input, where the region of interest is a preset range threshold. Only slow-time distance data within the specified distance threshold is used as input to the augmentation network. A U-Net augmentation network with an encoder-decoder structure can be introduced for augmentation processing. A schematic diagram of the U-Net augmentation network structure is shown below. Figure 4 As shown: (1) The encoder part of the augmented network consists of multiple convolutions and downsampling, extracting trajectory features and local texture features across frames layer by layer. The input of the augmented network is [Re(X); Im(X)]. Re(X) can be a matrix composed of the real part of the complex numerical target range image X, and Im(X) can be a matrix composed of the coefficients of the imaginary part of the complex numerical target range image X. The real and imaginary parts can be concatenated into a dual-channel input format by row / column. By separating the input through dual channels, the complex numerical physical features of the range image are completely preserved, avoiding the loss of phase information in the traditional single-channel amplitude input, and ensuring the physical interpretability of the augmented range image and the target positioning accuracy. Specifically, the network may include convolution operation modules such as conv3×3 + BatchNormalization (BN) + ReLU, MAX Pool 2×2, Up-Conv 2×2, Concatenation, and Conv1×1. (2) The decoder part of the augmented network gradually restores the resolution through upsampling and convolution, and makes skip connections with the corresponding layers of the encoder to fuse information at different scales and obtain the dual-channel output result corresponding to the augmented distance image [Re( ); Im( This allows the near-end gradient optimization process to be unfolded layer by layer into a deep network structure with learnable parameters, and combined with U-Net to form a two-stage range image reconstruction and enhancement network. Through the enhancement network, while keeping the main lobe structure intact, nonlinear compensation is performed on the trajectory holes, side lobes and residual clutter caused by amplitude flicker and data loss, thereby further suppressing residual clutter and compensating for amplitude flicker and loss.
[0052] Optionally, the iterative unfolding reconstruction network and the enhancement network adopt an end-to-end joint training method. The loss function corresponding to the training process of the end-to-end joint training method includes a first-stage loss and a second-stage loss. The first-stage loss is used to constrain the coarse reconstruction error of the iterative unfolding reconstruction network based on the energy ratio of the detected target energy to the background energy. The energy ratio is used to enhance the detected target and suppress residual static clutter in the detected scene. The second-stage loss is used to constrain the fitting error between the enhanced range image and the high-resolution range image of the label.
[0053] In some embodiments, a two-stage deep network can be constructed based on iterative expansion of the reconstruction network and the enhancement network. First, based on proximal gradient iteration, the optimization process of the imaging objective function can be expanded layer by layer and combined with a U-Net structure to construct a two-stage deep network, such as... Figure 5 The diagram illustrates the overall architecture of a two-stage deep network. It may include an iterative unfolding reconstruction network (Stage 1) for range image reconstruction (input Y, output X), and an augmentation network (Stage 2) for range image enhancement (input X, output X). The loss function is obtained through end-to-end joint training of the two-stage deep networks. oss1, oss2 is used to optimize the learnable network weights corresponding to the reconstruction network and the enhancement network in two iterations, thereby realizing the reconstruction and enhancement of the distance image.
[0054] For example, a two-stage deep network can be jointly trained and tested, using a high-quality, amplitude-stable range image as the labeled range image, and frequency domain echo data (observation data) that has undergone missing sampling, strong static background superposition, and amplitude flicker perturbation as input. Through backpropagation, the parameters such as stride and threshold in the iteratively expanded sub-network and the convolution kernel parameters in the U-Net augmented network are updated simultaneously.
[0055] Further optional, the estimation error of the first stage (preliminary reconstruction stage) and the target-to-background ratio (TBR) can be minimized.
[0056] For example, for a given training set Among them, the losses in the first phase It can be defined as follows: ; in, For inclusion The training set of the nth sample, the nth Each sample contains frequency domain sampled echoes from the network input. Phase 1 monitoring label Second phase of supervision labeling . , which is a weighting coefficient used to balance the relative importance of the MSE term and the TBR constraint term.
[0057] By introducing TBR in the first stage to construct the loss function corresponding to the first stage loss, the detection target is enhanced.
[0058] Specifically, TBR can be used to describe the energy discrimination between the target region and the background region corresponding to the detected target, and to evaluate the quality of the range image. For example, TBR can be defined as: ; in, For the image in pixels Complex values at the location, target region set Background area collection , , They are respectively , Number of pixels in the middle. The goal is to maximize the output of the first-stage coarse reconstruction range image. TBR is used to enhance the target while relatively suppressing the background.
[0059] Optionally, the estimation error corresponding to the second stage (range image enhancement stage) can be minimized, and the second stage loss can be minimized. It can be defined as follows: ; Therefore, the global loss function corresponding to a two-stage deep network It can be defined as follows: ; In the formula, These are the loss weights corresponding to the first-stage loss and the second-stage loss, respectively, to balance the loss items of different stages to the same scale.
[0060] During the testing phase, the network parameters of the pre-trained two-stage deep network were fixed, and frequency domain echo data collected in unknown scenarios were input into the network. First, the iterative expansion reconstruction network outputs a coarse reconstructed range image, which is then processed by the U-Net enhancement network for detail compensation and clutter suppression, ultimately yielding an enhanced high-resolution range image. This enhanced range image can serve as a high-fidelity input for subsequent through-wall imaging, target localization, and behavior analysis modules, enabling stable perception and applications in scenarios where target amplitude flicker is missing.
[0061] For example, the moving target display plus IDFT method, the singular value decomposition plus IDFT method, the robust principal component analysis method, the weighted nuclear norm minimization method, the low-rank prior and joint total variation method, and the method of this embodiment can be used respectively for processing, and the resulting distance image reconstruction comparison results are as follows: Figure 6 As shown, where, Figure 6(a) shows a simulation scene with a wall thickness of 0.2m. It illustrates a coordinate system with X (m) as the horizontal axis and Y (m) as the vertical axis, where X (m) represents the target's horizontal coordinate and Y (m) represents the target's vertical coordinate. The simulation trajectories corresponding to Target1, Target2, Target3, and Target4 are also shown. Figure 6 (b) shows the theoretical distance image corresponding to the simulated scene graph. Figure 6 (c)–(g) show the distance image reconstruction results corresponding to each contrast method. Figure 6 (h) represents the enhanced range image corresponding to the method of this embodiment, with frame as the horizontal axis and range as the vertical axis. It can be seen that, under the same noise level and missing sampling conditions, the range image produced by the comparison method still suffers from problems such as main lobe broadening, trajectory breakage, or strong background clutter. In contrast, the method of this embodiment can effectively suppress strong static background and noise interference while significantly mitigating the adverse effects of target amplitude flicker and missing sampling, reconstructing a high-resolution range image with a clear main lobe morphology, continuous trajectory, and good background suppression.
[0062] Correspondingly, Figure 7 A comparison diagram of the imaging results of the various comparison methods and the method of this embodiment at the 100th frame is shown, wherein... Figure 7 (a)–(e) are schematic diagrams of the imaging corresponding to each comparison method. Figure 7 (f) is an imaging schematic diagram of the method in this embodiment; Figure 8 A comparison chart of the localization results of various comparison methods and the method of this embodiment is shown, wherein... Figure 8 (a)–(e) show the localization results of each comparison method. Figure 8 (f) shows the positioning result of the method in this embodiment. It can be seen that the method in this embodiment can ensure the accuracy and reliability of subsequent imaging and positioning results.
[0063] Compared with related technologies, this embodiment proposes an ultra-wideband radar range image enhancement technique for scenarios with missing target amplitude flicker. It can achieve high-quality, high-resolution range image reconstruction and enhancement under complex conditions such as missing sampling, strong static backgrounds, and target amplitude flicker and local missing data. Specifically, considering the effect of the sampling matrix, an ultra-wideband radar imaging model is constructed based on the relationship between the echo measurement matrix, sampling matrix, dictionary matrix, and range image matrix. Then, the low-rank characteristics of strong static echoes between frames and the sparsity characteristics of the target range image in the range-time domain are introduced to establish an imaging objective function with joint low-rank and sparse constraints. A proximal gradient iteration algorithm is used to achieve the initial solution for range image reconstruction. Finally, the proximal gradient optimization process of the imaging objective function is unfolded layer by layer into a deep network structure with learnable parameters, and combined with U-Net to form a two-stage deep network to obtain the enhanced range image corresponding to the detected target. Under complex conditions such as target amplitude flicker, missing data, and strong static backgrounds, the continuity of the main lobe morphology and range-time trajectory of the detected target is maintained, providing stable and high-fidelity range image input for subsequent imaging and positioning applications. This is achieved through "low-rank and sparse physical model reconstruction +..." The two-stage joint design of "deep network enhancement" takes into account both physical interpretability and data-driven adaptability. It has good robustness to changes in sampling methods, noise levels and scene. It can be widely used in ultra-wideband radar scenarios such as through-wall detection, monitoring of vital signs behind walls and recognition of the behavior of concealed targets, providing effective technical support for high-precision perception and intelligent analysis in complex environments.
[0064] Furthermore, embodiments of this application provide an ultra-wideband radar range image enhancement device for scenarios with missing target amplitude scintillation, such as... Figure 9 As shown, the device includes: an acquisition module 31, a solution module 32, and an enhancement module 33.
[0065] The acquisition module 31 is configured to detect the target in the detection scene using a stepped-frequency ultra-wideband radar and acquire the frequency domain echo data corresponding to the target. The solver module 32 is configured to perform near-end iterative solution of the imaging objective function corresponding to the frequency domain echo data based on the iterative expansion reconstruction network, and obtain the coarse reconstructed range image corresponding to the detected target. The imaging objective function is a function established under the constraints of the imaging model, combining the low-rank characteristics of strong static echo between frames and the sparsity characteristics of the target range image in the range-time domain. The enhancement module 33 is configured to compensate and enhance the coarse reconstructed range image through the enhancement network to generate an enhanced range image corresponding to the detected target. The enhancement network is used to compensate for the trajectory hole of the detected target under the conditions of amplitude flicker and sampling loss of the detected target, and to suppress the sidelobe lift of the detected target and the residual static clutter of the detected scene.
[0066] In some embodiments, the solution module 32 is specifically configured to expand the proximal iterative solution process corresponding to the imaging objective function layer by layer to construct an iterative expansion reconstruction network. The iterative expansion reconstruction network is composed of multiple iterative modules connected in series. The iterative modules include at least an observation residual calculation unit, a single-channel multi-frame range image gradient update unit, a low-rank background term gradient update unit, a sparse proximal unit, and a low-rank proximal unit. Using the iterative expansion reconstruction network, a coarse reconstructed range image is obtained based on the single-channel multi-frame range image and the low-rank background term obtained from the proximal iterative solution of the imaging objective function. The method further includes: in the proximal gradient iterative solution process of the imaging objective function, performing extrapolation updates on the iterative variables corresponding to the imaging objective function based on a momentum acceleration strategy to obtain the single-channel multi-frame range image and the low-rank background term. The imaging objective function is used to perform residual calculation, gradient step size update, and soft thresholding operator and singular value thresholding operator through the proximal gradient iterative algorithm to alternately update the single-channel multi-frame range image and the low-rank background term.
[0067] In some embodiments, the solving module 32 is further configured to fuse the frequency dimension sampling matrix, DFT transform operator and low-rank background term in the detection scenario, and to construct an imaging model between the zero-padding frequency domain echo data and the target range image for frequency domain echo data with missing frequency dimensions.
[0068] In some embodiments, the enhancement module 33 is specifically configured to input the complex information corresponding to the coarse reconstructed range image into different channels of the enhancement network according to the real and imaginary parts respectively; use the enhancement network to perform amplitude normalization and dynamic range recalibration processing on the coarse reconstructed range image to generate an enhanced range image corresponding to the detection target; the enhancement network fuses features of different scales through multi-scale skip connections, which is used to compensate and enhance the trajectory holes, side lobes and residual clutter of the detection target while maintaining the main lobe structure of the detection target.
[0069] In some embodiments, the enhancement module 33 is specifically configured to employ an end-to-end joint training method for the iterative unfolding reconstruction network and the enhancement network. The loss function corresponding to the training process of the end-to-end joint training method includes a first-stage loss and a second-stage loss. The first-stage loss is used to constrain the coarse reconstruction error of the iterative unfolding reconstruction network and introduces an energy ratio term between the detected target energy and the background energy. The energy ratio term is used to enhance the detected target and suppress residual static clutter in the detected scene. The second-stage loss is used to constrain the fitting error between the enhanced range profile and the label range profile.
[0070] In some embodiments, the acquisition module 31 is specifically configured to detect the target in the detection scene using a stepped-frequency ultra-wideband radar, acquire multi-frame frequency domain echo data corresponding to the target at multiple acquisition times, and preprocess the multi-frame frequency domain echo data to acquire frequency domain echo data.
[0071] In some embodiments, the acquisition module 31 is specifically configured to perform DC component removal processing on multi-frame frequency domain echo data to obtain frequency domain echo data after removing DC bias; perform amplitude and phase consistency calibration on multi-channel data corresponding to multi-frame frequency domain echo data to obtain frequency domain echo data after eliminating channel errors; and perform amplitude normalization processing on multi-frame frequency domain echo data to obtain frequency domain echo data after eliminating inter-frame amplitude flicker.
[0072] It should be noted that other corresponding descriptions of the functional units involved in the ultra-wideband radar range image enhancement device for target amplitude scintillation missing scenarios provided in this application embodiment can be found in the following references. Figure 1 The corresponding descriptions in [the document] will not be repeated here.
[0073] Based on the above, Figure 1 As illustrated in the example, correspondingly, embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described... Figure 1 The example method shown.
[0074] Based on the above, Figure 1 As illustrated, correspondingly, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the above-described... Figure 1 The example method shown.
[0075] Based on this understanding, the technical solutions of the embodiments of this application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.
[0076] Based on the above, Figure 1 The method shown, and Figure 9 To achieve the above objectives, the present application also provides an electronic device, comprising a storage medium and a processor; the storage medium for storing a computer program; and the processor for executing the computer program to implement the above-described virtual device embodiments. Figure 1 The method shown.
[0077] Optionally, the aforementioned electronic device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, an input unit, etc.
[0078] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.
[0079] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.
[0080] Through the above description of the implementation methods, those skilled in the art can clearly understand that this application can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented in hardware. This application can construct an ultra-wideband radar imaging model based on the relationship between the echo measurement matrix, sampling matrix, dictionary matrix, and range image matrix, considering the effect of the sampling matrix; then, by introducing the low-rank characteristics of strong static echoes between frames and the sparsity characteristics of the target range image in the range-time domain, an imaging objective function with joint low-rank-sparse constraints is established, and a preliminary solution for range image reconstruction is achieved using a proximal gradient iteration algorithm; finally, the proximal gradient optimization process of the imaging objective function is expanded layer by layer into a deep network structure with learnable parameters, and combined with U-Net to form a two-stage deep network, obtaining the enhanced range image corresponding to the detected target. Under complex conditions such as target amplitude flicker, data loss, and strong static background, the continuity of the main lobe morphology and range-time trajectory of the detected target is maintained, providing stable and high-fidelity range image input for subsequent imaging and positioning applications. This is achieved through "low-rank-sparse physical model reconstruction +..." The two-stage joint design of "deep network enhancement" takes into account both physical interpretability and data-driven adaptability. It has good robustness to changes in sampling methods, noise levels and scene. It can be widely used in ultra-wideband radar scenarios such as through-wall detection, monitoring of vital signs behind walls and recognition of the behavior of concealed targets, providing effective technical support for high-precision perception and intelligent analysis in complex environments.
[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.
[0082] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A method for enhancing the range image of an ultra-wideband radar system in scenarios with missing target amplitude scintillation, characterized in that, include: The step-frequency ultra-wideband radar is used to detect targets in the detection scene and acquire frequency domain echo data corresponding to the targets. Based on the iterative unfolding reconstruction network, the imaging objective function corresponding to the frequency domain echo data is solved by near-end iterative solution to obtain the coarse reconstructed range image corresponding to the detected target; the imaging objective function includes a function established under the constraints of the imaging model, which combines the low-rank characteristics of strong static echo between frames and the sparsity characteristics of the target range image in the range-time domain. The coarse reconstructed range image is enhanced by an enhancement network to generate an enhanced range image corresponding to the detected target. The enhancement network is used to compensate for the trajectory hole of the detected target under the conditions of amplitude flicker and sampling loss, and to suppress the sidelobe lift of the detected target and the residual static clutter of the detected scene.
2. The method according to claim 1, characterized in that, The iterative expansion reconstruction network performs near-end iterative solution of the imaging objective function corresponding to the frequency domain echo data to obtain a coarse reconstructed range image corresponding to the detected target, including: The near-end iterative solution process corresponding to the imaging objective function is expanded layer by layer to construct an iterative expansion reconstruction network. The iterative expansion reconstruction network is composed of multiple iterative modules connected in series. The iterative module includes at least an observation residual calculation unit, a single-channel multi-frame range image gradient update unit, a low-rank background term gradient update unit, a sparse near-end unit, and a low-rank near-end unit. The coarse reconstructed range image is obtained by using the iterative unfolding reconstruction network based on the single-channel multi-frame range image and low-rank background term obtained by the near-end iterative solution of the imaging objective function; The method further includes: During the proximal gradient iteration solution of the imaging objective function, the iterative variables corresponding to the imaging objective function are extrapolated and updated based on the momentum acceleration strategy to obtain the single-channel multi-frame range image and low-rank background term. The imaging objective function is used to perform residual calculation, gradient step size update, and soft thresholding operator and singular value thresholding operator through the proximal gradient iteration algorithm to alternately update the single-channel multi-frame range image and low-rank background term.
3. The method according to claim 2, characterized in that, Before obtaining the coarse reconstructed range image corresponding to the detected target by performing near-end iterative solution of the imaging objective function corresponding to the frequency domain echo data based on the iterative unfolding reconstruction network, the method further includes: By integrating the frequency dimension sampling matrix, DFT transform operator, and low-rank background term in the detection scenario, an imaging model is constructed between the zero-padding frequency domain echo data and the target range image for frequency domain echo data with missing frequency dimensions.
4. The method according to claim 1, characterized in that, The step of compensating and enhancing the coarsely reconstructed range image through an enhancement network to generate an enhanced range image corresponding to the detected target includes: The complex information corresponding to the coarsely reconstructed distance image is input into different channels of the augmentation network according to the real and imaginary parts, respectively. The enhancement network is used to perform amplitude normalization and dynamic range recalibration on the coarse reconstructed range image to generate an enhanced range image corresponding to the detected target. The enhancement network fuses features of different scales through multi-scale skip connections to compensate for and enhance the trajectory holes, side lobes and residual clutter of the detected target while maintaining the main lobe structure of the detected target.
5. The method according to claim 1, characterized in that, The iterative unfolding reconstruction network and the enhancement network are trained in an end-to-end joint training manner. The loss function corresponding to the training process of the end-to-end joint training manner includes a first-stage loss and a second-stage loss. The first-stage loss is used to constrain the coarse reconstruction error of the iterative unfolding reconstruction network based on the energy ratio of the detected target energy to the background energy. The energy ratio is used to enhance the detected target and suppress residual static clutter in the detected scene. The second-stage loss is used to constrain the fitting error between the enhanced range image and the label range image.
6. The method according to claim 1, characterized in that, The step-frequency ultra-wideband radar is used to detect targets in the detection scene and acquire frequency domain echo data corresponding to the targets, including: The detection target in the detection scene is detected by step-frequency ultra-wideband radar, and multi-frame frequency domain echo data of the detection target at multiple acquisition times are obtained. The multi-frame frequency domain echo data is preprocessed to obtain the frequency domain echo data.
7. The method according to claim 6, characterized in that, The step of preprocessing the multi-frame frequency domain echo data to obtain the frequency domain echo data includes: The multi-frame frequency domain echo data is processed to remove the DC component, and the frequency domain echo data after removing the DC bias is obtained. Amplitude and phase consistency calibration is performed on the multi-channel data corresponding to the multi-frame frequency domain echo data to obtain frequency domain echo data after eliminating channel errors; The multi-frame frequency domain echo data is subjected to amplitude normalization processing to obtain frequency domain echo data after eliminating inter-frame amplitude flicker.
8. An ultra-wideband radar range image enhancement device for scenarios with missing target amplitude scintillation, characterized in that, include: The acquisition module is configured to detect targets in the detection scene using a stepped-frequency ultra-wideband radar and acquire frequency domain echo data corresponding to the targets. The solution module is configured to perform near-end iterative solution of the imaging objective function corresponding to the frequency domain echo data based on the iterative unfolding reconstruction network, and obtain the coarse reconstructed range image corresponding to the detected target; the imaging objective function includes a function established under the constraints of the imaging model, which combines the low-rank characteristics of strong static echo between frames and the sparsity characteristics of the target range image in the range-time domain; An enhancement module is configured to compensate and enhance the coarse reconstructed range image through an enhancement network to generate an enhanced range image corresponding to the detected target. The enhancement network is used to compensate for trajectory holes of the detected target under the conditions of amplitude flicker and sampling loss of the detected target, and to suppress sidelobe lift of the detected target and residual static clutter of the detected scene.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.
10. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.