Terahertz sar high-efficiency self-focusing imaging method and system based on RB-PCA

Abstract

The application provides a terahertz SAR high-efficiency self-focusing imaging method and system based on RB-PCA, comprising: obtaining original echo signal data and performing pretreatment; performing distance block in the distance direction on the pretreated echo signal data to obtain a plurality of distance blocks, performing sub-aperture segmentation in the distance block for each distance block to obtain a plurality of sub-aperture signal data; performing desquamation processing on each sub-aperture signal data to obtain desquamation signal data; performing PCA processing on the desquamation signal data and selecting a characteristic vector with the largest characteristic value as a target characteristic vector; obtaining each sub-aperture phase error gradient estimation value through adaptive PGA estimation on the target characteristic vector; obtaining two-dimensional space-varying phase errors of each distance block through global splicing and fusion on each sub-aperture phase error gradient estimation value; and performing phase compensation and azimuth compression on the pretreated echo signal data based on the two-dimensional space-varying phase errors to obtain focused imaging results.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and in particular to a high-efficiency autofocusing imaging method and system for terahertz SAR based on RB-PCA. Background Technology

[0002] Synthetic Aperture Radar (SAR) features all-weather, all-time operation and high resolution. In recent years, with the rapid development of semiconductor technology and radio frequency technology, the operating frequency band of SAR has gradually extended into the terahertz band. Compared to microwaves, terahertz waves have shorter wavelengths (sub-millimeter level), which endows terahertz SAR with extremely high spatial resolution and a keen ability to detect minute Doppler shifts. Furthermore, terahertz radar systems typically have small antenna apertures and are lightweight, making them easily integrated into UAVs and vehicle platforms, thus demonstrating broad application prospects in emerging fields such as short-range remote sensing, autonomous driving environmental perception, and security imaging.

[0003] To achieve centimeter-level or even sub-millimeter-level range resolution, radar systems typically require extremely large signal bandwidths (reaching several GHz or even tens of GHz). However, traditional frequency-modulated continuous wave (FMCW) systems impose extremely high hardware requirements on the sampling rate of analog-to-digital converters and the instantaneous processing bandwidth of the system when generating ultra-wide bandwidth signals, leading to a sharp increase in system cost and complexity. Stepped frequency continuous wave (SFCW) technology effectively circumvents the stringent hardware limitations imposed by transmitting a series of narrow-band sub-pulses with linearly increasing frequencies, coherently synthesizing a large-bandwidth signal in the frequency domain.

[0004] Despite the significant advantages of stepped-frequency terahertz SAR in high-resolution imaging, its practical application still faces severe technical challenges. The vehicle-mounted or airborne platforms on which the radar is mounted are subject to significant environmental disturbances, causing the radar track to deviate from its ideal straight-line motion. Existing inertial navigation systems (INS) and global positioning systems (GNSS) typically cannot provide the sub-millimeter-level motion measurement accuracy required for terahertz imaging. Furthermore, the extremely short wavelength of the terahertz band (e.g., 300 GHz corresponds to 1 mm wavelength) makes the system highly sensitive to platform motion errors. Millimeter-level vibrations that are negligible in the microwave band can introduce drastic phase errors exceeding one wavelength in the terahertz band, leading to severe image defocusing. Therefore, high-precision motion compensation (MOCO) for terahertz SAR must be performed using echo-based self-focusing algorithms. The two-dimensional space-variance (2D-SV) motion error caused by the large synthetic bandwidth required for high-resolution imaging presents an even greater challenge to MOCO.

[0005] Current motion compensation strategies can be broadly categorized into two types: compensation methods based on high-precision motion sensors and autofocus methods based on echo data. Motion sensor-based compensation methods primarily rely on inertial navigation systems (INS), inertial measurement units (IMUs), or global positioning systems (GNSS) mounted on the platform to record the platform's motion trajectory parameters in real time, thereby calculating and compensating for positional deviations. This type of method is intuitive in principle, independent of scene content, and is typically used as the first-level coarse compensation method in image processing.

[0006] Autofocusing methods based on echo data directly estimate phase errors from radar echoes, offering higher compensation accuracy. Classical parametric methods, such as the Map Drift (MD) algorithm and its derivatives, estimate error parameters by assuming the phase error follows a low-order polynomial model, resulting in high computational efficiency. Non-parametric methods, such as the Phase Gradient Autofocus (PGA) algorithm, can estimate phase errors of arbitrary forms and exhibit strong robustness. For low signal-to-noise ratio scenarios, researchers have further proposed improved algorithms such as Weighted PGA (WPGA) and Quality PGA (QPGA). For high-frequency motion errors, the high-performance Multi-aperture PGA (MS-PGA) has been proposed. This method utilizes sub-aperture partitioning, estimating the phase error for each individual sub-aperture and then stitching them together across the entire aperture range to achieve effective compensation for high-frequency vibrations. For the correction of spatially variable phase errors, existing solutions often employ a "divide and conquer" strategy, dividing the scene into multiple local regions and approximating the spatially invariant error in each local region.

[0007] However, given the extremely short wavelength of terahertz SAR and its high sensitivity, as well as the spatially varying errors encountered in various scenarios, existing technologies struggle to balance accuracy and efficiency. Traditional PGA algorithms cannot handle spatially varying errors, while scene segmentation, although capable of handling spatial variations, requires extremely fine sub-block partitioning to approximate the drastically changing error curve. This results in a linear increase in computational load with the number of blocks, making it difficult to guarantee the algorithm's real-time performance. Summary of the Invention

[0008] In view of this, in order to at least partially solve at least one of the aforementioned technical problems, this invention provides a high-efficiency autofocusing imaging method and system for terahertz SAR based on RB-PCA. Combining range block (RB) and sub-aperture segmentation strategies, the full aperture data is divided into sub-apertures for processing, achieving high-precision two-dimensional spatially varying motion error compensation for terahertz SAR. This algorithm can accurately compensate for phase errors caused by high-frequency vibrations in the terahertz band, while range block processing also addresses spatially varying phase errors caused by two-dimensional spatially varying motion errors. Furthermore, within this framework, Principal Component Analysis (PCA) dimensionality reduction and adaptive PGA are used to efficiently estimate phase gradient errors in the sub-apertures, greatly improving the algorithm's computational efficiency while ensuring estimation accuracy. The core scheme can be summarized as the RB-PCA method, aiming to combine the two-dimensional spatially varying error processing capability of range block with the fast computation capability of PCA dimensionality reduction. The technical solution is as follows:

[0009] According to one aspect of the present invention, a high-efficiency autofocusing imaging method for terahertz SAR based on RB-PCA is provided, comprising: S1: acquiring raw echo signal data and preprocessing it to obtain preprocessed echo signal data; S2: dividing the preprocessed echo signal data into range blocks in the range direction to obtain multiple range blocks, and performing sub-aperture segmentation within each range block to obtain multiple sub-aperture signal data; S3: performing deskewing processing on each sub-aperture signal data to obtain deskewing signal data; S4: performing PCA processing on the deskewing signal data and selecting the eigenvector with the largest eigenvalue as the target eigenvector; S5: obtaining the phase error gradient estimate of each sub-aperture by adaptive PGA estimation of the target eigenvector; S6: obtaining the two-dimensional spatially varying phase error of each range block by globally stitching and fusing the phase error gradient estimates of each sub-aperture; and S7: performing phase compensation and azimuth compression on the preprocessed echo signal data based on the two-dimensional spatially varying phase error to obtain the focused imaging result.

[0010] According to an embodiment of the present invention, S1: obtaining raw echo signal data and performing preprocessing to obtain preprocessed echo signal data includes: performing range compression on the raw echo signal data using fast Fourier transform; and performing range migration correction on the range-compressed echo signal data using a frequency scaling algorithm based on step frequency.

[0011] According to an embodiment of the present invention, S2: the preprocessed echo signal data is divided into multiple range blocks in the range direction. For each range block, sub-aperture segmentation is performed within the range block to obtain sub-aperture signal data, including: dividing the range data into multiple range blocks based on a near-field spatial variation threshold, such that the phase error within each range block is approximately empty and unchanged; and dividing the full aperture signal within each range block into multiple overlapping or non-overlapping sub-aperture signal data.

[0012] According to an embodiment of the present invention, the length of the sub-aperture segment is determined based on the condition that the residual phase error of the parabolic approximation is less than a preset threshold.

[0013] According to an embodiment of the present invention, S3: Obtaining de-skewed signal data by de-skewing the sub-aperture signal data includes: constructing a de-skewing function by parabolic approximating the slant distance equation of the sub-aperture; using the de-skewing function to de-skew the sub-aperture signal data to eliminate the second-order and higher-order terms in the sub-aperture signal data, thereby obtaining de-skewed signal data containing only the phase error term.

[0014] According to an embodiment of the present invention, S4: performing PCA processing on the deskewing signal data and selecting the eigenvector with the largest eigenvalue as the target eigenvector includes: constructing the autocorrelation matrix of the sub-aperture signal data, and performing eigenvalue decomposition on the autocorrelation matrix to extract the principal eigenvector corresponding to the largest eigenvalue as the target eigenvector.

[0015] According to an embodiment of the present invention, S5: obtaining the phase error gradient estimate of each sub-aperture by adaptive PGA estimation of the target feature vector includes: performing Fourier transform on the target feature vector to the frequency domain; determining the window width by adaptive windowing based on envelope; and obtaining the phase error gradient estimate of the sub-aperture by maximum likelihood estimation.

[0016] According to an embodiment of the present invention, S6: The two-dimensional spatially variable phase error of each range block is obtained by globally stitching and fusing the estimated values ​​of the phase error gradient of each sub-aperture, including: stitching the estimated values ​​of the phase error gradient of each sub-aperture within the same range block using a weighted least squares method based on the energy weight of the overlapping region to obtain the full aperture phase error gradient of the range block; integrating and smoothing the full aperture phase error gradient of each range block to obtain the full aperture phase error of each range block; and weighting and fusing the full aperture phase errors of different range blocks based on the distance from the range cell to the center of each range block to obtain the two-dimensional spatially variable phase error.

[0017] According to an embodiment of the present invention, the raw echo signal data is acquired by a vehicle-mounted or airborne terahertz synthetic aperture radar of the stepped frequency continuous wave (SFCW) system.

[0018] According to another embodiment of the present invention, a high-efficiency autofocusing imaging system for terahertz SAR based on RB-PCA is also provided to implement the above imaging method. The imaging system includes: a data preprocessing unit, a sub-aperture segmentation unit, a deskewing unit, a PCA processing unit, a PGA processing unit, a stitching and fusion unit, a compensation and compression unit, and an imaging display unit. The data preprocessing unit is configured to acquire raw echo signal data and perform preprocessing to obtain preprocessed echo signal data; the sub-aperture segmentation unit is configured to divide the preprocessed echo signal data into range blocks in the range direction to obtain multiple range blocks, and for each range block, perform sub-aperture segmentation within the range block to obtain multiple sub-aperture signal data; the de-skewing unit is configured to de-skewing each sub-aperture signal data to obtain de-skewing signal data; the PCA processing unit is configured to perform PCA processing on the de-skewing signal data and select the eigenvector with the largest eigenvalue as the target eigenvector; the PGA processing unit is configured to obtain the phase error gradient estimate of each sub-aperture by adaptive PGA estimation of the target eigenvector; the stitching and fusion unit is configured to obtain the two-dimensional spatially varying phase error of each range block by globally stitching and fusing the phase error gradient estimates of each sub-aperture; the compensation and compression unit is configured to perform phase compensation and azimuth compression on the preprocessed echo signal data based on the two-dimensional spatially varying phase error to obtain the focused imaging result; and the imaging display unit is configured to display the imaging result. Attached Figure Description

[0019] The objects, features, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0020] Figure 1 This is a geometric schematic diagram of a front-side-view stripe SAR radar according to an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the distance block and sub-aperture segmentation in an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram illustrating the advantages of PCA processing in an embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram of phase gradient stitching according to an embodiment of the present invention.

[0024] Figure 5a This is a schematic diagram of the efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to an embodiment of the present invention.

[0025] Figure 5b This is a schematic diagram of the workflow architecture of the high-efficiency autofocusing imaging method for terahertz SAR based on RB-PCA according to an embodiment of the present invention.

[0026] Figure 6 This is a schematic diagram of the composition of a high-efficiency autofocusing imaging system for terahertz SAR based on RB-PCA according to an embodiment of the present invention.

[0027] Figure 7 This diagram illustrates a comparison of the effects and performance indicators of the self-focusing imaging method of this invention with those of RDA and MS-PGA technologies.

[0028] Figure 8 This is a schematic diagram comparing the details of two point targets with the self-focusing imaging method of the present invention and the RDA and MS-PGA technical solutions.

[0029] Figure 9 This is a schematic diagram comparing the self-focusing imaging method of the present invention with the RDA and MS-PGA technical solutions regarding the azimuth cross section and indicators of two point targets. Detailed Implementation

[0030] This invention provides a high-efficiency autofocusing imaging method and system for terahertz SAR based on RB-PCA. Specifically, it relates to a high-efficiency motion compensation and imaging method for terahertz SAR on a vehicle-mounted or airborne platform, which is particularly suitable for solving the problem of high-precision real-time imaging under the coupling of high-frequency vibration and two-dimensional spatially variable phase error.

[0031] The main problems with existing technologies are: compensation methods based on high-precision motion sensors cannot meet the extremely high motion measurement accuracy requirements in the terahertz band. Existing commercial and even tactical sensors are often limited by issues such as measurement accuracy, drift accumulation, and insufficient sampling rate, making it difficult to capture the high-frequency mechanical vibrations common in vehicle platforms and failing to meet the sub-millimeter accuracy requirements of terahertz SAR imaging. Therefore, compensation relying solely on navigation data often leaves significant phase errors, necessitating the use of data-driven methods for fine correction.

[0032] Self-focusing methods based on echo data offer higher compensation accuracy, but still face significant challenges in the terahertz band. Parametric MOCO methods require rigorous parametric models to model motion errors. Low-order models commonly used in the microwave band often lack sufficient fitting ability when dealing with high-frequency nonlinear vibrations, while high-order models significantly increase algorithm complexity and the difficulty of model fitting. Non-parametric models such as PGA heavily rely on strong scattering points in the scene, and these algorithms are typically based on the assumption of "space-invariant error within the field of view." In some SAR imaging geometries, phase error varies drastically with slant range, exhibiting significant space-variability, causing the traditional space-invariant assumption to fail, and unified compensation across the entire aperture cannot achieve full-scene focusing. Meanwhile, while MS-PGA has strong high-frequency vibration compensation capabilities, its algorithmic complexity is high, and it also relies on strong scattering points within sub-apertures. For compensating space-variant motion errors, existing methods typically use range-based block processing, approximating space-invariant motion errors within a single range block. While this method solves the problem of spatial variation to some extent, it often requires extremely fine sub-block partitioning in order to approximate the drastically changing error curve, which leads to a linear increase in computational complexity with the number of blocks, severely restricting real-time processing capabilities.

[0033] The main technical problems to be solved by this invention include: addressing the difficulty of simultaneously and rapidly compensating for two-dimensional spatially varying phase errors caused by high-frequency platform vibration and large synthetic bandwidth in vehicle-mounted / airborne terahertz SAR imaging with high accuracy using existing algorithms; resolving the phase discontinuity and boundary effects problems existing in the traditional sub-aperture stitching process; solving the problem of inaccurate phase error estimation in low signal-to-noise ratio regions; and addressing the problem of excessive iterations during processing.

[0034] This invention presents a high-efficiency autofocusing imaging method and system for terahertz SAR based on RB-PCA. Using echo data acquired from a vehicle-mounted SAR system in a real-world scene, motion compensation is performed using this algorithm to obtain a clear terahertz SAR image. The imaging results achieve optimal focusing quality (image entropy, contrast) and optimal scene point target quality (3dB main lobe width, peak-to-side-lobe ratio, integral-to-side-lobe ratio), while maintaining approximately 10 times faster computation speed compared to the superior MS-PGA algorithm.

[0035] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0036] In this embodiment of the invention, combined with Figure 5a and Figure 5b As shown, a high-efficiency autofocusing imaging method and system for terahertz SAR based on RB-PCA is provided, which includes the following operations or steps:

[0037] S1: Acquire the raw echo signal data and perform preprocessing to obtain preprocessed echo signal data;

[0038] S2: Divide the preprocessed echo signal data into multiple range blocks in the range direction. For each range block, perform sub-aperture segmentation within the range block to obtain multiple sub-aperture signal data.

[0039] S3: Obtain deskewing signal data by deskewing the signal data of each sub-aperture;

[0040] S4: Perform PCA processing on the descrambling signal data and select the feature vector with the largest eigenvalue as the target feature vector;

[0041] S5: The phase error gradient estimate of each sub-aperture is obtained by adaptive PGA estimation of the target feature vector;

[0042] S6: The two-dimensional spatially varying phase error of each range block is obtained by globally stitching and fusing the estimated phase error gradient values ​​of each sub-aperture; and

[0043] S7: Based on the two-dimensional spatial phase error, phase compensation and azimuth compression are performed on the preprocessed echo signal data to obtain the focused imaging result.

[0044] According to an embodiment of the present invention, S1: Obtaining raw echo signal data and performing preprocessing to obtain preprocessed echo signal data includes:

[0045] Distance compression is performed on the raw echo signal data using Fast Fourier Transform; and

[0046] Range migration correction is performed on the range-compressed echo signal data using a frequency scaling algorithm based on step frequency.

[0047] According to an embodiment of the present invention, the raw echo signal data is acquired by a vehicle-mounted or airborne terahertz synthetic aperture radar of the stepped frequency continuous wave (SFCW) system.

[0048] Specifically, the terahertz SAR of this invention is a vehicle-mounted or airborne terahertz SAR. Taking a vehicle-mounted terahertz SAR system as an example, in a vehicle-mounted terahertz SAR system, the radar can adopt a stepped frequency configuration, such as... Figure 1 As shown, the radar uses a front-side-looking strip imaging mode, and under ideal conditions, it achieves a speed... It moves at a constant velocity in a straight line along the Y direction. The received echo signal has been sampled and mixed; the ideal signal model is:

[0049] (1);

[0050] in, Radar spatial location To target A Instantaneous slant distance, The shortest slant range between the radar and the target point. For radar altitude, This refers to the azimuth and time. In the radar system parameters, The starting frequency of the step signal. The step frequency, The number of sampling points in the distance direction. This is the sub-frequency index for the stepped frequency. The speed of light in free space is assumed, and the target backscattering coefficient is assumed to be 1. This signal is based on an ideal stepped-frequency SAR echo model, and the signal scale is... , Number of sampling points in the azimuth direction.

[0051] In formula (1) This is the radar azimuth window function, which is related to the antenna pattern. It can be approximated as a sinc-type function as follows:

[0052] (2);

[0053] in, This refers to the radar antenna beamwidth. However, the actual radar trajectory is not a uniform linear motion, such as... Figure 1 As shown by the red curve. At this point, there will be a shift compared to the ideal radar position; the actual position can be represented as:

[0054] (3);

[0055] in, , and These represent the actual track position offsets in the range, azimuth, and altitude directions, respectively. At this point, the actual slant range of the radar platform from the target will become:

[0056] (4);

[0057] Perform a Taylor expansion and simplify the above equation:

[0058] (5);

[0059] in:

[0060] (6);

[0061] (7);

[0062] (8);

[0063] in The angle between the radar line of sight and the vertical direction. , . The change in slant distance caused by motion error along the flight path. The quadratic term of the vertical motion error. Motion error perpendicular to the flight path and The projection component along the radar line of sight (LOS). and From point target A location In SAR systems, sensors such as IMU / GPS are used to achieve azimuth sampling at equal intervals by adjusting the PRF that transmits THz pulses, thus compensating for errors along the flight path. , Suppose that the range frequency of the signal is... Therefore, the actual echo is represented as follows:

[0064] (9);

[0065] The actual echo signal received at this time It exists in the range frequency domain and the azimuth time domain. The core of the autofocusing imaging method of this invention is handling the LOS error. Error with quadratic term These motion errors all vary with azimuth and time. Change, and and The high-frequency vibration error incorporated in this study is not negligible in the terahertz band. Specifically, the LOS error is related to the radar's downward viewing angle. This correlation results in a spatially variable phase error, which varies with distance. Quadratic term error. Introducing nonlinear phase changes can cause nonlinear distortion in the image.

[0066] Before performing phase error estimation and correction, range migration correction (RCMC) and range compression are required for the echo signal. Due to the nature of stepped-frequency radar, range compression can be achieved by performing a fast inverse Fourier transform (IFFT) on equation (9) in the range direction:

[0067] (10);

[0068] in For radar signal wavelength, This refers to the radar center frequency. Spatial phase error. for:

[0069] (11);

[0070] Distance to envelope for:

[0071] (12);

[0072] Subsequently, a frequency scaling algorithm based on stepped-frequency matrix (SF-FSA) was used to perform fast RCMC on the range-compressed signal. Compared to traditional RCMC using interpolation, SF-FSA performs range migration correction based on Fast Fourier Transform (FFT), resulting in significantly higher computational efficiency. The signal after RCMC is as follows:

[0073] (13);

[0074] in:

[0075] (14);

[0076] At this time, the distance envelope and the azimuth time It's irrelevant; the signal is compressed to the range unit. The echo signal preprocessing is now complete; the next step will be to analyze the spatially varying phase error. The specific correction method.

[0077] According to an embodiment of the present invention, S2: the preprocessed echo signal data is divided into multiple range blocks in the range direction. For each range block, sub-aperture segmentation is performed within the range block to obtain sub-aperture signal data, including:

[0078] Based on the near-field spatial variation threshold, the range data is divided into multiple range blocks, so that the phase error in each range block is approximately empty and invariant;

[0079] It divides the full aperture signal within each distance block into multiple sub-aperture signal data, which may or may not overlap.

[0080] Specifically, the distance-oriented block is used to address spatially varied phase errors due to... The resulting changes are addressed by sub-aperture segmentation, which is used to compensate for high-frequency phase errors caused by high-frequency vibrations of the radar platform. The self-focusing imaging method of this invention employs a method of first segmenting by range and then by sub-aperture segmentation, minimizing computational complexity.

[0081] From equation (8), it can be seen that different range units correspond to different downward viewing angles. Utilizing the spatial locality of spatially varying errors, the range direction is divided into... There are several blocks, and the downward viewing angle changes little within each block, with approximately consistent phase errors (unchanged in empty space). Let the length of each block be... meters, corresponding number of sampling points The requirement is that the change in phase error within the distance block is less than an acceptable threshold.

[0082] (15);

[0083] The change in the distance block angle can be obtained from geometric relationships. With distance block length satisfy:

[0084] (16);

[0085] The ground geometry corresponding to the joint distance block can be obtained as follows:

[0086] (17);

[0087] By combining known radar parameters with motion error scales for different scenarios, the approximate range of range block lengths can be obtained. Generally, the closer to the radar, the greater the range block length constraint, requiring the use of smaller range blocks; while the farther away from the radar, the smaller the range block length constraint, allowing the use of larger range blocks. In practice, considering both computational efficiency and the spatial invariance threshold of phase error, an intermediate value can be selected as the range block length.

[0088] By block length Divide the data along the distance into blocks. It can be represented as:

[0089] (18);

[0090] in To round up, This is the overlap coefficient between adjacent blocks, typically taken as 50%. To detect the farthest slant range, This is the nearest slope distance. For the ... A distance block, whose time range can be represented as:

[0091] (19);

[0092] No. Each distance block echo data is represented as :

[0093] (20);

[0094] The phase error within the range block can be approximated as: ,and Irrelevant.

[0095] To capture high-frequency vibrations, sub-aperture segmentation is performed within each range block. The length of each sub-aperture segment is determined based on the condition that the residual phase error of the parabolic approximation is less than a preset threshold. The full aperture signal within each range block is segmented into segments of length [missing information]. The sub-aperture signal at each sampling point, with a spacing between adjacent sub-apertures is... Sampling points. When the sub-aperture spacing At that time, there is overlap between adjacent sub-apertures. The number of sub-apertures divided. It can be represented as:

[0096] (twenty one);

[0097] Let the azimuth sampling rate be After the sub-aperture is divided, the first The center time of each aperture is , No. The azimuth time range of the individual aperture is ,in Therefore, the first result can be obtained through variable substitution. The distance block of the first Echo data of individual apertures :

[0098] (twenty two);

[0099] in The formula for the slope distance of the sub-aperture is given, and The azimuth time interval between the center of the sub-aperture and the target point. , At this point, within a single sub-aperture, not only can the influence of spatial variation error be ignored, but also higher-order phase errors... Within the sub-aperture, it can be approximated as a lower order. This improves the robustness and noise resistance of subsequent phase error estimation, and enhances its adaptability to high-frequency vibrations during motion. A schematic diagram of the distance block and sub-aperture segmentation is shown below. Figure 2 As shown.

[0100] According to an embodiment of the present invention, S3: the de-skewing processing of each sub-aperture signal data to obtain de-skewed signal data includes:

[0101] The slope distance equation of the sub-aperture is approximated by a parabola to construct a deslope function;

[0102] The deskewing function is used to deskew the sub-aperture signal data to eliminate the second-order and higher-order terms in the sub-aperture signal data, resulting in deskewed signal data containing only the phase error term.

[0103] Specifically, sub-aperture data It cannot yet be directly used for phase error gradient estimation because its phase contains... Quadratic and higher-order terms will affect The estimation produces significant interference. Echo phase de-chewing is required to eliminate this interference. The second and higher-order terms. Pair aperture slant distance equation. Perform a parabolic approximation and discard the curve. The higher-order terms yield:

[0104] (twenty three);

[0105] Among them, set The deslope function can be expressed as: This function can eliminate very precisely. Quadratic terms, but The parabola is approximately neglected The higher-order terms, while important, become non-negligible as the synthetic aperture length increases. Therefore, to ignore the residual error caused by the parabolic approximation, the sub-aperture length needs to be limited. The residual phase error caused by the ignored higher-order terms can be expressed as:

[0106] (twenty four);

[0107] To use a parabolic approximation, the residual phase error must be within an acceptable range:

[0108] (25);

[0109] This can be considered as the azimuth distance between the center of the sub-aperture and the target. Within the sub-aperture, this distance can be at most half the length of the sub-aperture. And based on the sub-aperture time... The range of values ​​can be obtained. The maximum value is also half the sub-aperture length. Therefore, to satisfy equation (25), the sub-aperture length must satisfy:

[0110] (26);

[0111] in For sub-aperture time, The azimuth sampling rate is used, and the maximum sub-aperture length can be obtained from the radar system parameters. The sub-aperture length cannot be too small, as an excessively small sub-aperture length will cause the phase error gradient within the sub-aperture to approximate a constant, leading to significant errors in subsequent PGA estimation. Generally, the closest slant range is selected. The upper limit of the sub-aperture length calculated at the location is used as the sub-aperture segmentation length.

[0112] De-skew the sub-aperture signal that satisfies the parabolic approximation to obtain a de-skewed signal that can be used for phase error estimation:

[0113] (27);

[0114] Thus, in the phase of the sub-aperture signal Both the second-order and higher-order terms originate from phase error. .

[0115] According to an embodiment of the present invention, S4: performing PCA processing on the descrambling signal data and selecting the feature vector with the largest eigenvalue as the target feature vector includes:

[0116] Construct the autocorrelation matrix of the sub-aperture signal data, and perform eigenvalue decomposition on the autocorrelation matrix to extract the principal eigenvector corresponding to the largest eigenvalue as the target eigenvector.

[0117] Specifically, directly using PGA to estimate the phase error gradient of the sub-aperture signal is the implementation of the MS-PGA algorithm. This not only requires calculating the phase error gradient of all range cells in the sub-aperture, which is extremely time-consuming, but also faces the problem of no obvious isolated strong scattering points in different range cells and the lack of concentrated energy in the main lobe. This problem can be solved by using PCA to reduce the dimensionality of the data and extract the feature vectors of the sub-aperture signal. The matrix form of the sub-aperture signal can then be constructed. as follows:

[0118] (28);

[0119] In the signal matrix elements Represents the sub-aperture signal in , The sampling point data values ​​at that time For the first The number of distance sampling points within each distance block needs to be noted. Derived from equation (19), representing the first The start time of each distance block.

[0120] Meanwhile, let the phase error be... Diagonal matrix of order:

[0121] (29);

[0122] diagonal elements Representative sub-aperture phase error exist The value at time. The sub-aperture signal matrix can be represented as ,in This represents an ideal sub-aperture signal with no phase error. To perform PCA dimensionality reduction, we first construct the autocorrelation matrix using the signal matrix:

[0123] (30);

[0124] in The autocorrelation matrix of the ideal image is given by the superscript. This represents the conjugate transpose of a matrix. For autocorrelation... Eigenvalue decomposition of the matrix yields:

[0125] (31);

[0126] in This is an eigenvalue diagonal matrix, with the eigenvalues ​​arranged in descending order. . Let be the eigenvector matrix when phase error exists. for The corresponding eigenvector. This completes the PCA processing of the sub-aperture signal. Combining equations (30) and (31), it can be seen that the phase error does not change the eigenvalues ​​of the signal. The eigenvector matrix satisfies ,in Let be the eigenvector matrix of the ideal sub-aperture signal.

[0127] At this point, the largest eigenvalue is selected. Corresponding feature vector It not only concentrates the main signal energy of all range cells, but also preserves the sub-aperture phase error. All the details. For Performing a PGA is equivalent to performing a PGA on a "virtual echo" that integrates the energy and phase information of all range cells. Furthermore... Its energy is more concentrated, which better meets the requirements of PGA.

[0128] According to an embodiment of the present invention, S5: obtaining the phase error gradient estimate of each sub-aperture by adaptive PGA estimation of the target feature vector includes:

[0129] Perform a Fourier transform on the target feature vector to the frequency domain;

[0130] The window width is determined by adaptive windowing based on the envelope; and

[0131] The gradient estimate of the sub-aperture phase error is obtained by maximum likelihood estimation.

[0132] Specifically, for eigenvectors containing complete phase error Using adaptive PGA to estimate the sub-aperture phase error gradient First of all, Perform a Fast Fourier Transform to the frequency domain, which is represented as: Searching for spectral peaks .

[0133] Determine the peak value Then, it is moved to the center of the spectrum using a circular shift operation to obtain... Subsequently, an envelope-based adaptive windowing method is employed to perform peak envelope detection on the spectrum. This method detects the first slowly changing point during the extension from both sides of the peak, and the width of the window is determined by the distance between these two points, thus obtaining the windowed signal. .

[0134] Finally, the phase history domain is returned through inverse fast Fourier transform:

[0135] (32);

[0136] The phase gradient of the sub-aperture is estimated by maximum likelihood estimation, and the phase gradient can be filtered by moving average to suppress noise.

[0137] (33);

[0138] in express The conjugate of [the data] is used. In the PCA-driven adaptive PGA method for estimating the phase error gradient of sub-apertures, the energy-concentrated eigenvectors are selected for estimation, while the point selection and windowing operations in PGA are implemented efficiently and adaptively. The PCA data dimensionality reduction operation concentrates the energy of strong scattering regions and improves the isolation of strong scattering points, while the envelope-based windowing can effectively cope with different scenarios and has better adaptability. Figure 3 This demonstrates the advantages of the PCA method. The figure compares the dominant eigenvector of the range cell with the highest energy peak value directly with that of the PCA-processed data from a specific sub-aperture. The isolation metric is defined as the ratio of the peak value to the mean value in the spectrum; this metric can be used to evaluate the energy concentration of the main lobe of the signal. It can be seen that the PCA-processed data outperforms different high-energy range cells in both isolation and peak energy ratio, while PGA processing shows better performance when the isolation of strong target points is even higher.

[0139] According to an embodiment of the present invention, S6: the two-dimensional spatially varying phase error of each range block is obtained by globally stitching and fusing the estimated phase error gradient values ​​of each sub-aperture, including:

[0140] The estimated phase error gradient values ​​of each sub-aperture within the same range block are stitched together using the weighted least squares method based on the energy weight of the overlapping region to obtain the full aperture phase error gradient of the range block.

[0141] The full aperture phase error gradient of each distance block is integrated and smoothed to obtain the full aperture phase error of each distance block;

[0142] The full aperture phase error of different distance blocks is weighted and fused based on the distance from the distance cell to the center of each distance block to obtain the two-dimensional spatially variable phase error.

[0143] The weighted fusion of the full aperture phase errors of different distance blocks is achieved by using a Gaussian function to calculate the weighting coefficients, so as to achieve a smooth transition of the phase errors of adjacent distance blocks.

[0144] Specifically, the first is obtained through PCA-PGA. Within the distance block, the first The phase error gradient of each aperture is Generally, the sub-aperture error can be approximated as a polynomial:

[0145] (34);

[0146] achievable ,in represent First-order and higher-order terms. PCA-PGA estimates the higher-order terms, so the relationship between the estimated value and the true gradient can be expressed as: . Effective estimation cannot be performed in PGA. It is necessary to estimate the linear shift of the gradient between sub-apertures by performing a weighted least squares (WLS) operation on the overlapping region between sub-apertures.

[0147] The weighted least squares requirement yields the energy weights of the overlapping regions of the sub-apertures, the th and the The weighting of individual apertures is calculated as follows:

[0148] (35);

[0149] in Indicates the first and the The overlapping area of ​​individual apertures. This reflects the signal-to-noise ratio and coherence of the two sub-apertures within the overlapping region. A higher weight indicates a more reliable phase estimation between the sub-apertures. Let the overall offset vector to be estimated for each sub-aperture be:

[0150] (36);

[0151] The global weighted least squares optimization problem for the overall offset matrix is ​​as follows:

[0152] (37);

[0153] in This represents the set of sub-aperture pairs with overlapping regions. To eliminate overall translational uncertainty, the first sub-aperture is chosen as... Then, by constructing a sparse matrix, the solution is obtained. Finally, the phase gradient of each sub-aperture is offset and corrected, and then globally stitched together to obtain the first... Full aperture phase error gradient estimate for each distance block .

[0154] A smooth full-aperture phase error is obtained by integrating the gradient function:

[0155] (38);

[0156] The phase error obtained by integration will contain redundant linear phase. Linear fitting is needed to remove the linear phase in the phase error of each distance block in order to ensure that the phase error after stitching is more accurate.

[0157] After splicing the sub-apertures, the distance block is obtained. Full aperture phase error Cross-distance block fusion needs to be achieved to construct the final two-dimensional spatially variable phase error. .

[0158] A Gaussian weighted fusion method is used for distance cells. Weighted according to its distance to the center of each distance block:

[0159] (39);

[0160] Distance unit here With angle Satisfying Relationships , For distance resolution. At the level of the sampled discrete signals, they are essentially the same; to maintain consistency with the formula, we still use... To express.

[0161] The normalized weighting function adopts a Gaussian form that incorporates the distance block energy:

[0162] (40):

[0163] The kernel function is:

[0164] (41):

[0165] For the first The center of the distance block, For the first Energy weight of each distance block, smoothing factor This is an adjustable parameter, typically set to the distance between blocks. A larger value... This results in a smoother transition but at the cost of phase detail. Gaussian weights achieve a smooth transition in overlapping regions, avoiding distance-oriented seams in phase errors. Simultaneously, each distance cell primarily receives contributions from the nearest block, with the weights of more distant blocks decaying rapidly. For example... Figure 4 The diagram shows a global stitching and fusion of phase error gradient estimates.

[0166] According to an embodiment of the present invention, S7: Based on the two-dimensional spatially variable phase error, phase compensation and azimuth compression are performed on the preprocessed echo signal data to obtain the focused imaging result.

[0167] Specifically, the fused two-dimensional spatially variable phase error simultaneously satisfies the continuity in both the azimuth and range directions. The two-dimensional phase correction function is applied to compensate for the phase of equation (13):

[0168] (42);

[0169] Finally, azimuth compression is performed to obtain the focused SAR image:

[0170] (43);

[0171] The azimuth-directed matched filtering is expressed as follows:

[0172] (44);

[0173] Ultimately, the image correction and output of the focused imaging results are achieved.

[0174] In another embodiment of the present invention, a high-efficiency autofocusing imaging system for terahertz SAR based on RB-PCA is also provided to implement the above-described imaging method, such as... Figure 6 As shown, the imaging system includes:

[0175] The data preprocessing unit is configured to acquire raw echo signal data and perform preprocessing to obtain preprocessed echo signal data.

[0176] The sub-aperture segmentation unit is configured to divide the preprocessed echo signal data into multiple range blocks in the range direction, and to perform sub-aperture segmentation within each range block to obtain multiple sub-aperture signal data.

[0177] The deskewing unit is configured to deskew the sub-aperture signal data to obtain deskewed signal data.

[0178] The PCA processing unit is configured to perform PCA processing on the descrambling signal data and select the feature vector with the largest eigenvalue as the target feature vector.

[0179] The PGA processing unit is configured to obtain the phase error gradient estimate of each sub-aperture by adaptive PGA estimation of the target feature vector;

[0180] The stitching and fusion unit is configured to obtain the two-dimensional spatially variable phase error of each range block by globally stitching and fusing the phase error gradient estimates of each sub-aperture;

[0181] The compensation and compression unit is configured to perform phase compensation and azimuth compression on the preprocessed echo signal data based on the two-dimensional spatially variable phase error to obtain a focused imaging result; and

[0182] An imaging display unit is configured to display the imaging results.

[0183] Example 1:

[0184] Pedestrian walkway imaging was performed using a 220 GHz stepped-frequency vehicle-mounted terahertz SAR. (Radar altitude...) The radar is positioned on the roof of the vehicle, pointing downwards. The angle between the radar's line of sight and the horizontal plane is 35 degrees. The radar's azimuth resolution is... Distance resolution Data with a length of 10m in the azimuth direction is collected and displayed. The echo data is compensated using the autofocusing imaging method of this invention, wherein the range block and sub-aperture segmentation requirements of the autofocusing imaging method of this invention are set as follows: sub-aperture length Spacing between adjacent sub-apertures Distance block distance to sampling points Distance block overlap rate The imaging effects of the RDA-based scheme (without motion compensation), the MS-PGA-based scheme, and the autofocusing imaging method of this invention are compared as follows: Figure 7 As shown in the figure, the MS-PGA iteration count is 10, while the self-focusing imaging method of this invention has only 1 iteration. All algorithms are run on the same device. Comparing the image entropy, image contrast, and computation time of the images after imaging, it can be seen that the self-focusing imaging method of this invention is optimal in terms of image entropy and contrast, and its processing speed is extremely fast, improving by about 90% compared to MS-PGA, and even faster than RDA without motion compensation. Figure 6Two point targets are circled in red. Point target analysis yields the main lobe width (IWR), peak-to-side-lobe ratio (PSLR), and peak-to-integral ratio (ISLR). The RDA and MS-PGA schemes are also compared. Detailed point target analysis for the three schemes is shown below. Figure 8 As shown, the azimuth cross-sectional diagram of the point target is compared with specific indicators, for example... Figure 8 As shown. Figure 9 As shown, the autofocusing imaging method of the present invention is optimal in all three indicators (the smaller the better for all three indicators), and the 3dB width of the main lobe at point 1 is also very close to the ideal azimuth resolution. Although the performance of the self-focusing imaging method of the present invention is not significantly improved compared to MS-PGA, the self-focusing imaging method of the present invention only requires one iteration, while MS-PGA requires 10 iterations, and the computational efficiency of the self-focusing imaging method of the present invention is much higher than that of MS-PGA.

[0185] As can be seen from the above, the RB-PCA-based terahertz SAR high-efficiency autofocusing imaging method and system of the present invention adopts the following main technical solutions:

[0186] Distance-based partitioning strategy: Based on the near-field spatial variation threshold, the data is partitioned into blocks along the distance, decomposing the two-dimensional spatial variation problem into a locally spatial invariant problem.

[0187] PCA dimensionality reduction phase estimation: In sub-aperture processing, instead of processing each distance cell individually, the covariance matrix is ​​constructed and the principal eigenvector corresponding to the largest eigenvalue is extracted. This vector is then used as the sole representative signal of the sub-aperture for PGA phase estimation, thereby reducing the computational complexity by orders of magnitude.

[0188] Gradient domain weighted stitching: Phase gradient stitching is performed based on the least squares method with energy weighting in the overlapping region to ensure phase continuity across the entire aperture.

[0189] Gaussian weighted distance fusion: The phase error of adjacent distance blocks is smoothly fused using a Gaussian function to eliminate the block boundary effect in the range direction.

[0190] Adaptive PGA combination: Combining PCA and adaptive PGA greatly improves the efficiency of phase gradient estimation and solves the pain points of slow PGA iteration and high requirements for strong scattering points.

[0191] It should be noted that the principal eigenvector extracted using PCA (Eigenvalue Decomposition) in this scheme can also be extracted using Singular Value Decomposition (SVD) to extract the left singular vector corresponding to the largest singular value, or principal components can be extracted using Independent Component Analysis (ICA). The mathematical principles and technical effects are similar. During sub-aperture stitching, gradient domain stitching can use either weighted least squares or a simple weighted average method, or direct stitching in the phase domain (requiring phase unwrapping), although smoothness and noise resistance may decrease slightly. For distance block fusion: Gaussian weighted fusion can be replaced with linear weighted fusion (trapezoidal window) or sinc function weighted fusion. Any substitution of the above technical solutions according to actual application requirements falls within the scope of this invention.

[0192] Compared with the closest sub-aperture PGA algorithm to date (MS-PGA), the RB-PCA-based terahertz SAR efficient autofocusing imaging method and system proposed in this application have the following significant advantages:

[0193] The computational efficiency is improved by orders of magnitude: Traditional MS-PGA requires PGA iteration for all distance cells within each sub-aperture. This invention utilizes PCA technology to compress information into a single feature vector, and combined with adaptive PGA, achieves extremely high efficiency and very few iterations. With a data volume of 10m, the computational speed is improved by approximately 90%, and the improvement is even greater with larger data volumes.

[0194] Stronger robustness in low signal-to-noise ratio: Traditional methods rely on strong scattering points, which are prone to failure in weak texture regions. This invention uses PCA to automatically converge the energy of all range cells within the sub-aperture, significantly improving the signal-to-noise ratio of the equivalent signal and the isolation of strong points, while maintaining high accuracy even in weak texture regions.

[0195] Consistent Focus Across All Scenarios: Combining distance segmentation and Gaussian fusion, it accurately compensates for drastic spatial variation errors from near to far, solving the problem that traditional PGAs cannot handle spatial variation errors and edge defocus.

[0196] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. It should be noted that implementations not illustrated or described in the drawings or the main text of the specification are forms known to those skilled in the art and have not been described in detail. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A high-efficiency autofocusing imaging method for terahertz SAR based on RB-PCA, comprising: S1: Acquire the raw echo signal data and perform preprocessing to obtain preprocessed echo signal data; S2: Divide the preprocessed echo signal data into multiple range blocks in the range direction. For each range block, perform sub-aperture segmentation within the range block to obtain multiple sub-aperture signal data. S3: Obtain deskewing signal data by deskewing the signal data of each sub-aperture; S4: Perform PCA processing on the descrambling signal data and select the feature vector with the largest eigenvalue as the target feature vector; S5: The phase error gradient estimate of each sub-aperture is obtained by adaptive PGA estimation of the target feature vector; S6: The two-dimensional spatially varying phase error of each range block is obtained by globally stitching and fusing the estimated phase error gradient values ​​of each sub-aperture; and S7: Based on the two-dimensional spatial phase error, phase compensation and azimuth compression are performed on the preprocessed echo signal data to obtain the focused imaging result.

2. The efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to claim 1, characterized in that, S1: Acquiring raw echo signal data and performing preprocessing to obtain preprocessed echo signal data includes: Distance compression is performed on the raw echo signal data using Fast Fourier Transform; and Range migration correction is performed on the range-compressed echo signal data using a frequency scaling algorithm based on step frequency.

3. The efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to claim 1, characterized in that, S2: The preprocessed echo signal data is divided into range blocks in the range direction to obtain multiple range blocks. For each range block, sub-aperture segmentation is performed within the range block to obtain sub-aperture signal data, including: Based on the near-field spatial variation threshold, the range data is divided into multiple range blocks, so that the phase error in each range block is approximately empty and invariant; It divides the full aperture signal within each distance block into multiple sub-aperture signal data, which may or may not overlap.

4. The efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to claim 3, characterized in that, The length of the sub-aperture segment is determined based on the condition that the residual phase error of the parabolic approximation is less than a preset threshold.

5. The efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to claim 4, characterized in that, S3: Deskewing the sub-aperture signal data to obtain deskewing signal data includes: The slope distance equation of the sub-aperture is approximated by a parabola to construct a deslope function; The deskewing function is used to deskew the sub-aperture signal data to eliminate the second-order and higher-order terms in the sub-aperture signal data, resulting in deskewed signal data containing only the phase error term.

6. The efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to claim 1, characterized in that, Step S4: Performing PCA processing on the descrambled signal data and selecting the feature vector with the largest eigenvalue as the target feature vector includes: Construct the autocorrelation matrix of the sub-aperture signal data, and perform eigenvalue decomposition on the autocorrelation matrix to extract the principal eigenvector corresponding to the largest eigenvalue as the target eigenvector.

7. The efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to claim 1, characterized in that, Step S5: Obtain the phase error gradient estimate of each sub-aperture by adaptive PGA estimation of the target feature vector. include: Perform a Fourier transform on the target feature vector to the frequency domain; The window width is determined by adaptive windowing based on the envelope; as well as The gradient estimate of the sub-aperture phase error is obtained by maximum likelihood estimation.

8. The efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to claim 1, characterized in that, S6: The two-dimensional spatially varying phase error of each range block is obtained by globally stitching and fusing the estimated phase error gradient values ​​of each sub-aperture, including: The estimated phase error gradient values ​​of each sub-aperture within the same range block are stitched together using the weighted least squares method based on the energy weight of the overlapping region to obtain the full aperture phase error gradient of the range block. The full aperture phase error gradient of each distance block is integrated and smoothed to obtain the full aperture phase error of each distance block; The full aperture phase error of different distance blocks is weighted and fused based on the distance from the distance cell to the center of each distance block to obtain the two-dimensional spatially variable phase error.

9. The efficient autofocusing imaging method for terahertz SAR based on RB-PCA according to claim 1, characterized in that, The raw echo signal data was acquired by a vehicle-mounted or airborne terahertz synthetic aperture radar using a stepped-frequency continuous wave (SFCW) system.

10. A high-efficiency autofocusing imaging system for terahertz SAR based on RB-PCA, used to implement the imaging method according to any one of claims 1-9, the imaging system comprising: The data preprocessing unit is configured to acquire raw echo signal data and perform preprocessing to obtain preprocessed echo signal data. The sub-aperture segmentation unit is configured to divide the preprocessed echo signal data into multiple range blocks in the range direction, and to perform sub-aperture segmentation within each range block to obtain multiple sub-aperture signal data. The deskewing unit is configured to deskew the sub-aperture signal data to obtain deskewed signal data. The PCA processing unit is configured to perform PCA processing on the descrambling signal data and select the feature vector with the largest eigenvalue as the target feature vector. The PGA processing unit is configured to obtain the phase error gradient estimate of each sub-aperture by adaptive PGA estimation of the target feature vector; The stitching and fusion unit is configured to obtain the two-dimensional spatially variable phase error of each range block by globally stitching and fusing the phase error gradient estimates of each sub-aperture; The compensation and compression unit is configured to perform phase compensation and azimuth compression on the preprocessed echo signal data based on the two-dimensional spatially variable phase error to obtain the focused imaging result. as well as An imaging display unit is configured to display the imaging results.

Images