A radar raw echo data blind focusing imaging method

Abstract

The application discloses a radar original echo data blind focusing imaging method, which comprises the following steps: based on the approximate periodicity model of one-dimensional original data amplitude signal and the PCM method, impulse repetition interval coarse estimation and fine estimation are carried out, so that the one-dimensional data is restored to a two-dimensional data matrix; the approximate matching filter model of the radar image is established by using the reference point echo and the approximate shift invariance model; the original two-dimensional echo data is divided into multiple blocks, the energy of each block is normalized, the reference echo is estimated by using the PCM method; and the radar image is restored by using the estimated reference echo through the matching filter method. The application can restore high-quality radar images from the original data without depending on the radar system parameters.

Description

Technical Field

[0001] This invention belongs to the field of radar technology, specifically relating to a blind focusing imaging method based on raw radar echo data. Background Technology

[0002] Radar imaging, such as synthetic aperture radar (SAR), can acquire high-resolution ground images, playing a vital role in fields such as military reconnaissance, environmental monitoring, and topographic mapping. The core of radar imaging is the focusing algorithm, which processes raw data to generate high-quality images.

[0003] Traditional radar image focusing algorithms are mainly divided into two categories: frequency domain algorithms, including range-Doppler algorithms, linear frequency modulated (LFM) scaling algorithms, range migration algorithms, and their variants; and time domain algorithms, including back projection algorithms and their variants. Existing algorithms all share a common limitation: they require knowledge of radar system parameters to generate images, such as wavelength, center slant range, fast time sampling rate, pulse repetition interval (PRI), waveform parameters (e.g., the modulation frequency when using a LFM signal), and platform velocity. However, in certain scenarios, such as when the user cannot obtain system parameters or when the illumination source is a non-cooperative target, these algorithms cannot generate images. Summary of the Invention

[0004] The purpose of this invention is to provide a blind focusing imaging method for raw radar echo data, which can recover high-quality radar images from raw data without relying on radar system parameters.

[0005] The technical solution for achieving the objective of this invention is: a blind focusing imaging method using raw radar echo data, comprising:

[0006] Step 1: Based on the approximate periodic model of the one-dimensional original data amplitude signal and the principal component maximization method, perform coarse and fine estimation of PRI to restore the one-dimensional data to a two-dimensional data matrix;

[0007] Step 2: Using the reference point echo and the approximate translation invariance model, establish an approximate matched filtering model for the radar image;

[0008] Step 3: By dividing the original two-dimensional echo data into multiple blocks and normalizing the energy of each block, the PCM method estimates the reference echo.

[0009] Step 4: Using the estimated reference echo, the radar image is recovered through matched filtering.

[0010] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0011] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the above-described method.

[0012] A computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method.

[0013] Compared with existing technologies, the advantages of this invention are as follows: This invention proposes an approximate matched filtering model for representing radar images using the echoes of a reference point, leveraging the approximate translation invariance of the echoes; this model uses unknown reference echoes for image formation; to estimate this reference, this invention develops a Principal Component Maximization (PCM) method, utilizing the low-dimensional structure of the reference echo; PCM processes the original two-dimensional data by dividing it into blocks, normalizing their energy, and maximizing the principal component energy of all blocks to accurately estimate the reference echo under non-stationary clutter; furthermore, forming the 2-D original echo form of 1-D original data requires knowledge of the PRI; for cases where the PRI is unknown, this invention proposes a two-step estimation method that can adaptively recover the 2-D original echo form of 1-D original data. This invention can effectively recover high-quality images from raw radar data without knowing any radar system parameters. Attached Figure Description

[0014] To more clearly illustrate the embodiments of this application or the existing technical solutions, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 The diagram illustrates the transformation of one-dimensional data into two-dimensional data, where (a) represents one-dimensional data; (b) shows the alignment of the integer portions of the sample PRI to form two-dimensional data; and (c) shows the shifting of each pulse observation data by a fractional sampling interval to align with the start of each observation interval.

[0016] Figure 2 This diagram illustrates the coarse estimate of PRI obtained from a single-wavelength signal peak. The horizontal axis of the peak value represents the estimate, which is 0.0017812. The coarse estimate of PRI is the reciprocal of this value, i.e.

[0017] Figure 3To accurately estimate PRI using PCM, the peak horizontal axis represents the estimated value, and lines 1, 2, and 3 represent the three estimation results, with the third obtaining the final accurate estimate.

[0018] Figure 4 This is an example of dividing the original two-dimensional data into blocks. In this example, the block size is 500x500.

[0019] Figure 5 Example of echo estimation for reference.

[0020] Figure 6 Here is an example of the imaging results. Detailed Implementation

[0021] This invention proposes a blind focusing imaging method using raw radar echo data. The method first proposes an approximate matched filtering model to represent the radar image using the echo from a reference point, leveraging the approximate translation invariance of the echo. This model uses an unknown reference echo for image formation. To estimate this reference, this invention develops a Principal Component Maximization (PCM) method, utilizing the low-dimensional structure of the reference echo. PCM processes the raw 2D data by dividing it into blocks, normalizing their energy, and maximizing the principal component energy of all blocks to accurately estimate the reference echo under non-stationary clutter. Furthermore, forming the 2D raw echo form of 1D raw data requires knowledge of the pulse repetition interval (PRI). For cases where the PRI is unknown, this invention proposes a two-step estimation method that adaptively recovers the 2D raw echo form of 1D raw data. Compared to existing imaging methods, this invention can effectively recover high-quality images from raw radar data without knowing any radar system parameters.

[0022] The present invention provides a blind radar image recovery method based on principal component maximization, comprising:

[0023] Step 1: Based on the approximate periodic model of the one-dimensional original data amplitude signal and the principal component maximization (PCM) method, perform coarse and fine estimation of PRI to restore the one-dimensional data to a two-dimensional data matrix;

[0024] Recovering a two-dimensional (2-D) data matrix from a one-dimensional (1-D) raw data vector, with the pulse repetition interval (PRI) unknown; the radar echo data stream is initially 1-D raw data. Only when the pulse repetition interval is multiplied by the sampling rate (denoted as P = Tf) can a two-dimensional (2-D) data matrix be recovered. s Where T is the pulse repetition interval, f sOnly when the sampling rate is known can it be formed into a 2-D original data matrix; when this parameter is unknown, it must be estimated from the data to reconstruct the 2-D original data for subsequent PCM and image reconstruction processing; to this end, we propose a two-step method to estimate the sample PRI and recover the 2-D original data matrix from the 1-D original data.

[0025] Parametric models for converting 1-D raw data into 2-D raw data are presented. First, we discuss the 2-D data formation model when the PRI (Primary Indicator) is known. The sample PRI is a parameter of this model, and we can view the estimation of PRI as a model parameter inversion problem. The relationship between the time axis t' of the 1-D observed signal and the time axis of its corresponding radar 2-D observed signal is as follows:

[0026] t′=t+τ=lT+τ (1)

[0027] Here, the slow time of the radar is represented as t = lT, where l represents the l-th pulse observation interval, considering that the slow time is discrete; τ is the fast time of the synthetic aperture radar; the start time of the l-th pulse observation interval is t = lT; therefore, the 2-D signal can be expressed using a 1-D signal as follows:

[0028] Y(l,τ)=y(t') (2)

[0029] Here, Y(l,τ) is the radar's two-dimensional (slow-time-fast-time) observation signal. The above model uses a continuous signal, but the sampled data is discrete; therefore, we need to construct a discrete 2D data formation model. The sample PRI can be represented in the discrete domain as...

[0030] P = Tf s =P int +P dec (3)

[0031] Where P int It is the integer part, P dec It is the decimal part; the starting point of the l-th observation interval can be represented as

[0032]

[0033] This point may not typically be on the sampling grid, and the derivation of this point to nearby sampling grids varies between different l; we must align the starting points in the 2-D data matrix to maintain coherent information between different observation intervals.

[0034] The calibration of the l-th echo can be performed by sampling in the discrete domain at fractional sampling intervals N. l,dec This is achieved by moving data slices; this process can be represented as...

[0035] Y(l,:)=shift(y(N l,int +(0:P int )),N l,dec (5)

[0036] Shift indicates shifting the sampled sequence to the right by N on the time axis. l,dec The sampling time is 1; Y(l,:) represents the sampled data of the l-th slow-time observation. The shift operation can be performed using the Fast Fourier Transform (FFT).

[0037]

[0038] Where W = exp(-j2π / P) int Using a mapping, Y = M 1D→2D (y,P) represents the process of forming a 2D data matrix from 1-D data given a known sample P (pulse repetition interval multiplied by sampling rate);

[0039] A rough estimate of the sample PRI is made by approximating the periodicity of the amplitude of the 1-D raw data; the amplitude signals at different observation intervals are similar, therefore the entire 1-D signal can be approximated as a periodic signal with a period of P (pulse repetition interval multiplied by sampling rate); this property can be expressed as

[0040] |y(nf s )|≈|y((n+P)f s (7) where n is the sampling point number, f s The sampling rate is denoted by . This periodicity can be shown by comparing a slice of data with the remaining data, with expected periodic peaks.

[0041] Due to periodicity, the spectrum of the 1-D amplitude will have a peak at 1 / P on the normalized frequency axis [0,1]; therefore, we can find the non-zero frequency peak position f of the FFT spectrum. peak and its reciprocal as the PRI estimate However, such an estimate is rough and requires further, more precise estimation.

[0042] Fine estimation of sample PRI using PCM; fine estimation using PCM can be expressed as an optimization problem.

[0043]

[0044] Where y sub It is a selected subset of the complete 1-D raw data; Y sub =M 1D→2D (y subP) represents the transformation of 1-D data ysub into a 2-D data matrix Y based on P (pulse repetition interval multiplied by sampling rate). sub ;Y sub,pc =P(Y sub ) indicates taking Y sub The principal component; This is an estimate of P obtained by solving this optimization problem; this problem is equivalent to maximizing the data covariance matrix. The maximum eigenvalue ρ1 is found; this problem is solved by performing a parameter search on P.

[0045] Sample PRI is estimated through a two-step process based on an approximate periodic model of 1-D data and principal component maximization: 1) coarse sample PRI is estimated using the peak of the Fast Fourier Transform (FFT) of the amplitude; 2) fine sample PRI is estimated using PCM. In this way, we can recover the 2-D data matrix from the incomplete 1-D original data vector for subsequent processing steps.

[0046] Step 2: Using the reference point echo and the approximate translation invariance model, establish an approximate matched filtering model for the radar image;

[0047] An image representation model is constructed using approximate translation invariance; an approximate matched filtering model is proposed to describe radar images, which will serve as the basis for the radar image restoration algorithm in subsequent steps.

[0048] Approximate translation invariance of point target echoes; consider a reference point p = [η0, τ0] = [x0 / v, 2R0 / c] in the slow-time-fast-time domain, where η0 is the slow-time coordinate, τ0 is the fast-time coordinate, x0 is the azimuth coordinate, R0 is the range coordinate, v is the platform velocity, and c is the speed of light; consider a point target p+Δp at a distance Δp from the reference point, the coordinates of which are given by p+Δp = [(x0+Δx) / v, 2(R0+ΔR) / c]; Δx and ΔR are the differences between the azimuth and range coordinates of this point from the reference point; the process of converting this point p+Δp into its corresponding 2-D radar echo can be represented as a system function H p+Δp Although this function changes with p+Δp, we assume it is invariant relative to p+Δp; the modeling process will be shown next; using a quadratic approximation of the distance history R(η, p+Δp), we obtain

[0049]

[0050] This equation shows that the distance history of a universal point can be approximated as the distance history of a reference point plus the distance difference ΔR; with this approximation, the echo Y of p+Δp... p+Δp(η,τ) can be represented as a shifted version of the reference echo; the following derivation illustrates this relationship.

[0051]

[0052] Among them, w a (·) is the slow-time window function of the radar signal, s(·) is the transmitted signal, and Y p (η-Δx / v, τ-2ΔR / c) represents the translational versions of the target echo at point P in the slow and fast time domains, respectively, where λ is the radar signal wavelength. By neglecting the exponential phase term containing exp(-j4πΔR / λ), we can consider the point echo response as a translation-invariant response relative to p, i.e.

[0053] Y p+Δp (η,τ)≈Y p (η-Δη,τ-Δτ)(11) Therefore, the mapping from a point scatterer to its echo can be approximately modeled as a translation-invariant system H.

[0054] Image representation of reference echoes based on approximate translation invariance. Radar image focusing can be conceptualized as a process that focuses the echoes of each resolution element to form a complex image. The amplitude of this image represents the intensity of the focused echoes, while the phase contains range information relative to a reference point. The focusing process can be described as follows. For a reference point p, the focused pixel value is given by the inner product.

[0055]

[0056] in(·) * Y(η,τ) represents the conjugate operation, where Y(η,τ) represents all echoes received from the target region, including clutter and noise.

[0057] For any point p + Δp, the amplitude of the focused pixel is determined by... The phase term representing the range difference relative to the reference point is given by exp(-j4πΔR / λ), where λ is the radar signal wavelength. Therefore, the focusing intensity at point p+Δp can be expressed as...

[0058]

[0059] Discretize the observation scene using a grid.

[0060] Δp[n,k]=[nρ x / v,k2ρ R / c] (14)

[0061] Where Δp[n,k] represents the discretized grid points, n and k represent the grid point labels, and ρ x Range resolution, ρ RThis refers to azimuth resolution; radar images can be represented as a 2D focusing response of grid points.

[0062]

[0063] Where I[n,k] represents the focusing response of the grid point, Δ η =ρ x / v is the slow time sampling interval, Δ τ =2ρ R / c is the fast-time sampling interval; since the sampled data is discrete, the above equation can be rewritten in discrete form.

[0064]

[0065] Based on the image representation model described above, we can recover radar images by correlating them with reference echoes. This process can be efficiently implemented using fast convolution.

[0066] Step 3: By dividing the original two-dimensional echo data into multiple blocks and normalizing the energy of each block, the PCM method estimates the reference echo.

[0067] 2D Data Block Segmentation and Normalization. To effectively estimate the reference echo in the presence of clutter, we must consider the time-domain characteristics of clutter. Clutter is the sum of echoes from various background scattering elements, and its intensity is affected by factors such as ground target type, incident angle, and beam pattern. Clutter is typically non-stationary. To handle the non-stationarity of clutter, we segment the observation data and normalize the Frobenius norm of each block. The k-th original data block Y k The normalization process can be expressed as:

[0068]

[0069] Among them, Y k,nm Represents the normalized data, ||Y k || F This represents the Frobenius norm of the matrix. Normalization is used to make the energy of each data block equal, so that the effects of non-stationary clutter between different blocks can be suppressed when estimating the reference echo via PCM.

[0070] The reference echo is recovered by principal component maximization. The reference echo is estimated using PCM, which maximizes principal components over all normalized raw data blocks. This optimization problem can be formulated as follows:

[0071]

[0072] in, It is the estimated reference echo matrix, g k It is matrix Yk,pc Y is the square of the Frobenius norm, where K represents the total number of blocks. k,pc =P(Y k,nm ) represents the k-th normalized raw data block Y k,nm The principal component.

[0073] By observing problem (18), we can find Equivalent to data covariance matrix The maximum eigenvalue; therefore, the physical concept of the optimization problem (18) is to extract the maximum principal component from all normalized blocks.

[0074] If the data mean is removed, this problem is actually solved by estimating the principal components Y of all normalized blocks. k,pc And select the principal component with the largest Frobenius norm as the echo estimate, i.e.

[0075] Step 4: Using the estimated reference echo, the radar image is recovered through matched filtering.

[0076] Using the estimated reference 2-D echo matrix Convolving or matched filtering with the 2D observation data matrix Y generates radar image I, i.e.

[0077]

[0078] Furthermore, the PCM method includes: dividing the original 2-D data into multiple data blocks and performing energy normalization on each data block; estimating the reference point echo by maximizing the principal component energy of all data blocks; and recovering the radar image by using the estimated reference point echo through matched filtering.

[0079] The reference point echo is estimated by maximizing the principal component energy of all data blocks. Specifically, singular value decomposition (SVD) or eigenvalue decomposition is performed on each data block, and the data component with the largest singular value or eigenvalue is selected as the principal component of that data block. From the principal components of all data blocks, the eigenvector with the largest energy norm is selected as the reference point echo.

[0080] This method utilizes the approximate periodicity of the original 1-D data to perform coarse PRI estimation; and uses the PCM method to perform fine PRI estimation.

[0081] Furthermore, the coarse PRI estimation is performed through FFT spectral peak search.

[0082] Furthermore, it also includes: performing matched filtering using the estimated reference point echo.

[0083] Furthermore, the image restoration method can be applied to airborne systems, spaceborne systems, ground systems, or other imaging systems.

[0084] Furthermore, the image restoration method is applied to the design and manufacture of data processing devices.

[0085] Furthermore, the approximate translation invariance model is used in radar imaging hardware and software design and manufacturing.

[0086] Furthermore, the approximate translation invariance model is used in the design and manufacturing of optical imaging hardware and software.

[0087] Example

[0088] The present invention provides a blind focusing imaging method for raw radar echo data, further illustrated by examples and accompanying drawings.

[0089] 1) Convert one-dimensional data into two-dimensional data

[0090] Estimate PRI using the two-step estimation method in step one and transform the one-dimensional data into two dimensions. Figure 1 A flowchart is provided to describe the transformation of one-dimensional data into two-dimensional data. Figure 2 An example of a process for coarsely estimating PRI using PCM is given. Figure 3 An example of the process for accurately estimating PRI is given.

[0091] 2) Data block division and normalization

[0092] The two-dimensional data is divided into blocks, with a block size of 500x500 in this example. Figure 4 A schematic diagram of data partitioning is provided.

[0093] 3) Estimate the reference echo.

[0094] The maximum eigenvalue of each block is estimated using the PCM method. The block corresponding to the maximum eigenvalue is selected, and the principal component of that block is used as the reference echo. Figure 5 An example of the estimated reference echo is given.

[0095] 4) Image generation.

[0096] Matched filtering imaging is performed using the reference echo and the two-dimensional raw data. Figure 6 An example of the imaging results is given.

Claims

1. A blind focusing imaging method using raw radar echo data, characterized in that, Includes the following steps: Step 1: Based on the approximate periodic model of the one-dimensional original data amplitude signal and the principal component maximization method, perform coarse and fine estimation of PRI to restore the one-dimensional data to a two-dimensional data matrix; Step 2: Using the reference point echo and the approximate translation invariance model, establish an approximate matched filtering model for the radar image; Step 3: The reference echo is estimated by dividing the original two-dimensional echo data into multiple blocks and normalizing the energy of each block. Step 4: Using the estimated reference echo, the radar image is recovered through matched filtering.

2. The blind focusing imaging method for raw radar echo data according to claim 1, characterized in that, Step 1 is as follows: Recovering a two-dimensional data matrix from a one-dimensional raw data vector, with an unknown pulse repetition interval; the radar echo data stream initially consists of one-dimensional raw data, where only P = Tf. s Only when this information is known can it be formed into a two-dimensional original data matrix, where T is the pulse repetition interval and f... s The sampling rate is P; when the parameter P is unknown, it must be estimated from the data to reconstruct the original two-dimensional data for subsequent principal component maximization and image reconstruction processing. A parametric model is constructed by converting one-dimensional raw data into two-dimensional raw data. When PRI is known, the two-dimensional data model is constructed as follows: the sample PRI is a parameter of the model, and the estimation of PRI is considered as a model parameter inversion problem. The relationship between the time axis t' of the one-dimensional observed signal and the time axis of its corresponding two-dimensional radar observed signal is as follows: t′=t+τ=lT+τ (1) The slow time of radar is expressed as t = lT, where l represents the l-th pulse observation interval; τ is the fast time of synthetic aperture radar; the start time of the l-th pulse observation interval is denoted as lT; the two-dimensional signal is expressed using a one-dimensional signal as follows: Y(l,τ)=y(t') (2) Where Y(l,τ) is the radar two-dimensional observation signal; the above model uses a continuous signal, but the sampled data is discrete; therefore, a discrete two-dimensional data formation model needs to be constructed; the sample PRI is represented in the discrete domain as P=Tf s =P int +P dec (3) Where P int It is the integer part, P dec It is the decimal part; the starting point of the l-th observation interval is denoted as Align the starting points in the two-dimensional data matrix to maintain coherent information between different observation intervals; The calibration of the l-th echo is performed by sampling in the discrete domain at fractional sampling intervals N. l,dec This is achieved by moving data slices; this process is represented as... Y(l,:)=shift(y(N l,int +(0:P int )),N l,dec ) (5) Shift indicates shifting the sampled sequence to the right by N on the time axis. l,dec The sampling time is 1; Y(l,:) represents the l-th slow-time observation sample data; the shift operation can be performed using the Fast Fourier Transform (FFT). Where W = exp(-j2π / P) int Using a mapping, Y = M 1D→2D (y,P) represents the process of forming a two-dimensional data matrix from one-dimensional data given a known sample P. A rough estimate of the sample PRI is made by approximating the periodicity of the amplitude of the one-dimensional raw data; the amplitude signals at different observation intervals are similar, therefore the entire 1-D signal can be approximated as a periodic signal with period P; this property is expressed as |y(nf s )|≈|y((n+P)f s )|(7) Where n is the sampling point number, f s Sampling rate; Due to periodicity, the spectrum of the one-dimensional amplitude will have a peak at 1 / P on the normalized frequency axis [0,1]; therefore, by finding the non-zero frequency peak position f of the FFT spectrum... peak and its reciprocal as the PRI estimate However, such an estimate is rough and requires further, more precise estimation. A fine-grained estimate of the sample PRI is performed using Principal Component Maximization (PCM); this fine-grained estimate is expressed as an optimization problem. Where y sub It is a selected subset of the complete 1-D raw data; Y sub =M 1D→2D (y sub ,P) means that the one-dimensional data y is divided according to P. sub Transformed to a two-dimensional data matrix Y sub ;Y sub,pc =P(Y sub ) indicates taking Y sub Principal components; This is an estimate of P obtained by solving this optimization problem; this problem is equivalent to maximizing the data covariance matrix. The maximum eigenvalue ρ1; this problem is solved by performing a parameter search on P; Sample PRI is estimated using a two-step process based on an approximate periodic model of 1-D data and principal component maximization: 1) coarse sample PRI is estimated using the fast Fourier transform peak of the amplitude; 2) fine sample PRI is estimated using principal component maximization.

3. The blind focusing imaging method for raw radar echo data according to claim 2, characterized in that, In step 2, an image representation model is constructed using approximate translation invariance, and an approximate matched filtering model is proposed to describe radar images; Consider a reference point p = [η0, τ0] = [x0 / v, 2R0 / c] in the slow-time-fast-time domain, where η0 is the slow-time coordinate, τ0 is the fast-time coordinate, x0 is the azimuth coordinate, R0 is the range coordinate, v is the platform velocity, and c is the speed of light. Consider a point target p+Δp at a distance Δp from the reference point. The coordinates of this point are given by p+Δp = [(x0+Δx) / v, 2(R0+ΔR) / c], where Δx and ΔR are the differences between the azimuth and range coordinates of this point from the reference point. The process of converting this point p+Δp into its corresponding 2D radar echo can be represented as a system function H. p+Δp Assuming this function is invariant with respect to p+Δp, the modeling process will be shown below; using a quadratic approximation of the distance history R(η,p+Δp), we obtain... Formula (9) shows that the distance history of the universal point can be approximated as the distance history of the reference point plus the distance difference ΔR; the echo Y of p+Δp p+Δp (η,τ) represents the shifted version of the reference echo; Among them, w a (·) is the slow-time window function of the radar signal, s(·) is the transmitted signal, and Y p (η-Δx / v, τ-2ΔR / c) represents the translational versions of the target echo at point P in the slow and fast time domains, respectively, where λ is the radar signal wavelength. By neglecting the exponential phase term containing exp(-j4πΔR / λ), the point echo response is considered as a translationally invariant response relative to p. Y p+Δp (η,τ)≈Y p (η-δη,τ-δτ) (11) Therefore, the mapping from a point scatterer to its echo is approximately modeled as a translation-invariant system H; Image representation of reference echoes based on approximate translation invariance; radar image focusing is conceptualized as a process that focuses the echoes of each resolution element to form a complex image; the amplitude of this image represents the intensity of the focused echoes, while the phase contains range information relative to the reference point; the focusing process is described as follows: for reference point p, the focused pixel value is given by the inner product. in(·) * Y(η,τ) represents the conjugate operation, where Y(η,τ) represents all echoes received from the target region, including clutter and noise. For any point p + Δp, the amplitude of the focused pixel is determined by... The phase term, representing the range difference relative to the reference point, is given by exp(-j4πΔR / λ); therefore, the focused intensity at point p+Δp can be expressed as... Discretize the observation scene using a grid. Δp[n,k]=[nρ x / v,k2ρ R / c] (14) Where Δp[n,k] represents the discretized grid points, n and k represent the grid point labels, and ρ x Range resolution, ρ R This refers to the azimuth resolution; radar images are represented as a 2D focusing response of grid points. Where I[n,k] represents the focusing response of the grid point, Δ η =ρ x / v is the slow time sampling interval, Δ τ =2ρ R / c is the fast time sampling interval; since the sampled data is discrete, formula (15) is rewritten in discrete form. Based on the image representation model described above, the radar image is recovered by correlating it with the reference echo.

4. The blind focusing imaging method for raw radar echo data according to claim 3, characterized in that, Step 3 is as follows: 2D data block partitioning and normalization; dividing the observed data into blocks and normalizing the Frobenius norm of each block; the k-th original data block Y k The normalization process is expressed as Among them, Y k,nm Represents the normalized data, ||Y k || F This represents the Frobenius norm of the matrix; The reference echo is recovered by principal component maximization; the reference echo is estimated by principal component maximization, which maximizes the principal components over all normalized raw data blocks; this optimization problem is formulated as follows: in, It is the estimated reference echo matrix, g k It is matrix Y k,pc The square of the Frobenius norm, where K represents the total number of blocks; Y k,pc =P(Y k,nm ) represents the k-th normalized raw data block Y k,nm Principal components; By observing problem (18), one can discover Equivalent to data covariance matrix The maximum eigenvalue; therefore, the physical concept of the optimization problem (18) is to extract the maximum principal component from all normalized blocks; If the data mean is removed, this problem is actually solved by estimating the principal components Y of all normalized blocks. k,pc And select the principal component with the largest Frobenius norm as the echo estimate, i.e.

5. The blind focusing imaging method for raw radar echo data according to claim 4, characterized in that, Step 4 is as follows: Using the estimated reference 2-D echo matrix Convolving or matched filtering with the 2D observation data matrix Y generates radar image I, i.e. .

6. The blind focusing imaging method for raw radar echo data according to claim 5, characterized in that, The principal component maximization method includes: dividing the original 2D data into multiple data blocks and normalizing the energy of each data block; estimating the reference point echo by maximizing the principal component energy of all data blocks; and recovering the radar image by using the estimated reference point echo through matched filtering.

7. The blind focusing imaging method for raw radar echo data according to claim 6, characterized in that, The reference point echo is estimated by maximizing the principal component energy of all data blocks. Specifically, singular value decomposition or eigenvalue decomposition is performed on each data block, and the data component with the largest singular value or eigenvalue is selected as the principal component of that data block. From the principal components of all data blocks, the eigenvector with the largest energy norm is selected as the reference point echo.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program performs the steps of the method described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-7.

Images