A method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction

Abstract

A method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction includes the following steps: preprocessing a two-dimensional ISAR image by zeroing out areas surrounding a target scattering point that do not contain a target; performing eigenvalue decomposition on the autocorrelation matrices of different range cells and azimuth cells, and pre-reconstructing a one-dimensional range image and a one-dimensional azimuth image using the maximum eigenvalue obtained for each cell; then, zeroing out corresponding image cells whose maximum eigenvalue is less than a certain threshold, thereby obtaining a two-dimensional ISAR image with some noise areas zeroed out; setting a threshold, and considering range and azimuth cells above the threshold to contain target scattering points; identifying points in the image with amplitudes above the threshold as target scattering points, and extracting scattering points for range and azimuth respectively; and determining scattering points by combining range and azimuth to eliminate false scattering points. The present invention is characterized by efficient extraction of weak target scattering points, and can achieve the goal of effectively reducing noise while fully preserving weak scattering points.

Description

Technical Field

[0001] The present invention belongs to the field of radar technology, and in particular relates to an InISAR imaging scattering center extraction method based on range-azimuth image reconstruction. Background Art

[0002] Inverse Synthetic Aperture Radar (ISAR), with its all-day, all-weather capabilities, plays a vital role in non-cooperative target recognition. However, two-dimensional ISAR images are merely two-dimensional projections of targets on the imaging plane and cannot reveal their altitude. In contrast, interferometric ISAR (InISAR) can obtain target altitude information through interferometry between two antennas, significantly improving target recognition capabilities.

[0003] The extraction of scattering centers is a key step in InISAR three-dimensional imaging. Currently, the traditional method for extracting scattering centers is the CLEAN algorithm and its extensions, which are widely used in radar imaging. However, the traditional CLEAN algorithm only compares the amplitude of each pixel of the ISAR image with an empirical threshold, which is easily affected by background noise and sidelobes, especially some weak scattering points are difficult to extract. In response to the shortcomings of the traditional CLEAN algorithm, an improved CLEAN algorithm was proposed, which uses the one-dimensional high-resolution range profile (HRRP) of the target to determine the peak of the backscattering area. The results show that the improved CLEAN algorithm can extract more weak scattering points, but it also extracts some sidelobes and noise with higher amplitudes, resulting in more false scattering points in the three-dimensional reconstruction results. Summary of the Invention

[0004] In order to overcome the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide an InISAR imaging scattering center extraction method based on range-azimuth image reconstruction, which has the characteristics of efficiently extracting target weak scattering points and can achieve the purpose of effectively reducing noise while fully retaining weak scattering points.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction includes the following steps:

[0007] Step 1, obtaining a two-dimensional ISAR image, and preprocessing the two-dimensional ISAR image by setting the target scattering point and the area without the target to zero, thereby obtaining an ISAR image after zero preprocessing;

[0008] Step 2: Based on the ISAR image after zeroing preprocessing, perform eigenvalue decomposition on the autocorrelation matrices of different range units and azimuth units, and pre-reconstruct one-dimensional range images and one-dimensional azimuth images respectively using the maximum eigenvalue obtained for each unit. Then, zero the corresponding image units whose maximum eigenvalue is less than a certain threshold, thereby obtaining a two-dimensional ISAR image with some noise areas zeroed.

[0009] Step 3: Based on the two-dimensional image matrix after the noise area is zeroed, repeat step 2 to reconstruct the one-dimensional range image and the one-dimensional azimuth image, set a threshold, and consider the range and azimuth cells above the threshold to contain target scattering points, and define them as the main range cell and the main azimuth cell, respectively;

[0010] Step 4: Based on the main range and main azimuth units, different range unit extraction thresholds and different azimuth unit extraction thresholds are determined by threshold updating, and points in the image with amplitudes higher than the thresholds are considered to be target scattering points. Scattering points are extracted for range and azimuth respectively.

[0011] Step 5: Based on the scattering points extracted in the range and azimuth directions in step 4, the scattering points are determined by combining the range and azimuth directions and false scattering points are eliminated.

[0012] The step 1 is to obtain a two-dimensional ISAR image by using an imaging algorithm on the ISAR echo data. The step 2 is specifically as follows:

[0013] (2a) Assume that the dimension of an ISAR image A is M×N, where M and N represent the number of azimuth units and the number of range units, respectively; let X n is the nth column vector of A, then the autocorrelation matrix of the nth range cell is expressed as:

[0014]

[0015] R n After eigenvalue decomposition, the maximum eigenvalue λ of each distance unit is obtained n (n=1,2,3...N); then use the maximum eigenvalue λ of each distance unit n Pre-reconstruct a one-dimensional range image; define the range threshold as q1, and record the position of any range unit whose peak value in the reconstructed one-dimensional range image is lower than q1;

[0016] (2b) Pre-reconstruct one-dimensional azimuth image; let Y m is the mth row vector of A, then the autocorrelation matrix of the mth orientation unit is expressed as:

[0017]

[0018] To G mAfter eigenvalue decomposition, the maximum eigenvalue μ of each orientation unit is obtained m (m=1,2,3...M), and then use the maximum eigenvalue μ of each orientation unit m Pre-reconstruct the one-dimensional azimuth image, define the azimuth threshold as q2, and record the position of any azimuth unit with a peak value lower than q2 in the reconstructed one-dimensional azimuth image.

[0019] In step (2a), the range cells at the same position in the two-dimensional ISAR image of step 1 are set to zero, while the range cells above q1 remain unchanged;

[0020] In the step (2b), the azimuth units at the same position in the two-dimensional ISAR image of step 1 are set to zero, while the azimuth units higher than q2 remain unchanged, and finally the image matrix B after the zeroing process is obtained.

[0021] The step 3 is specifically as follows:

[0022] (3a) Let X' n is the nth column vector of B, then the autocorrelation matrix of the nth range cell is expressed as:

[0023]

[0024] R' n After eigenvalue decomposition, the maximum eigenvalue λ of each distance unit is used n '(n=1,2,3...N) reconstruct one-dimensional range image;

[0025] (3b) Let Y' m is the mth row vector of B, then the autocorrelation matrix of the mth orientation unit is expressed as:

[0026]

[0027] To G' m After eigenvalue decomposition, the maximum eigenvalue μ' of each orientation unit is used m (m=1,2,3...M) reconstruct a one-dimensional azimuth image.

[0028] In the step (3a), the range cells with peak values ​​greater than q1 are marked as the main range cells containing the target scattering points.

[0029] In the step (3b), the azimuth unit with a peak value greater than q2 is marked as the main azimuth unit containing the target scattering point.

[0030] The step 4 is specifically as follows:

[0031] (4a) Let the number of main distance units be N main , find the maximum amplitude δ of each main distance unit i(i=1,2,3...N main ), different coefficients are set according to the different maximum eigenvalues ​​after eigenvalue decomposition of the autocorrelation matrix of each main distance unit:

[0032]

[0033] Where φ1 and φ2 are constants, λ i ' is the maximum eigenvalue corresponding to the i-th principal distance unit, k r and b r max(λ i ') and min(λ i '); c i Multiply by the maximum amplitude δ i Get the threshold value of the corresponding distance unit:

[0034]

[0035] For the i-th main distance unit, it is preliminarily determined that the amplitude is greater than θ i The elements of are the scattering point positions;

[0036] (4b) In view of the fact that some false scattering points in (4a) are extracted as scattering points, let the number of main azimuth units be M main , the jth main orientation unit threshold is:

[0037]

[0038] where χ' j is the maximum amplitude of the jth main azimuth unit, u' j is the maximum eigenvalue of the jth main orientation unit, k a and b a max(μ' j ) and min(μ' j ).

[0039] In step 5, the target scattering points are extracted along the range direction and the azimuth direction respectively, and the final scattering point extraction result is determined by combining the extraction results of the range direction and the azimuth direction.

[0040] Beneficial effects of the present invention:

[0041] The present invention utilizes a pre-reconstructed one-dimensional image and performs a zeroing operation on some noise areas, thereby reducing the overall noise energy, increasing the distinction between subsequent scattering point units and noise units, and having better noise suppression capability.

[0042] The present invention reconstructs a one-dimensional range image and a one-dimensional azimuth image from two dimensions, so that when finding the main unit where the scattering point is located, it can better suppress the erroneous extraction of noise units, and is more robust than the traditional method of only reconstructing the one-dimensional range image to suppress noise.

[0043] The present invention proposes a threshold update mechanism to extract the positions of scattering points for the main range unit and the main azimuth unit respectively. The relationship between the maximum eigenvalue after the eigenvalue decomposition of the autocorrelation matrix of different main units and the threshold coefficient is used to set a threshold for each main unit, thereby achieving accurate extraction of scattering points.

[0044] The present invention integrates the scattering point extraction results in the range and azimuth directions. Only when a scattering point is extracted at a certain position in the image in both the range and azimuth directions will this scattering point be used as the final extraction result. Otherwise, it is considered that there is no target scattering point at that position. This greatly reduces the probability of erroneous extraction of scattering points. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart for implementing the present invention.

[0046] Figure 2 is the original two-dimensional ISAR simulation image.

[0047] Figure 3 Schematic diagram of the ship target model for the simulation data experiment in the present invention.

[0048] Figure 4 Schematic diagram of a one-dimensional range image reconstructed without setting some noise units to zero and a one-dimensional range image reconstructed after setting some noise units to zero in a simulation data experiment in the present invention.

[0049] Figure 5 Schematic diagram of scattering point extraction results of the An_Improved_CLEAN algorithm and the scattering center extraction algorithm based on range-azimuth image reconstruction in the simulation data experiment of the present invention.

[0050] Figure 6 Schematic diagram of the three-dimensional reconstruction results of different algorithms in the simulation data experiment in the present invention.

[0051] Figure 7 This is the ISAR imaging result of the Boeing 737 aircraft.

[0052] Figure 8 The scattering point extraction results of the ISAR measured data imaging results of the Boeing 737 aircraft in the present invention are respectively obtained by using the An_Improved_CLEAN algorithm and the scattering center extraction algorithm based on range-azimuth image reconstruction. DETAILED DESCRIPTION

[0053] The present invention will be described in further detail below with reference to the accompanying drawings.

[0054] Reference Figures 1-8 The specific implementation steps of this example are as follows:

[0055] Step 1: Perform zero preprocessing on the acquired two-dimensional ISAR image.

[0056] The specific implementation of this step is as follows:

[0057] Since most areas in ISAR images are devoid of target scattering points, these noise regions are distributed around the scattering points, hindering the subsequent determination of the primary range and azimuth cells containing scattering points. Before extracting scattering points, the noise regions surrounding the target in the ISAR image are preprocessed by zeroing them. This preprocessing is achieved by zeroing out the regions outside the azimuth and range ranges of the target scattering points. This preprocessing preserves the rectangular region containing the target, reduces the overall number of noise cells in the ISAR image, and facilitates the subsequent distinction between range and azimuth cells containing and excluding scattering points.

[0058] Step 2: calculate the autocorrelation matrix of different distance units in the ISAR image after zero preprocessing, and obtain the maximum eigenvalue of different distance units through eigenvalue decomposition, pre-reconstruct the one-dimensional range image, and then pre-reconstruct the one-dimensional azimuth image along the azimuth direction through similar steps, and then pre-process some noise units by zeroing through threshold.

[0059] The specific implementation of this step is as follows:

[0060] (2a) Assume that the dimension of an ISAR image A is M×N, where M and N represent the number of azimuth units and range units, respectively. Let X n is the nth column vector of A, then the autocorrelation matrix of the nth range cell can be expressed as:

[0061]

[0062] R n After eigenvalue decomposition, the maximum eigenvalue λ of each distance unit is obtained n (n=1,2,3...N). Then use the maximum eigenvalue λ of each distance unit n Pre-reconstruct the one-dimensional range image. Define the range threshold as q1. Record the location of any range bin in the reconstructed one-dimensional range image whose peak is below q1. Set the range bins at the same location in the two-dimensional ISAR image from step 1 to zero, while leaving the range bins above q1 unchanged. The threshold q1 is typically set to 0.01–0.05 multiplied by the maximum value of the maximum eigenvalue corresponding to each range bin.

[0063] (2b) Based on the same principle, pre-reconstruct the one-dimensional azimuth image. Let Y m is the mth row vector of A, then the autocorrelation matrix of the mth orientation unit is expressed as:

[0064]

[0065] To G m After eigenvalue decomposition, the maximum eigenvalue μ of each orientation unit is obtained m (m=1,2,3...M). Then use the maximum eigenvalue μ of each orientation unit m Pre-reconstruct the one-dimensional azimuth image. Define the azimuth threshold as q2. Record the location of any azimuth cell in the reconstructed one-dimensional azimuth image whose peak is lower than q2. Set the azimuth cells at the same location in the two-dimensional ISAR image from step 1 to zero, while leaving the azimuth cells higher than q2 unchanged. Finally, obtain the zeroed image matrix B. The threshold q2 is usually set to 0.01-0.05 multiplied by the maximum value of the maximum eigenvalue corresponding to each azimuth cell.

[0066] Step 3: Based on the image matrix B after zeroing constructed in step 2, reconstruct the one-dimensional range image and the one-dimensional azimuth image on the basis of the image matrix B, and find the main range and main azimuth cells containing the target scattering points.

[0067] The specific implementation of this step is as follows:

[0068] (3a) Similar to the process in step 2, let X' n is the nth column vector of B, then the autocorrelation matrix of the nth range cell is expressed as:

[0069]

[0070] R' n After eigenvalue decomposition, the maximum eigenvalue λ of each distance unit is used n '(n=1,2,3...N) reconstruct the one-dimensional range image and mark the range cells with peak values ​​greater than q1 as the main range cells containing the target scattering points.

[0071] (3b) Let Y' m is the mth row vector of B, then the autocorrelation matrix of the mth orientation unit is expressed as:

[0072]

[0073] To G' m After eigenvalue decomposition, the maximum eigenvalue μ' of each orientation unit is used m (m=1,2,3...M) reconstruct a one-dimensional azimuth image and mark the azimuth unit with a peak value greater than q2 as the main azimuth unit containing the target scattering point.

[0074] Step 4: Based on the main distance and main azimuth units containing scattering points in step 3, extract scattering points from each main unit in turn.

[0075] This step is specifically implemented as follows:

[0076] (4a) Let the number of main distance units be N main , find the maximum amplitude δ of each main distance unit i (i=1,2,3...N main ), a threshold update mechanism is proposed, which sets different coefficients according to the different maximum eigenvalues ​​after eigenvalue decomposition of the autocorrelation matrix of each main distance unit:

[0077]

[0078] Where φ1 and φ2 are constants, λ i ' is the maximum eigenvalue corresponding to the i-th principal distance unit, k r and b r max(λ i ') and min(λ i '). i Multiply by the maximum amplitude δ i Get the threshold value of the corresponding distance unit:

[0079]

[0080] For the i-th main distance unit, it can be preliminarily determined that the amplitude is greater than θ i The elements of are the scattering point positions.

[0081] (4b) In order to extract some false scattering points as scattering points in (4a), similar operations need to be performed on each main azimuth unit. Let the number of main azimuth units be M main , the jth main orientation unit threshold is:

[0082]

[0083] where χ' j is the maximum amplitude of the jth main azimuth unit, u' j is the maximum eigenvalue of the jth main orientation unit, k a and b a max(μ' j ) and min(μ' j ).

[0084] In step 5, based on the target scatter points extracted along the range and azimuth directions in step 4, the final scatter point extraction results are determined by combining the range and azimuth extraction results. Only scatter points extracted along the range or azimuth directions are considered false scatter points. Only scatter points extracted along both the range and azimuth directions at a given unit location are retained as the final extracted scatter points, thus eliminating false scatter points.

[0085] like Figure 1 As shown in the flowchart of the invention, the purpose of pre-reconstructing the one-dimensional range image and the one-dimensional azimuth image is to remove some noise units with lower amplitude, which is conducive to finding the main range and main azimuth units containing target scattering points after reconstructing the one-dimensional range image and the one-dimensional azimuth image.

[0086] like Figure 2 As shown: This figure is an ISAR image generated by simulating the echoes of 64 point targets using the RD algorithm.

[0087] like Figure 4 As shown: The left picture is the one-dimensional range image obtained without zero preprocessing, and the right picture is the one-dimensional range image obtained after zero preprocessing. It can be found that the amplitude difference between the noise unit and the main range unit containing the target scattering point in the right picture is larger, which makes it easier to extract the main range unit and more flexible when setting the threshold.

[0088] like Figure 5 The left and right figures show the target scattering point extraction results obtained by the An_Improved_CLEAN algorithm and the algorithm of the present invention, respectively. It can be clearly seen that the left figure misses scattering points in region 1 and extracts too many scattering points in region 2, while the algorithm of the present invention performs well in regions 1 and 2.

[0089] like Figure 6 The top and bottom figures show the 3D reconstruction results after scatter point extraction using the An_Improved_CLEAN algorithm and the algorithm of our invention, respectively. It can be seen that many scatter points are not reconstructed in the top figure, and multiple false scatter points are reconstructed at some points, while the 3D reconstruction of the scatter points in the bottom figure is good.

[0090] like Figure 7 As shown in the figure: This figure is the imaging result obtained by using the RD algorithm to obtain the ISAR measured data of the Boeing 737 aircraft.

[0091] like Figure 8 Figures (a) and (b) show the target scattering point extraction results of the measured data obtained by the An_Improved_CLEAN algorithm and the algorithm of the present invention, respectively. In the corresponding parts circled in red in the two figures, more scattering points are recovered in Figure (b).

Claims

1. A method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction, characterized in that: The following steps are included: Step 1, obtaining a two-dimensional ISAR image, and preprocessing the two-dimensional ISAR image by setting the target scattering point and the area without the target to zero, thereby obtaining an ISAR image after zero preprocessing; Step 2: Based on the ISAR image after zeroing preprocessing, perform eigenvalue decomposition on the autocorrelation matrices of different range units and azimuth units, and pre-reconstruct one-dimensional range images and one-dimensional azimuth images respectively using the maximum eigenvalue obtained for each unit. Then, zero the corresponding image units whose maximum eigenvalue is less than a certain threshold, thereby obtaining a two-dimensional ISAR image with some noise areas zeroed. Step 3: Based on the two-dimensional image matrix after the noise area is zeroed, repeat step 2 to reconstruct the one-dimensional range image and the one-dimensional azimuth image, set a threshold, and consider the range and azimuth cells above the threshold to contain target scattering points, and define them as the main range cell and the main azimuth cell, respectively; Step 4: Based on the main range and main azimuth units, different range unit extraction thresholds and different azimuth unit extraction thresholds are determined by threshold updating, and points in the image with amplitudes higher than the thresholds are considered to be target scattering points. Scattering points are extracted for range and azimuth respectively. Step 5: Based on the scattering points extracted in the range and azimuth directions in step 4, the scattering points are determined by combining the range and azimuth directions and false scattering points are eliminated.

2. The method for extracting scattering centers from InISAR images based on range-azimuth image reconstruction according to claim 1, wherein: In the step 1, an imaging algorithm is used to obtain a two-dimensional ISAR image on the ISAR echo data.

3. The method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction according to claim 1, wherein: The step 2 is specifically as follows: (2a) Assume that the dimension of an ISAR image A is M×N, where M and N represent the number of azimuth units and the number of range units, respectively; let X n is the nth column vector of A, then the autocorrelation matrix of the nth range cell is expressed as: R n After eigenvalue decomposition, the maximum eigenvalue λ of each distance unit is obtained n , n=1,2,3...N; then use the maximum eigenvalue λ of each distance unit n Pre-reconstruct a one-dimensional range image; define the range threshold as q1, and record the position of any range unit whose peak value in the reconstructed one-dimensional range image is lower than q1; (2b) Pre-reconstruct one-dimensional azimuth image; let Y m is the mth row vector of A, then the autocorrelation matrix of the mth orientation unit is expressed as: To G m After eigenvalue decomposition, the maximum eigenvalue μ of each orientation unit is obtained m , m=1,2,3...M, and then use the maximum eigenvalue μ of each orientation unit m Pre-reconstruct the one-dimensional azimuth image, define the azimuth threshold as q2, and record the position of any azimuth unit with a peak value lower than q2 in the reconstructed one-dimensional azimuth image.

4. The method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction according to claim 3, wherein: In the step (2a), the distance cells at the same position in the two-dimensional ISAR image in step 1 are set to zero, while the distance cells higher than q1 remain unchanged.

5. The method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction according to claim 3, wherein: In the step (2b), the azimuth units at the same position in the two-dimensional ISAR image of step 1 are set to zero, while the azimuth units higher than q2 remain unchanged, and finally the image matrix B after the zeroing process is obtained.

6. The method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction according to claim 5, characterized in that: The step 3 is specifically as follows: (3a) Let X' n is the nth column vector of B, then the autocorrelation matrix of the nth range cell is expressed as: R' n After eigenvalue decomposition, the maximum eigenvalue λ′ of each distance unit is used n (n=1,2,3...N) reconstruct one-dimensional range image; (3b) Let Y' m is the mth row vector of B, then the autocorrelation matrix of the mth orientation unit is expressed as: To G' m After eigenvalue decomposition, the maximum eigenvalue μ′ of each orientation unit is used m , m=1,2,3...M, reconstruct a one-dimensional azimuth image.

7. The method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction according to claim 6, wherein: In the step (3a), the range cells with peak values ​​greater than q1 are marked as the main range cells containing the target scattering points.

8. The method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction according to claim 6, wherein: In the step (3b), the azimuth unit with a peak value greater than q2 is marked as the main azimuth unit containing the target scattering point.

9. The method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction according to claim 1, wherein: The step 4 is specifically as follows: (4a) Let the number of main distance units be N main , find the maximum amplitude of each main distance unit δ, i = 1, 2, 3...N main , according to the different maximum eigenvalues ​​after eigenvalue decomposition of the autocorrelation matrix of each main distance unit, different coefficients are set: where φ1 and φ2 are constants, λ′ i is the maximum eigenvalue corresponding to the i-th principal distance unit, k r and b r max(λ′ i ) and min(λ′ i ); c i Multiply by the maximum amplitude δ i Get the threshold value of the corresponding distance unit: For the i-th main distance unit, it is preliminarily determined that the amplitude is greater than θ i The elements of are the scattering point positions; (4b) In view of the fact that some false scattering points in (4a) are extracted as scattering points, let the number of main azimuth units be M main , the jth main orientation unit threshold is: where χ' j is the maximum amplitude of the jth main azimuth unit, u' j is the maximum eigenvalue of the jth main orientation unit, k a and b a max(μ' j ) and min(μ' j ).

10. The method for extracting scattering centers from InISAR imaging based on range-azimuth image reconstruction according to claim 1, characterized in that: In step 5, the target scattering points are extracted along the range direction and the azimuth direction respectively, and the final scattering point extraction result is determined by combining the extraction results of the range direction and the azimuth direction.

Images