A Multi-channel SAR Moving Target Detection Method Based on Fusion of Multi-polarization Images

Abstract

Classic multi-channel SAR moving target detection methods such as DPCA and STAP all use interference phase for clutter suppression and moving target detection. However, in non-uniform scenes, the performance of the STAP method is limited due to the small number of samples and inaccurate covariance matrix estimation; the DPCA method has limited suppression ability for strong scatterers in non-uniform scenes, such as buildings, because of the inevitable existence of noise, its interference phase is not strictly zero, and a large number of false alarms will be caused after DPCA processing. Therefore, the moving target detection performance of traditional multi-channel SAR moving target detection methods in non-uniform scenes needs to be improved. In view of the problem that traditional multi-channel moving target detection methods have many false alarms of buildings in complex terrain scenes, the present invention proposes a moving target detection method that fuses multi-polarization images, and replaces and fuses different types of objects of cross-polarization images and co-polarization images to reduce scene non-uniformity and improve moving target detection performance.

Description

Technical Field

[0001] The invention belongs to the field of radar detection, and in particular relates to a multi-channel SAR moving target detection method integrating multi-polarization images. Background Art

[0002] Synthetic Aperture Radar Ground Moving Target Detection (SAR-GMTI) can detect moving targets in the image based on the SAR image and estimate its position. Multi-channel SAR-GMTI deployed along the track is currently the most widely used SAR-GMTI mode. Compared with the traditional single-channel system, the multi-channel SAR-GMTI system has more system degrees of freedom. Generally, the moving target is detected according to the interference phase generated by the target movement after clutter suppression processing. Multi-channel SAR-GMTI methods mainly include Along Track Interferometry (ATI), Displaced Phase Center Antenna (DPCA) and Space-time Adaptive Processing (STAP). Among them, ATI technology uses the interferometric phase between channels to detect and estimate the speed of moving targets, but since no clutter suppression is performed before detection, the detection effect is greatly affected by ground clutter; STAP method is a two-dimensional joint adaptive processing technology in spatial and temporal domains, which can effectively suppress clutter interference. When the clutter is non-uniform and the signal-to-clutter ratio is low, many scholars have proposed many improved algorithms, but the implementation structure of STAP method is complex and the signal processing calculation is very large; DPCA technology can complete clutter suppression and detect moving targets by performing simple subtraction operations under the condition that the channel spacing, carrier speed and pulse repetition frequency meet a certain relationship. Due to its simple structure and easy implementation in engineering, it is widely used in practical systems. Polarization can describe the propagation trajectory of electromagnetic waves in space. In addition to amplitude, phase and Doppler frequency, the target polarization characteristics are another key information of SAR echoes. For example, polarization information can be used to classify the types of objects in SAR images, and then determine the precise location of each type of object in the image. SAR-GMTI combined with polarization information will be a major trend in the future development of SAR. Classic multi-channel SAR moving target detection methods such as DPCA and STAP all use interference phase to suppress clutter and detect moving targets. However, in non-uniform scenes, the performance of the STAP method is limited due to the small number of samples and inaccurate covariance matrix estimation; the DPCA method has limited suppression capabilities for strong scatterers in non-uniform scenes, such as buildings. Due to the inevitable presence of noise, the interference phase is not strictly zero, and a large number of false alarms will be caused after DPCA processing. Therefore, the moving target detection performance of traditional multi-channel SAR moving target detection methods in non-uniform scenes needs to be improved. Summary of the invention

[0003] In order to overcome the defects in the prior art, the present invention aims at the problem that the traditional multi-channel moving target detection method has many false alarms of buildings in complex terrain scenes, and proposes a moving target detection method that fuses multi-polarization images. By replacing and fusing different types of objects in cross-polarization images and co-polarization images, the scene non-uniformity is reduced and the moving target detection performance is improved. Specifically, the following steps are included:

[0004] Step 1: Perform SAR imaging processing on the full polarization data of the two channels respectively to obtain the co-polarization complex image s1 and the cross-polarization complex image s2 of the reference channel and the co-polarization complex image s3 and the cross-polarization complex image s4 of the other channel;

[0005] Step 2: perform image registration and channel equalization. For image registration, the cross-correlation method is used to obtain the offset of different channel images in the azimuth direction, and then the data of the other channel is interpolated and translated so that the pixels at the same position on the two channel images correspond to the same target on the ground. For channel equalization, a two-dimensional frequency domain adaptive channel equalization method is used to calculate the equalization weight coefficients in the range and azimuth directions respectively, and the image of the other channel is multiplied by the equalization weight coefficients in the range and azimuth directions respectively, and the amplitude and phase errors of each pixel are corrected to obtain the co-polarization complex image s3' and the cross-polarization complex image s4' after processing of the other channel.

[0006] Step 3: Use the full polarization data of the reference channel to perform polarization classification, and use the H / α-Wishart iterative classification method to identify buildings and determine their specific locations in the image;

[0007] Step 4: Using the co-polarization image as the reference image, the pixels at the same position of the co-polarization image are replaced with the pixels where the buildings are located in the two-channel cross-polarization images to obtain the image s after the two-channel data is replaced. 12 and 34 ';

[0008] Step 5: Replace the two channels with the image s 12 、s 34 'Subtraction completes the image domain DPCA processing; Step six, perform CFAR detection, use the CA-CFAR method to design a hollow sliding window, calculate the statistical characteristics of the clutter in the window, calculate the threshold based on the established random distribution assumption and threshold setting principle, slide the window to traverse the image, determine the threshold based on the local sample adaptively, and determine whether the moving target exists based on the set threshold to obtain the final moving target detection result.

[0009] The beneficial effects of the present invention are:

[0010] Compared with the prior art, the present invention has the following significant benefits:

[0011] (1) Based on the classic image domain DPCA method, the method of the present invention fuses multi-polarization images. In complex environments, it can reduce scene non-uniformity, thereby reducing false alarms such as man-made buildings in complex scenes and improving the moving target detection performance.

[0012] (2) The method of the present invention is explained based on the image domain DPCA method, but the idea of ​​replacing different polarization images can be used in other full-polarization multi-channel SAR-GMTI detection results to improve the detection performance of moving targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flow chart of the present invention.

[0014] Figure 2(a) is a SAR image of the HH polarization measured data of the radar reference channel.

[0015] FIG. 2( b ) is the DPCA processing result of the present invention.

[0016] Figure 3 This is the CFAR detection result without data replacement.

[0017] Figure 4 This is the CFAR detection result of the present invention. DETAILED DESCRIPTION

[0018] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0019] A flow chart of a multi-channel SAR moving target detection method that fuses multi-polarization images is shown in the figure. Figure 1 As shown, the specific method is as follows:

[0020] Step 1, perform SAR imaging processing on the two channel data respectively to obtain the reference channel HH polarization complex image s1, VH polarization complex image s2 and the other channel HH polarization complex image s3, VH polarization complex image s4. The reference channel HH polarization SAR imaging result is shown in Figure 2(a), where the vertical direction is the azimuth direction, the horizontal direction is the range direction, the azimuth point number Na=8192, and the range point number Nr=1300. According to the imaging results, it can be seen that there are many building areas with strong scattering in the figure, as shown by the marks in the figure. The effectiveness of the method proposed in the present invention will be verified by the buildings at the marked locations.

[0021] Step 2: Perform image registration and channel equalization. The image registration uses the cross-correlation method to obtain the offset of the two images in the azimuth direction, and then interpolate and translate the image of the other channel outside the reference channel so that the pixels at the same position on the two images correspond to the same ground scene. Here, 10 times the cubic spline interpolation is performed, and then the registration amount calculated by the cross-correlation method is 1.0 pixel. Channel equalization uses a two-dimensional frequency domain adaptive channel equalization method to calculate the equalization weight coefficients in the range and azimuth directions respectively, and multiply the image of the other channel outside the reference channel by the equalization weight coefficients in the range and azimuth directions respectively, and correct the amplitude and phase errors of each pixel on the image of the other channel outside the reference channel to obtain the co-polarization complex image s3' and cross-polarization complex image s4' of the other channel outside the reference channel.

[0022] Step three, use the full polarization data of the reference channel for polarization classification, and adopt the H / α-Wishart iterative classification method to divide the ground objects into 8 categories based on the polarization scattering entropy H and the polarization angle α. Low-entropy multiple scattering and low-entropy dipole scattering are identified as building features to obtain the specific location of the building pixels in the image.

[0023] Step 4: Using the co-polarization image as the reference image, replace the pixels at the same position in the co-polarization image with the pixels where the building is located in the cross-polarization image for the two channels, and obtain the image s after the two-channel data is replaced 12 and 34 '.

[0024] Step 5: Replace the two channels with the image s 12 、s 34 'Subtract to complete the image domain DPCA processing dpca =s 12 -s 34 '.

[0025] The DPCA processing result after replacement is shown in Figure 2(b), from which we can see that most of the stationary clutter is well suppressed.

[0026] Step 6: Perform CFAR detection and design a hollow sliding window. Here, the sliding radius is set to 25, the protection unit radius is set to 5, and the statistical characteristics of the clutter in the window are calculated. Here, based on the Rayleigh distribution assumption, the false alarm rate P is set fa =10 -6 , the sliding window traverses the image, determines the threshold according to the local samples, and obtains the final moving target detection result. For easy observation, the amplitude of the remaining pixels is set to 1.

[0027] The HH polarization CFAR results are as follows Figure 3 As shown in Figure 1, it can be seen that the building at the marked position has many bright spots because the remaining energy after cancellation is still much stronger than the surrounding clutter. The CFAR detection results after data replacement are as follows: Figure 4 As shown, the energy of the building is significantly reduced, and the corresponding data unit detection threshold does not drop significantly, so the false alarm caused by the building is reduced.

[0028] In order to further verify the effectiveness of the algorithm proposed in the present invention, the images of region A and region B in the CFAR detection results are partially enlarged (region A occupies about 1000 azimuth units and 200 distance units; region B occupies about 1260 azimuth units and 250 distance units) and the number of false alarms is calculated. It can be clearly seen from the figure that after data replacement, the number of false alarm points is significantly reduced. There are about 82 false alarm points in the HH polarization CFAR result in region A, and about 10 false alarm points in the method proposed in the present invention, and the number of false alarms is reduced by 87.8%; there are about 131 false alarm points in the HH polarization CFAR result in region B, and about 60 false alarm points in the method proposed in the present invention, and the number of false alarms is reduced by 54.2%. The number of false alarm points is summarized in the following table.

[0029] Region A Region B Number of HH polarization false alarms 82 131 The number of false alarms of the present invention 10 60 False alarm reduction (%) 87.8 54.2

[0030] The present invention is not limited to the above specific implementations, and the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made to the above implementations based on the technical essence of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-channel SAR moving target detection method integrating multi-polarization images, characterized in that: The following steps are involved: Step 1: Perform SAR imaging processing on the full polarization data of the two channels respectively to obtain the co-polarization complex image s1 and the cross-polarization complex image s2 of the reference channel and the co-polarization complex image s3 and the cross-polarization complex image s4 of the other channel; Step 2: perform image registration and channel equalization. For image registration, the cross-correlation method is used to obtain the offset of different channel images in the azimuth direction, and then the data of the other channel is interpolated and translated so that the pixels at the same position on the two channel images correspond to the same target on the ground. For channel equalization, a two-dimensional frequency domain adaptive channel equalization method is used to calculate the equalization weight coefficients in the range and azimuth directions respectively, and the image of the other channel is multiplied by the equalization weight coefficients in the range and azimuth directions respectively, and the amplitude and phase errors of each pixel are corrected to obtain the co-polarization complex image s3' and the cross-polarization complex image s4' after processing of the other channel. Step 3: Use the full polarization data of the reference channel to perform polarization classification, and use the H / α-Wishart iterative classification method to identify buildings and determine their specific locations in the image; Step 4: Using the co-polarization image as the reference image, the pixels at the same position of the co-polarization image are replaced with the pixels where the buildings are located in the two-channel cross-polarization images to obtain the image s after the two-channel data is replaced. 12 and 34 '; Step 5: Replace the two channels with the image s 12 、s 34 'Subtract and complete the image domain DPCA processing; Step six, perform CFAR detection, use the CA-CFAR method to design a hollow sliding window, calculate the statistical characteristics of the clutter in the window, calculate the threshold based on the established random distribution assumption and threshold setting principle, slide the window to traverse the image, determine the threshold based on local samples, determine whether the moving target exists based on the set threshold, and obtain the final moving target detection result.

Images