A hybrid baseline tomographic sar image registration method

Abstract

The application belongs to the technical field of synthetic aperture radar, and provides a mixed baseline tomographic SAR image registration method, which can simultaneously solve the high-precision registration problem of long baseline images and short baseline images in tomographic SAR processing. The specific idea is that the stretching problem between short baseline image pairs can be ignored, and the registration parameters can be accurately extracted through the cross-correlation function; the stretching problem of long baseline image pairs is obvious, and the direct extraction method of pixel offset is abandoned, and the pixel offsets of multiple short baseline images are accumulated to indirectly obtain.

Description

Technical Field

[0001] This invention belongs to the field of synthetic aperture radar technology, and in particular relates to a hybrid baseline tomographic SAR image registration method, which can simultaneously solve the problem of high-precision registration of long and short baseline images in tomographic SAR processing. Background Technology

[0002] As an extension of Interferometric Synthetic Aperture Radar (InSAR) technology, Tomosynthetic Aperture Radar (TomoSAR) can solve the inherent overlapping problem in SAR imaging, enabling three-dimensional (3D) imaging. It is an effective method for 3D mapping, providing a robust solution for urban monitoring.

[0003] For TomoSAR, its 3D imaging processing is based on acquired SAR images, requiring precise registration of SAR images from different observation locations. Due to the slant-range projection mode of SAR imaging, image distortion will exist between different SAR images, often manifested as image stretching. The degree of image stretching is related to the difference in the downward viewing angle of the acquired SAR images. For short-baseline SAR images, the viewing angle change is small, so the stretching in its local areas is consistent and can be equivalent to pixel offset. However, in the case of long baselines, the image distortion problem becomes severe, and the obvious pixel stretching problem makes it difficult to extract pixel offset, rendering the most commonly used image registration algorithm based on local window cross-correlation ineffective.

[0004] Therefore, a robust tomographic SAR image registration method is needed that can simultaneously meet the high-precision registration requirements of long and short baseline image pairs. Summary of the Invention

[0005] This invention proposes a hybrid baseline tomography SAR image registration method that can simultaneously guarantee SAR image registration under both long and short baselines. The specific approach is as follows: the stretching problem between short baseline image pairs can be ignored, and registration parameters can be accurately extracted using a cross-correlation function; however, the stretching problem of long baseline image pairs is significant. Therefore, instead of directly extracting pixel offsets, the pixel offsets of multiple short baseline images are accumulated to achieve indirect acquisition. Furthermore, to avoid misalignment of pixel offsets for pixels with the same name in different images, the pixel offsets of pixels with the same name are registered using a step-by-step recursive approach.

[0006] Beneficial effects:

[0007] This invention addresses the issue of mixed long and short baseline tomographic SAR image stacks, employing both indirect and direct pixel offset extraction methods to achieve accurate image registration. Specifically, for short baselines, a cross-correlation function is used to estimate registration parameters; for long baselines, the accumulation of pixel offsets from multiple short baselines is used to extract registration parameters, with a step-by-step recursive process to correct pixel offset misalignments between corresponding pixels. Compared to methods that directly register using the cross-correlation function of two SAR images, this invention can more robustly estimate the registration parameters of long baseline image pairs and simultaneously achieve high-precision registration of both long and short baseline image pairs. Attached Figure Description

[0008] Figure 1 Schematic diagram of image stretching problem and short baseline registration parameter accumulation.

[0009] Figure 2 Flowchart of Hybrid Baseline Tomography SAR Image Registration Method

[0010] Figure 3 Detailed illustration of hybrid baseline tomography SAR image registration in the case of 4 images Detailed Implementation

[0011] The present invention will now be clearly and completely described in conjunction with implementation examples and corresponding drawings.

[0012] The core issue that needs to be explained in this method is the pre-registration processing of image pixel offsets. During the pixel offset extraction process for image pairs, the pixel offset represents the resampling parameter from image registration to the master image, and this offset corresponds one-to-one with the pixel units of the master image. When it is necessary to obtain the pixel offset of the long baseline through the accumulation of short baselines, if the accumulation process is performed directly, due to the pixel offset between the master images of different image pairs, there will be a misalignment problem between pixel offsets corresponding to the same coordinate position. In this case, the accuracy of pixel offset accumulation will be affected, thus affecting the registration accuracy of the long baseline.

[0013] Therefore, this method adds a pre-registration process for pixel offsets before accumulating the pixel offsets of different short baseline image pairs. Due to the one-to-one correspondence between the main image pixel units and the offsets, if the current main image can be registered with other main images, its corresponding pixel offset can be resampled according to the registration parameters of the current main image to correct misalignment. Simultaneously, since the short baselines are interconnected, the subordinate image of the current baseline is precisely the main image of the next-level baseline. Here, taking the first-level baseline as an example, the pixel offset of the second-level baseline can be resampled according to the registration parameters of the first-level baseline to complete the pre-registration of the pixel offsets of these two baselines. After pre-registration, the pixel offsets can be accumulated to obtain the registration parameters of the first and second-level baselines. Based on this, the pre-registration of the third-level pixel offset can be performed, and the registration parameters of the first to third-level baselines can be accumulated. And so on, through step-by-step recursive and accumulative processing, long baseline image registration can be completed. After the registration of the longest baseline image pair is completed, the registration of all short baseline image pairs is also completed.

[0014] The main processing steps of this method are divided into three parts: First, the pixel offset of the short baseline image is obtained by cross-correlation measurement of the local window of the image; then, in order to achieve pixel-by-pixel alignment of the short baseline pixel offset, the pixel offset is resampled in a stepwise recursive manner; finally, the pixel offset of the long baseline image is obtained by accumulating several connected short baseline images, and the image is resampled according to the accumulated pixel offset to complete the image registration.

[0015] Among them, the image stretching problem caused by long baselines is as follows: Figure 1 As shown, the processing flowchart of the tomographic SAR image stack registration method based on short baseline recursion is as follows: Figure 2 As shown, the specific steps include:

[0016] Step 1: Extracting Short Baseline Pixel Offsets

[0017] First, all images are sorted according to the lower viewpoint. The registration method is as follows: Figure 3 As shown, here, image-1 is set as the master image, and all other images should be registered to image-1. Next, images with adjacent indices form short baseline pairs, and several sample points from the master image are selected (e.g., ...). Figure 3 Registration parameters (A, B, F) from (a), (b), (c), and (d) are extracted. Then, the pixel offsets of these sample points in the paired images (e, (f), and (g)) are extracted based on the real correlation algorithm. Finally, the pixel offsets of the entire image are obtained by interpolation based on the pixel offsets of the sample points.

[0018] The pixel offsets in the range and azimuth directions between image-1 and image-2 are respectively set as follows: and (r,a) are pixel coordinates. and Represents the pixel offsets in the range and azimuth directions between images image-2 and image-3.

[0019] Step 2: Recursively accumulate pixel offsets

[0020] First, the pixel offset of the current baseline is resampled based on the extracted pixel offset of the previous baseline. For example, the pixel offset resampling between image-2 and image-3 is performed based on the pixel offset between image-1 and image-2 (e.g., ...). Figure 3 (as shown in (e), (f), and (h)). Assume the resampled distance and orientation pixel offsets are respectively... and At this point:

[0021]

[0022] Secondly, based on the resampled pixel offset, a longer baseline pixel offset can be obtained through accumulation processing (such as...). Figure 3 (as shown in (e), (h), and (j)). For example, the pixel offset between images -1 and -3 can be obtained using the following formula:

[0023]

[0024] Following the above method, the pixel offsets of image-1 and image-4 can also be achieved through recursive accumulation (e.g., Figure 3 (as shown in (j), (g), (i), (k)). When the number of images is greater than 3, a more general recursive formula can be given, namely:

[0025]

[0026]

[0027] Step 3: Image resampling

[0028] Based on the obtained pixel offsets, image resampling can be performed to achieve accurate image registration. The registered image is shown below. Figure 3 As shown in (l), (m), (n), and (o). Image resampling is essentially an interpolation process, which can be performed using the following formula: Where S i S is the i-th original image before resampling. i,j It is the i-th image that was resampled after being registered to the j-th image.

[0029]

[0030] Following the above method, long and short baseline image pair registration can be completed simultaneously. Furthermore, in this method, only a limited number of short baseline image pairs are directly estimated using sliding window operations, while the registration of long baseline image pairs only involves accumulation and interpolation. This reduces the computational load in the image registration process, improves processing efficiency, and effectively ensures the efficient construction of high-quality tomographic SAR data stacks.

Claims

1. A hybrid baseline tomography SAR image registration method, characterized in that, Includes the following steps: Step 1: Extract short baseline pixel offset; Step 2: Resample the pixel offset of the current baseline based on the extracted pixel offset of the previous baseline. Based on the resampled pixel offset, obtain the pixel offset of the longer baseline through recursive accumulation. Step 3: Image resampling.

2. The hybrid baseline tomography SAR image registration method as described in claim 1, characterized in that, In step two, the pixel offset of the current baseline is resampled based on the extracted pixel offset of the previous baseline. The specific formula is as follows: ； Then, based on the resampled pixel offset, the pixel offset of the longer baseline is obtained through recursive accumulation. The specific recursive formula is as follows: 。 3. The hybrid baseline tomography SAR image registration method as described in claim 1, characterized in that, In step three, image resampling is performed based on the obtained pixel offsets to achieve accurate image registration. Image resampling is essentially an interpolation process, and its formula is as follows: ； in, It is the first before resampling One original image; Is it registered to the first The first image after resampling Image.

Images