A method for fine registration and stereo measurement of spaceborne InSAR images

Abstract

This invention relates to a method for fine registration and stereo measurement of spaceborne InSAR images, comprising: coarsely registering primary and secondary single-view complex images using a cross-correlation registration method based on polynomial fitting, and dividing the primary single-view complex image and the coarsely registered secondary single-view complex image into grid blocks; calculating the coherence of the coarsely registered secondary single-view complex image under different range offsets based on an improved mean-normalized de-flattening coherence coefficient, and extracting the offset of the peak value of the grid block coherence coefficient distribution; smoothing the offset using a coherence mask and filtering algorithm, resampling the coarsely registered secondary single-view complex image and using it for interferometric processing; summing the smoothed stereo measurement relative offset with the polynomial registration offset, and converting it into a stereo measurement relative elevation map after de-flattening processing. This invention is used to achieve fine registration of long-baseline primary and secondary SAR images in InSAR applications to generate high-quality interferometric phase maps.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a method for fine registration and stereo measurement of spaceborne InSAR images. Background Technology

[0002] A Digital Image Model (DEM) is a three-dimensional digital model describing the shape of the Earth's surface. It consists of a series of datasets containing geographic plane coordinates and elevations, and has significant application value in scientific research, economic construction, and military fields. In certain specific applications, such as seismic deformation extraction and terrain monitoring, high-resolution, high-precision DEMs are particularly important; however, their extraction is usually very complex. InSAR (Interferometric Synthetic Aperture Radar), due to its all-weather, all-day operation and as an active sensor, has become one of the methods for accurately acquiring DEMs. The quality of the interferometric phase map obtained after processing, such as the registration of primary and secondary SAR images, directly determines the accuracy of DEM acquisition using InSAR technology. Therefore, to obtain high-quality interferograms, it is necessary to perform sub-pixel-level fine registration of the primary and secondary single-view complex images, typically requiring a registration error of less than 1 / 8 pixel. Larger registration errors will result in severe decoherence, leading to a large amount of noise in the interferometric phase map.

[0003] Currently, common registration processes typically employ cross-correlation and polynomial fitting methods, which generally meet registration requirements to a certain extent. However, for SAR sensors with lower carrier frequencies, such as L-band sensors, longer baselines are often used for interferometric measurements. When the terrain is steep, a longer baseline can cause varying degrees of distance-to-pixel shift at different elevations. In such cases, common cross-correlation and polynomial fitting methods will fail to meet the required accuracy, leading to decoherence in parts of the interferogram and preventing the acquisition of the correct interferometric phase.

[0004] Furthermore, traditional stereo measurement typically achieves high vertical measurement accuracy by registering amplitude information under conditions of large angular differences (greater than 5°). However, the angular difference between the primary and secondary single-view complex images used for InSAR measurement is usually less than 0.1°, making it difficult to obtain stereo measurement results due to limitations in registration accuracy. Therefore, high-precision, high-performance registration methods require further research and development. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for fine registration and stereo measurement of spaceborne InSAR images. This method achieves fine registration of long-baseline primary and secondary SAR images in InSAR applications to generate high-quality interferometric phase maps. Simultaneously, using the offset obtained from the fine registration, stereo measurement results are calculated under minimal viewing angle differences.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for fine registration and stereo measurement of spaceborne InSAR images includes the following steps:

[0008] Step 101: Using a registration method based on cross-correlation and polynomial fitting, calculate the polynomial registration offset map and the auxiliary single-view complex image of coarse registration;

[0009] Step 102: Divide the main single-view complex image and the coarsely registered auxiliary single-view complex image into grid blocks, and calculate the coherence coefficient of the auxiliary single-view complex image block under different offsets based on the improved mean normalization to remove the flat ground coherence coefficient.

[0010] Step 103: Calculate the precise position of the peak value as the distance offset of the grid block by interpolating the coherence coefficient distribution under different offsets in Step 102, record the peak value as the optimal coherence coefficient, and estimate the grid block offset by weighting for the case of multiple peak values.

[0011] Step 104: Use coherence mask interpolation and filtering algorithms to smooth the offset of the mesh block in the distance direction to obtain the relative offset map of the stereo measurement in the distance direction. Combine the offset map of polynomial registration to calculate the absolute offset of the stereo measurement. After removing the flat ground, obtain the relative elevation map of the stereo measurement.

[0012] Step 105: Based on the relative offset map of the stereo measurement distance, resample the coarsely registered auxiliary single-view complex image to obtain a finely registered auxiliary single-view complex image, and then perform interferometric processing to obtain a high-quality interferometric phase map.

[0013] Further, step 101 specifically includes:

[0014] First, select a certain number of blocks in the same region of the primary and secondary single-view complex images, and register them using the cross-correlation method. Then, use a polynomial to fit the registration offset result to obtain the coarse registration offset. Resample the secondary single-view complex image to obtain the coarsely registered secondary SLC image, thereby reducing the search range for fine registration.

[0015] Further, step 102 specifically includes:

[0016] The main SLC image and the coarsely registered auxiliary SLC image are divided into grid blocks with a fixed window size. Then, the search range and spacing of the offset are estimated. Based on the mean normalization of the flat ground coherence coefficient, the coherence coefficient of the auxiliary single-view complex image block under different offsets is calculated.

[0017] Furthermore, step 103 specifically includes:

[0018] Calculate the peak location of the mean-normalized coherence coefficient distribution in flat terrain. Interpolate the peak point and several nearby points to calculate the precise location of the maximum value as the distance offset of the grid block, and use this maximum value as the optimal coherence coefficient. For the case of multiple peaks, use the peak size as the weight to estimate the position of the peak as the distance offset of the grid block.

[0019] Further, step 104 specifically includes:

[0020] The lower coefficient values ​​in the phase coefficient map recorded during peak estimation are masked, and the masked positions are interpolated. Then, the distance offset of the grid blocks is smoothed using median filtering and nonlocal mean filtering algorithms to obtain the stereo measurement relative offset map. Subsequently, the stereo measurement absolute offset is calculated by combining the polynomial registration offset map, and the stereo measurement relative elevation map in the radar coordinate system is obtained after de-flattening.

[0021] Further, step 105 specifically includes:

[0022] The stereo measurement relative offset map is interpolated to make it the same size as the coarsely registered auxiliary SLC image. The coarsely registered auxiliary single-view complex image is resampled according to the offset value of each pixel to obtain the finely registered auxiliary single-view complex image. Then, the main single-view complex image and the finely registered auxiliary single-view complex image are subjected to interferometry to obtain a high-quality interferometric phase map.

[0023] Furthermore, the calculation method for the mean-normalized flatland coherence coefficient is as follows:

[0024] The first step is to uniformly divide the image block into N grids. p For each sub-block, calculate the coherence coefficient γ. q :

[0025]

[0026] Among them, I m I s These are the primary and secondary single-view complex images within the sub-image block region, respectively. For I s The conjugate of , E{} represents the mean over all elements. Describing the natural constant e Power, |I m | 2 Indicates the calculation of I m The square of the modulus of each element, |I s | 2 Indicates the calculation of I s The square of the modulus of each element, γ q In this context, 'q' represents the sub-block number.

[0027] Then, these sub-blocks are averaged to obtain the mean-normalized flatland coherence coefficient.

[0028]

[0029] Where q is the sub-block number, ranging from 1 to N. p integers, N p The initial number of sub-blocks, N, is the number of sub-blocks. p The mean is 1, when the coherence coefficient of flat land is normalized. When the value is less than the set coherence threshold, increase N. p The number of N, where N p ∈{n 2 ..., n = 1, 2, 3, ..., until... If the coherence threshold is met, then select [the appropriate threshold]. N at its maximum p The data is segmented to calculate the mean-normalized flatland coherence coefficient.

[0030] This invention achieves fine registration and stereo measurement of SAR coherent images, and compared with existing technologies, it has the following significant advantages:

[0031] (1) This invention utilizes mean normalization to remove the coherence coefficient of flat terrain and multi-peak estimation method to perform fine grid estimation of offset, which significantly improves the accuracy of single-look complex SAR image registration in InSAR applications.

[0032] (2) The present invention estimates the peak position by performing one-dimensional interpolation on the coherence coefficient of discrete sampling, which greatly reduces the amount of computation and significantly improves the computation speed.

[0033] (3) Under the condition of extremely small viewing angle difference, the present invention uses the offset of fine registration and adopts median filtering and nonlocal mean filtering methods to obtain stereo measurement results, which can be used for verification of interferometric unwinding phase, assistance of interferometric phase diagram unwinding and other applications. Attached Figure Description

[0034] Figure 1This is a flowchart of the spaceborne InSAR image fine registration and stereo measurement method of the present invention;

[0035] Figure 2 This is a geometric schematic diagram of the present invention;

[0036] Figure 3 It is a polynomial registration interferometric phase diagram;

[0037] Figure 4 This is the fine registration interferometric phase diagram in this invention;

[0038] Figure 5 This is a partial detail of the polynomial registration interferometric phase diagram;

[0039] Figure 6 This is a partial detail view of the fine registration interferometric phase diagram in this invention;

[0040] Figure 7 It is the coherence coefficient of the polynomial registration interferometric phase diagram;

[0041] Figure 8 This is a schematic diagram of the coherence coefficients of the finely registered interferometric phase diagram in this invention;

[0042] Figure 9 This is the result of the filtered stereoscopic measurement relative elevation in this invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0044] The spaceborne InSAR image fine registration and stereo measurement method of this invention first uses a registration method based on cross-correlation and polynomial fitting to coarsely register the main and auxiliary single-view complex images. Then, the main single-view complex image and the coarsely registered auxiliary single-view complex image are divided into grid blocks. Next, based on an improved mean normalization method for removing flat terrain coherence coefficients, the coherence coefficients of the auxiliary single-view complex image at different offsets are calculated, and the peak value of the coherence coefficient distribution for each grid block is extracted. Then, the precise location of the actual peak value is estimated by interpolation as the offset of the grid block, and the grid block offset is estimated by weighting for multi-peak cases. Then, the range offset of the grid blocks is smoothed using coherence mask interpolation and filtering algorithms to obtain a stereo measurement relative offset map. The stereo measurement absolute offset is calculated by combining the polynomial registration offset map, and after flat terrain processing, a stereo measurement relative elevation map is obtained. Finally, based on the smoothed stereo measurement relative offset map, the coarsely registered auxiliary single-view complex image is resampled to obtain a finely registered auxiliary single-view complex image, which is then subjected to interferometric processing to obtain a high-quality interferometric phase map.

[0045] Specifically, such as Figure 1 As shown, the spaceborne InSAR image fine registration and stereo measurement method of the present invention includes the following steps:

[0046] Step 101: Using a registration method based on cross-correlation and polynomial fitting, calculate the polynomial registration offset map and the auxiliary single-view complex image of coarse registration.

[0047] Specifically, a certain number of control points are selected in the same region of both the primary and secondary single-view complex images. A fixed-size window is then selected centered on each control point; the window size is determined by the image resolution, typically set to 128×128 pixels. Subsequently, cross-correlation calculations are performed on the amplitude information within the windows of the primary and secondary single-view complex images. The position of the maximum value of the cross-correlation function is used as the offset of that control point. The registration offset results are then fitted, and the least squares method is used to calculate the polynomial coefficients to obtain the coarse registration offset Δx in the range direction. Nr Coarse registration offset Δx in azimuth direction Na :

[0048] Δx Nr =a0+a1i+a2j+a3ij+a4i+a5j

[0049] Δx Na =b0+b1i+b2j+b3ij+b4i 2 +b5j 2

[0050] Where a0-a5 are the range polynomial coefficients obtained by least squares fitting, b0-b5 are the azimuth polynomial coefficients obtained by least squares fitting, i is the azimuth pixel position of the auxiliary single-view complex image, and j is the range pixel position of the auxiliary single-view complex image.

[0051] Then, the offset is used to resample and move the auxiliary single-view complex image to obtain a coarsely registered auxiliary single-view complex image, thereby reducing the search range for fine registration.

[0052] Step 102: Divide the main single-view complex image and the coarsely registered auxiliary single-view complex image into grid blocks, and calculate the coherence coefficient of the auxiliary single-view complex image block under different offsets based on the improved mean normalization to remove the flat ground coherence coefficient.

[0053] Specifically, firstly according to Figure 2 Calculate the flat-ground phase based on the satellite geometry shown. The calculation method is as follows:

[0054]

[0055] Where λ is the wavelength of the electromagnetic wave emitted by the radar, and B len θ is the baseline length. m The viewpoint is the angle between the baseline and the horizontal plane in a single-view complex image, α is the angle between the baseline and the horizontal plane, and π is the value of pi.

[0056] Subsequently, the primary single-view complex image and the coarsely registered secondary single-view complex image are divided into grid blocks. Then, the search range for the range offset is estimated, and the secondary single-view complex image is moved along the range direction at fixed intervals. For non-integer pixel movements, an interpolation movement method is used to resample the complex image. Interpolation movement methods include, but are not limited to, sinc interpolation, cyclic displacement matrix, and other sub-pixel image movement methods. An improved mean-normalized de-flattening coherence coefficient is calculated for each image block after each movement.

[0057] The mean-normalized de-flattening coherence coefficient calculation method is as follows: First, the image block is uniformly divided into N grids. p For each sub-block, calculate the coherence coefficient γ. q :

[0058]

[0059] Among them, I m I s These are the primary and secondary single-view complex images within the sub-image block region, respectively. For I s The conjugate of , E{} represents the mean over all elements. Describing the natural constant e Power, |I m | 2 Indicates the calculation of I m The square of the modulus of each element, |I s | 2 Indicates the calculation of I s The square of the modulus of each element, γ q In this context, 'q' represents the sub-block number.

[0060] Then, these sub-blocks are averaged to obtain the mean-normalized flatland coherence coefficient.

[0061]

[0062] Where q is the sub-block number, ranging from 1 to N. p integers, N p The initial number of sub-blocks, N, is the number of sub-blocks. p The value is 1, the number of blocks is 1, and the mean is normalized to remove the coherence coefficient of flat land. If the value is less than the set coherence threshold (e.g., 0.6), increase N. p The number of N, where N p ∈{n 2 ..., n = 1, 2, 3, ..., until... If the coherence threshold is met, then select [the appropriate threshold]. N at its maximum p The data is segmented to calculate the mean-normalized flatland coherence coefficient.

[0063] Step 103: Calculate the precise location of the peak as the distance offset of the grid block by interpolating the distribution of the mean-normalized flat coherence coefficient, record the peak size as the optimal coherence coefficient, and estimate the grid block offset by weighting for the case of multiple peaks.

[0064] Specifically, the mean normalized de-flattening coherence coefficient of each grid block under different distance offsets k was obtained. Then, the peak extraction method is first used to obtain... The size of all peak points and their offset k are recorded, with the largest peak value being the optimal coherence coefficient. Subsequently, interpolation is performed on the peak point and several nearby points, typically five. The interpolation method includes, but is not limited to, cubic spline interpolation and cubic interpolation. The precise location of the peak value is then obtained from the interpolation result as the distance offset of the grid block.

[0065] Subsequently, for the multi-peak case, a weighted method is used to estimate the distance offset of the grid blocks.

[0066]

[0067] Where P is the set of peak points, k is an element in set P, and Δx off (k) represents the offset corresponding to the peak point.

[0068] Step 104: Use coherence mask interpolation and filtering algorithms to smooth the offset of the grid blocks in the distance direction to obtain the stereo measurement relative offset map. Combine the polynomial registration offset map to calculate the stereo measurement absolute offset. After de-flattening, obtain the stereo measurement relative elevation map.

[0069] Specifically, the lower coefficient values ​​recorded in the phase coefficient diagram during peak estimation are considered invalid regions, and a mask is used to mark these regions as invalid. Two-dimensional interpolation is then performed on the invalid locations filtered out by the mask. The interpolation methods include, but are not limited to, conventional interpolation methods such as quadratic spline interpolation and cubic interpolation.

[0070] Subsequently, median filtering and nonlocal mean filtering algorithms were applied sequentially to smooth the distance offset of the grid blocks, resulting in a stereoscopic relative offset map. Then, the absolute stereoscopic offset was calculated using the polynomial registration offset map, followed by geometric transformation and terrain removal, and the relative elevation h at each point was calculated.

[0071]

[0072] Among them, h flt The ground leveling height can be obtained from the ground leveling process of interferometry, Δx. Nr The distance-oriented coarse registration offset mentioned in step 101; Δh stereo The height resolution corresponding to the stereo measurement offset is expressed as follows:

[0073]

[0074] Where ρ is the range sampling resolution of the SAR main single-view complex image, and θ m The viewpoint under the main single-view complex image, R m R s B represents the slant distance of the main and secondary single-view complex images, respectively. len Let be the total baseline length, and Δθ be the incident angle difference between the primary and secondary single-view complex images. Finally, the stereoscopic relative elevation map h in the radar coordinate system is obtained.

[0075] Step 105: Based on the relative offset map of the stereo measurement distance, resample the coarsely registered auxiliary single-view complex image to obtain a finely registered auxiliary single-view complex image, and then perform interferometric processing to obtain a high-quality interferometric phase map.

[0076] Specifically, the stereo range relative offset map is interpolated to the same pixel size as the coarsely registered auxiliary SLC image. The interpolation method includes, but is not limited to, common methods such as two-dimensional bilinear interpolation and two-dimensional cubic interpolation. Based on the queried range relative offset map, the coarsely registered auxiliary single-view complex image is resampled and shifted to obtain the finely registered auxiliary single-view complex image.

[0077] Subsequently, conventional interferometric processing operations were performed on the primary single-view complex image and the finely registered secondary single-view complex image to obtain a high-quality interferometric phase map with fine registration.

[0078] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0079] Example 1

[0080] The technical solution of this invention was verified using field measurement data from the ALOS satellite SAR sensor (PALSAR) in mountainous areas. Figure 3 It is an interferometric phase map obtained by the polynomial registration method. Figure 4 The interferometric phase diagram is obtained by the fine registration method in this invention. Figure 5 This is a partial detail of the polynomial registration interferometric phase map. Figure 6 This is a partial detail view of the fine registration interferometric phase map in this invention. Figure 7 These are the coherence coefficients of the polynomial registration interferometric phase diagram. Figure 8 It is the coherence coefficient of the finely registered interferometric phase map in this invention. Figure 9 The results show the filtered stereoscopic relative elevation of the interferometric phase map obtained in this invention. It can be seen that compared to the interferometric phase map obtained by the cross-correlation polynomial registration method, the fine registration method proposed in this invention significantly improves the quality of the interferometric phase map, with clearer fringes and a higher coherence coefficient. Statistical results show that the average coherence coefficient of the interferometric phase map increases from 0.51 to 0.60. Furthermore, the stereoscopic measurement results detected by this invention can reflect the absolute changes in terrain, providing error detection capabilities for InSAR processing.

[0081] The above description is only a partial embodiment of the present invention and is not intended to limit the scope of protection of the present invention.

Claims

1. A method for fine registration and stereo measurement of spaceborne InSAR images, characterized in that, Includes the following steps: Step 101: Using a registration method based on cross-correlation and polynomial fitting, calculate the polynomial registration offset map and the auxiliary single-view complex image of coarse registration; Step 102: Divide the main single-view complex image and the coarsely registered auxiliary single-view complex image into grid blocks. Calculate the coherence coefficient of the auxiliary single-view complex image blocks at different offsets based on the improved mean normalization and flat-ground coherence coefficient, including: The main single-view complex image and the coarsely registered auxiliary single-view complex image are divided into grid blocks with a fixed window size. Then, the search range and spacing of the offset are estimated. Based on the mean normalization of the flat ground coherence coefficient, the coherence coefficient of the auxiliary single-view complex image block under different offsets is calculated. The method for calculating the mean-normalized coherence coefficient for flat terrain is as follows: First, the image patch is divided again into a grid. For each sub-block, calculate the coherence coefficient. : ； in, , These are the primary and secondary single-view complex images within the sub-image block region, respectively. for conjugate, This means calculating the mean of all elements. Represents the natural constant of Power of 1 Indicates calculation The square of the modulus of each element in the equation. Indicates calculation The square of the modulus of each element in the equation. In Indicates the sub-block number; Then, these sub-blocks are averaged to obtain the mean-normalized flatland coherence coefficient. : ； in, The sub-blocks are numbered from 1 to... integers, The initial number of sub-blocks. The mean is 1, when the coherence coefficient of flat land is normalized. When the value is less than the set coherence threshold, increase The number of, among which ,until If the coherence threshold is met, then select [the appropriate threshold]. Maximum As the optimal result; Step 103: Calculate the precise position of the peak as the distance offset of the grid block by interpolating the coherence coefficient distribution under different offsets calculated in Step 102, record the peak size as the optimal coherence coefficient, and estimate the grid block offset by weighting for the multi-peak case. Step 104: Use coherence mask interpolation and filtering algorithms to smooth the offset of the mesh block in the distance direction to obtain the relative offset map of the stereo measurement in the distance direction. Combine the offset map of polynomial registration to calculate the absolute offset of the stereo measurement. After removing the flat ground, obtain the relative elevation map of the stereo measurement. Step 105: Based on the relative offset map of the stereo measurement distance, resample the coarsely registered auxiliary single-view complex image to obtain a finely registered auxiliary single-view complex image, and then perform interferometric processing to obtain a high-quality interferometric phase map.

2. The method for fine registration and stereo measurement of spaceborne InSAR images according to claim 1, characterized in that, Step 101 specifically includes: First, select a certain number of blocks in the same region of the primary and secondary single-view complex images, and register them using the cross-correlation method. Then, use a polynomial to fit the registration offset result to obtain the coarse registration offset. Resample the secondary single-view complex image to obtain the coarsely registered secondary single-view complex image, thereby reducing the search range for fine registration.

3. The method for fine registration and stereo measurement of spaceborne InSAR images according to claim 1, characterized in that, Step 103 specifically includes: Calculate the peak location of the mean-normalized coherence coefficient distribution in flat terrain. Interpolate the peak point and several nearby points to calculate the precise location of the maximum value as the distance offset of the grid block, and use this maximum value as the optimal coherence coefficient. For the case of multiple peaks, use the peak size as the weight to estimate the position of the peak as the distance offset of the grid block.

4. The method for fine registration and stereo measurement of spaceborne InSAR images according to claim 3, characterized in that, Step 104 specifically includes: The lower coefficient values ​​in the phase coefficient map recorded during peak estimation are masked, and the masked positions are interpolated. Then, the distance offset of the grid blocks is smoothed using median filtering and nonlocal mean filtering algorithms to obtain the stereo measurement relative offset map. Subsequently, the stereo measurement absolute offset is calculated by combining the polynomial registration offset map, and the stereo measurement relative elevation map in the radar coordinate system is obtained after de-flattening.

5. The method for fine registration and stereo measurement of spaceborne InSAR images according to claim 4, characterized in that, Step 105 specifically includes: The stereo measurement relative offset map is interpolated to make it the same size as the coarsely registered auxiliary single-view complex image. The coarsely registered auxiliary single-view complex image is resampled according to the offset value of each pixel to obtain the finely registered auxiliary single-view complex image. Then, the main single-view complex image and the finely registered auxiliary single-view complex image are subjected to interferometry to obtain a high-quality interferometric phase map.

Images