A small unmanned aerial vehicle InSAR interferogram image pair registration method using external DEMs

Abstract

The application discloses a small unmanned aerial vehicle-borne InSAR interferogram image pair registration method using external DEM, divides images into two different types of regions by using external DEM data, adopts different registration strategies respectively, and realizes high-precision registration of small unmanned aerial vehicle-borne InSAR interferogram image pairs. The application can realize high-precision registration, improve registration efficiency, and provide high-precision inversion results for terrain surveying and mapping.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of interferometric radar, in particular to a small unmanned aerial vehicle-borne interferometric synthetic aperture radar interferogram pair registration method using external DEM. BACKGROUND

[0002] Airborne interferometric synthetic aperture radar (InSAR) has been widely used in geological disaster monitoring and warning, ground subsidence monitoring, topographic mapping and other fields. With the gradual increase of monitoring flexibility requirements, miniaturized and lightweight unmanned aerial vehicle-borne synthetic aperture radar (SAR) systems have gradually become a research hotspot.

[0003] The small unmanned aerial vehicle motion platform is easily affected by weather factors such as wind, and its motion trajectory often cannot follow a uniform straight line, showing nonlinear random disturbance, and the two flight tracks in the repeated track flight process cannot be completely parallel, and the coupling with the terrain causes the radar images of the two flights to not be completely matched, that is, the pixel points at the same position in the two images do not correspond to the same ground object target, and there is a non-negligible offset between the two pixel points corresponding to the same ground object target.

[0004] The traditional registration method is to use a window-based automatic matching method, select the position with the best evaluation index as the position of the homonym according to a certain evaluation function, and then establish a polynomial fitting model according to the coordinates of the selected homonym, and calculate the fitting coefficients using the least squares method, thereby establishing the coordinate mapping relationship from the main image to the auxiliary image.

[0005] The nonlinearity of the single flight track of the small unmanned aerial vehicle-borne InSAR, the inability of the two flight tracks to be parallel, and the offset between the two images and the terrain coupling will affect the accuracy of obtaining the global offset based on the polynomial fitting method, so the traditional registration method is not suitable for small unmanned aerial vehicle-borne InSAR image registration. Therefore, there is an urgent need for a small unmanned aerial vehicle-borne InSAR interferogram pair registration method to obtain high-quality interferometric phase images and thereby to invert high-precision terrain information. SUMMARY

[0006] Therefore, the present application provides a small unmanned aerial vehicle-borne InSAR interferogram pair registration method using external DEM, which can realize high-precision registration of non-parallel baseline interferogram pairs.

[0007] The small unmanned aerial vehicle-borne InSAR interferogram pair registration method using external DEM provided by the present disclosure includes the following steps:

[0008] A right-angle coordinate system of imaging space is constructed, and the two groups of SAR echo data are respectively processed by using a backward projection algorithm in the same right-angle coordinate system of imaging space;

[0009] The obtained DEM is converted into a corresponding right-angle coordinate system of imaging space;

[0010] The converted DEM is processed by block, and is divided into a small-elevation region close to an imaging plane and a large-elevation region far from the imaging plane, and the results after the block processing are analyzed by using a connected region analysis to remove a smaller part;

[0011] Any image is taken as a main image, a point with a higher signal-to-noise ratio in a small-elevation region in the main image is selected as a reference point, a window matching is selected, and a shift amount of the reference point between the main image and an auxiliary image is calculated; a global registration shift amount of the region is obtained by fitting, a mapping relationship between coordinates of the main image and the auxiliary image is established, the auxiliary image is interpolated, and registration of the small-elevation region is realized;

[0012] For the large-elevation region, the converted DEM and twice track information are used to jointly solve a registration shift amount, the auxiliary image is interpolated and resampled, coarse registration is realized; a pixel point with a high correlation coefficient after the coarse registration is selected as the reference point, a global registration shift amount of each connected region is obtained, the auxiliary image is interpolated, and registration of the large-elevation region is realized.

[0013] Further, the step of converting the obtained DEM into a corresponding right-angle coordinate system of imaging space specifically includes:

[0014] The original DEM in a geographic coordinate system is projected into a geocentric coordinate system;

[0015] The DEM result in the geocentric coordinate system is projected into a corresponding right-angle coordinate system of an imaging plane.

[0016] Further, the step of processing the converted DEM by block adopts an image segmentation method based on k-means.

[0017] Further, the step of calculating the shift amount of the reference point between the main image and the auxiliary image based on the window matching specifically includes:

[0018] A matching window of a certain size is determined with the reference point of the main image as the center, and a search window larger than the matching window is selected on the auxiliary image;

[0019] The matching window is moved pixel by pixel in rows and columns in the search window, and a correlation coefficient of the two matching windows in the main image and the auxiliary image is calculated;

[0020] The position of the same-named point is selected as the position of the maximum correlation coefficient in the search window, and a shift amount between the main image and the auxiliary image is obtained.

[0021] Further, the step of obtaining the global registration offset of the region by fitting adopts polynomial fitting, and specifically includes:

[0022]

[0023] Wherein, Δx k , Δy k respectively represent the registration offset of the reference point in the row and column, x k , y k respectively represent the coordinates of the reference point in the row and column, a1-a3, b1-b3 represent the registration coefficients to be estimated.

[0024] After the least square method is used to estimate the to-be-estimated transformation model coefficients, the global offset can be estimated according to the row and column coordinates of all points in the image.

[0025] Further, the step of jointly solving the registration offset by using the converted DEM and the two flight path information includes:

[0026] Taking the first flight path, the non-linear flight path is linearly fitted, and each terrain point on the DEM is rotated around the linear flight path, and the intersection of the imaging plane in the distance direction is the projection point position of the terrain point under the non-linear flight path.

[0027] Taking the second flight path, the above operation is repeated to obtain the projection point position of the terrain point under the second flight path.

[0028] The projection point positions obtained twice are subtracted, and the registration offset jointly solved by using the converted DEM and the two flight path information is obtained.

[0029] Further, the step of obtaining the global registration offset of the region adopts a multilayer perception machine, and the specific steps include:

[0030] The network is trained, the two-dimensional coordinates of the selected reference point are taken as the input, the prediction value of the two-dimensional offset is taken as the output, the two-dimensional offset of the reference point is taken as the label, the parameters in the network are corrected through the loss function feedback, and the network with the minimum loss function is obtained.

[0031] The global offset is obtained, the two-dimensional coordinates of each point in the region are taken as the input by using the trained network, and the global registration offset in the region is obtained.

[0032] The method for registering small unmanned aerial InSAR interferogram pairs provided by the disclosure utilizes external DEM data for terrain division; utilizes connected domain analysis to eliminate small areas; different registration strategies are adopted for two different types of regions, and high-precision registration of small unmanned aerial InSAR interferogram pairs is realized.

[0033] Compared with the prior art, the present disclosure has the following advantages: ① With external DEM data, the image is divided into regions, and different registration strategies are adopted, which improves the registration efficiency while achieving high-precision registration; ② Provides high-precision inversion results for topographic mapping; ③ Utilize the combination of DEM and two flight path information to improve the calculation registration accuracy; ④ Use multilayer perception neural network to improve the calculation speed. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The figure is a schematic diagram of DEM projection imaging geometry.

[0035] Figure 2 It is a search window and a matching window schematic diagram.

[0036] Figure 3 It is the flow chart of the present application. DETAILED DESCRIPTION

[0037] The present application will be described in detail below with reference to the accompanying drawings and examples.

[0038] The present application provides a small unmanned aerial InSAR interferogram image pair registration method using external DEM, as shown in the accompanying drawings, specifically comprising the following steps: Figure 3 The specific steps are as follows:

[0039] S1, construct an imaging space rectangular coordinate system, and image process two groups of SAR echo data under the same imaging rectangular coordinate system.

[0040] In the embodiment of the present application, each group of SAR echo must be imaged under the same imaging coordinate system, and the imaging algorithm is a backward projection algorithm.

[0041] The construction method of the imaging space rectangular coordinate system is as follows:

[0042] Take the center position of the radar track as the plumb line, find a point with a distance of H from the center of the radar track along the downward direction as the coordinate origin, where H is the flight height of the radar platform, take the plumb line as the z-axis, and the direction from the coordinate origin to the center position of the radar track as the x-axis direction, according to the right-hand rule, construct a three-dimensional rectangular coordinate system, at this time the xy plane is the radar imaging plane, the x axis is the azimuth axis, and the y axis is the range axis.

[0043] S2, convert the obtained DEM to the imaging space rectangular coordinate system xyz.

[0044] In the embodiment of the present application, the principle of converting the obtained external DEM to the space rectangular coordinate system corresponding to the imaging plane is as follows, and the schematic diagram is shown in the accompanying drawings. Figure 1 ​

[0045] The first step is to project the original DEM in the latitude-longitude height coordinate system to the geocentric coordinate system, that is, to convert the spatial coordinates (lat, lon, alt) of each point in the DEM, latitude (lat), longitude (lon) and height (alt), into the geocentric coordinate system (x, y, z) using formula (1)

[0046]

[0047] wherein N = 6378137

[0048] The second step is to project the DEM result (x, y, z) in the geocentric coordinate system to the corresponding spatial rectangular coordinate system of the imaging plane using formula (2).

[0049]

[0050] wherein (u x , u y , u z ), (v x , v y , v z ), (n x , n y , n z ) are the direction vectors of the x-axis, y-axis and z-axis in the geocentric coordinate system, respectively, and (x0, y0, z0) is the position of the coordinate origin in the geocentric coordinate system.

[0051] S3, the converted DEM is processed by block division, and is divided into two parts, which are the elevation area close to the imaging plane and the other elevation terrain area. The results of the block division are subjected to connected domain analysis, and the parts with small areas are removed.

[0052] In the embodiment of the application, the region block processing method can be a k-means-based image segmentation method.

[0053] The steps of the external DEM region block method based on k-means are as follows:

[0054] First step: arbitrarily select two values as the initial cluster centers;

[0055] Second step: calculate the difference between the height h of each point in the target region and each cluster center, and classify these points into the class in which the cluster center is closest to the point;

[0056] Third step: recalculate the average value of each class and update the cluster center of each class;

[0057] Fourth step: repeat the second and third steps, and output the clustering result when the cluster center no longer changes.

[0058] In the embodiment of the present application, the connected domain analysis method can be a connected domain analysis method based on two traversals or a connected domain analysis method based on seed filling.

[0059] S4, taking either of the two images for one homonym as the main image and the other as the auxiliary image, selecting a point with high signal-to-noise ratio in a small height region in the main image as the reference point, and using a window-based automatic matching method to obtain the offset of the reference point between the main and auxiliary images; then obtaining the global registration offset of the region by fitting, and establishing the mapping relationship of the coordinates of the main and auxiliary images. According to the mapping relationship, the information of the main image is interpolated to the auxiliary image, so that the registration of the small height region is realized.

[0060] In the embodiment of the present application, the window-based automatic matching method uses the maximum coherence coefficient as the criterion, and the specific steps are as follows, and the schematic diagram is shown in Figure 2 .

[0061] Step 1: Determine a matching window of a certain size with the reference point of the main image as the center, and select a search window larger than the matching window on the auxiliary image.

[0062] Step 2: Move the matching window in the search window by row and column pixel by pixel, and calculate the coherence coefficients of the two matching windows in the main and auxiliary images. Select the position of the maximum coherence coefficient in the search window as the position of the homonym, and obtain the offset between the main and auxiliary images.

[0063] In the embodiment of the present application, the polynomial fitting method is used to obtain the global registration offset, and the method is as follows:

[0064]

[0065] Where Δx k , Δy k represent the registration offsets of the reference point in the row and column respectively, x k , y k represent the coordinates of the reference point in the row and column respectively, and a1-a3, b1-b3 represent the registration coefficients to be estimated. After the least square method is used to estimate the to-be-estimated transformation model coefficients, the global offset can be estimated according to the row and column coordinates of all points in the image. Then, the auxiliary image is interpolated to obtain the image after fine registration.

[0066] S5, for large elevation area, the converted DEM and twice flight path information are used to jointly solve the registration offset, the auxiliary image is interpolated and resampled, and coarse registration is realized. The pixel points with high correlation coefficients after coarse registration are selected as reference points, and the global registration offset of each connected domain is obtained. The auxiliary image is interpolated, and the registration of the large elevation area is realized.

[0067] In the embodiment of the application, the method for jointly solving the registration offset by using the converted DEM and twice flight path information is as follows:

[0068] Firstly, the first flight path is taken, the nonlinear flight path is linearly fitted, each terrain point on the DEM is rotated around the linear flight path, and the intersection of the imaging plane in the distance direction is positive, that is, the projection point position of the terrain point under the nonlinear flight path; the second flight path is taken, and the above operation is repeated to obtain the projection point position of the terrain point under the second flight path. The difference between the two obtained projection point positions is the registration offset solved by using the converted DEM and twice flight path information.

[0069] In the embodiment of the application, the multilayer perception machine is used to obtain the global registration offset in the region, and the specific process is divided into two steps. The first step is to train the network, and the two-dimensional coordinates of the selected reference points are used as the input, and the two-dimensional offset prediction value is used as the output. The two-dimensional offset of the reference point is used as the label, the parameters in the network are corrected through the loss function feedback, and the network with the minimum loss function is obtained; the second step is to obtain the global offset, the two-dimensional coordinates of each point in the region are used as the input by using the trained network, and the global registration offset in the region is obtained.

[0070] In summary, the above is only a preferred embodiment of the application, and is not used to limit the protection scope of the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A method for registering InSAR interferometric images on a small UAV using an external DEM, characterized in that, include: An imaging spatial rectangular coordinate system is constructed, and the two sets of SAR echo data are processed by the back projection algorithm under the same imaging rectangular coordinate system. The acquired DEM is converted to the corresponding imaging space rectangular coordinate system; The converted DEM is divided into blocks, namely small elevation regions close to the imaging plane and large elevation terrain regions far from the imaging plane. Connectivity analysis is performed on the block results to remove the smaller parts. Using any image as the main image, select points with high signal-to-noise ratio in the low-elevation region of the main image as reference points. Based on window matching, calculate the offset of the reference points between the main and auxiliary images. By fitting, obtain the global registration offset of the region, establish the mapping relationship between the coordinates of the main and auxiliary images, and interpolate the auxiliary image to achieve registration in the low-elevation region. For high-elevation areas, the registration offset is jointly calculated using the converted DEM and two flight track information. The auxiliary image is then interpolated and resampled to achieve coarse registration. Pixels with high correlation coefficients after coarse registration are selected as reference points to obtain the global registration offset of each connected component. The auxiliary image is then interpolated to achieve registration in high-elevation areas. The step of jointly calculating the registration offset using the converted DEM and the two track information includes: Take the first trajectory, perform linear fitting on the nonlinear trajectory, rotate each terrain point on the DEM around the linear trajectory, and the intersection point with the imaging plane in the positive distance direction is the projection point position of the terrain point under the nonlinear trajectory. Repeat the above operation on the second flight path to obtain the projection point position of the terrain point under the second flight path; The difference between the two projection point positions can be used to obtain the registration offset calculated by combining the transformed DEM and the two track information.

2. The registration method as described in claim 1, characterized in that, The step of converting the acquired DEM to the corresponding imaging space rectangular coordinate system includes the following specific methods: Project the original DEM in the latitude, longitude and height coordinate system onto the geocentric and earth-fixed coordinate system; The DEM results in the geocentric-fixed coordinate system are projected onto the spatial rectangular coordinate system corresponding to the imaging plane.

3. The registration method according to claim 1, characterized in that, The step of dividing the converted DEM into blocks uses a k-means-based image segmentation method.

4. The registration method according to claim 1, characterized in that, The step of calculating the offset of the reference point between the main and secondary images based on window matching specifically includes: A matching window of a certain size is determined with the reference point of the main image as the center, and a search window larger than the matching window is selected on the auxiliary image; The matching window is moved pixel by pixel in rows and columns within the search window, while the coherence coefficients of the two matching windows in the main image and the auxiliary image are calculated simultaneously. The position of the maximum coherence coefficient within the search window is selected as the position of the corresponding point, and its offset between the main and auxiliary images is obtained.

5. The registration method according to claim 1, characterized in that, The step of obtaining the global registration offset of the region through fitting employs polynomial fitting and specifically includes: in These represent the registration offsets of the reference point in the row and column, respectively. These represent its coordinates in the row and column, respectively. This represents the registration coefficient to be estimated; After estimating the coefficients of the transformation model to be estimated using the least squares method, the global offset can be estimated based on the row and column coordinates of all points in the image.

6. The registration method according to claim 1, characterized in that, The steps for obtaining the global registration offset of each connected component are performed using a multilayer perceptron, and include: The network is trained by taking the two-dimensional coordinates of the selected reference point as input and outputting the predicted value of the two-dimensional offset. The two-dimensional offset of the reference point is used as the label, and the parameters in the network are corrected by feedback through the loss function to obtain the network with the minimum loss function. Global offset acquisition: Using the trained network, the two-dimensional coordinates of each point in the region are used as input to obtain the global registration offset of each connected component.

Images