An automatic top-bottom offset correction method for building extraction graph patches of remote sensing images

Abstract

The application provides a kind of automatic top and bottom offset correction method for building extraction of remote sensing image building extraction, comprising the following steps, obtain the DSM image and DEM image of target area with the same resolution of building top image spot image of target area, difference operation is carried out based on pixel value, i.e.DSM-DEM, obtain nDSM image containing building height data of target area;According to the satellite collection posture information when the original remote sensing image of extraction image spot is collected, the nDSM image is offset reconstructed, so that the reconstructed nDSM image has the top and bottom offset information caused by the satellite collection posture of building;The height value corresponding to each building top image spot is obtained by position matching, and the top and bottom offset correction is carried out on the position of each building top image spot combined with satellite collection posture information, to obtain the bottom image spot of each building.The application effectively improves the position accuracy of building extraction image spot, provides a basis for building high-precision modeling and positioning, and improves the efficiency of image spot extraction.

Description

Technical Field

[0001] This invention belongs to the field of post-processing technology for extracted data from remote sensing images, and specifically relates to an automated method for correcting top and bottom offsets in extracted building patches from remote sensing images. Background Technology

[0002] In the field of remote sensing image information extraction, the extraction of ground feature information has always been a research hotspot. Traditional remote sensing image information extraction algorithms mainly extract information based on the texture, shape, spatial, and spectral features of ground features. In recent years, deep learning-based information preview technology has also provided new solutions for remote sensing image information extraction due to its ability to utilize the deep features of images.

[0003] During in-orbit imaging, remote sensing satellites experience changes in flight attitude due to various reasons, and the sensor's shooting perspective is often sideways rather than frontally, causing image distortion. Among these distortions, the top-to-bottom offset of features with a certain height caused by satellite tilt has always been an inherent problem in remote sensing imagery. When extracting buildings from remote sensing imagery, the outline of the building's top surface is often extracted. However, due to the influence of the satellite tilt angle, buildings exhibit top-to-bottom offset in the actual image, causing a discrepancy between the extracted building outline and the geographical location of the building's base, thus affecting the subsequent use of the building outline.

[0004] The existing remote sensing image information extraction has the following shortcomings: there is a certain error between the extracted building location and the actual geographical location, which affects the accuracy of the building location; at the same time, for the same area, the top and bottom of buildings in the images are different due to the different orientation of the sensors at different imaging times, which makes the extracted building locations in the same area different at different times. This requires repeated extraction of buildings in the same area, which is labor-intensive and inefficient. Summary of the Invention

[0005] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide an automated top and bottom offset correction method for extracting building patches from remote sensing images, thereby improving the accuracy and efficiency of extracting building patches from remote sensing images.

[0006] The technical solution of this invention is:

[0007] An automated method for correcting top and bottom offsets in building extraction patches from remote sensing imagery includes the following steps:

[0008] 1) Acquire satellite remote sensing images of the target area, and obtain satellite acquisition attitude information during image acquisition, including satellite orbital altitude, satellite side tilt angle, and image azimuth angle; extract building top patch images of the target area from the satellite remote sensing images;

[0009] 2) Obtain DSM and DEM images of the target area with the same resolution as the building top patch image of the target area, and perform a difference operation on the DSM and DEM images based on the pixel values ​​to obtain an nDSM image containing building height data of the target area.

[0010] 3) The nDSM image is offset and reconstructed based on the attitude information and building height data acquired by the satellite, so that the reconstructed nDSM image contains the top and bottom offset information of the building caused by the attitude of the satellite acquisition.

[0011] 4) Match the location of the building top patch images with the reconstructed nDSM images to obtain the building height values ​​corresponding to each building top patch;

[0012] 5) Based on the building height value corresponding to the top patch of each building, and combined with the attitude information collected by the satellite, the top and bottom offset of the top patch of each building is corrected to obtain the bottom patch of each building.

[0013] Preferably, in step 3), the nDSM image is offset and reconstructed based on the satellite-acquired attitude information and building height data, specifically as follows:

[0014] 31) Based on the maximum height value of each pixel in the nDSM image, the satellite orbital altitude, and the satellite side tilt angle, the maximum offset distance of the image is estimated to be d pixels;

[0015] 32) Expand the nDSM image outward by d pixels in all directions (top, bottom, left, and right);

[0016] 33) Calculate the offset distance and offset direction of each pixel based on the height value of each pixel and the attitude information acquired by the satellite, and perform offset reconstruction on the expanded nDSM image.

[0017] Preferably, in step 31), the parameter d satisfies the following expression:

[0018]

[0019]

[0020] In the formula, H max Here, r is the maximum height value of each pixel in the nDSM image, r is the resolution of the nDSM image, int is the floor function, R is the average radius of the Earth, and h is the maximum height value of each pixel in the nDSM image. sat α is the satellite's orbital altitude, and α is the satellite's side yaw angle.

[0021] Preferably, in step 33), the offset distance of each pixel is calculated based on the height value of each pixel and the attitude information acquired by the satellite, specifically as follows:

[0022] The offset distance for each pixel is calculated using the following expression:

[0023]

[0024]

[0025] In the formula, m xi m is the horizontal offset distance of pixel i. yi β is the vertical offset distance of pixel i, β is the image azimuth angle, and h is the vertical offset distance of pixel i. i This is the height value of pixel i.

[0026] Preferably, in step 33), the offset direction of each pixel is calculated based on the height value of each pixel and the attitude information acquired by the satellite, specifically as follows:

[0027] When m xi When m > 0, shift to the left; when m xi When <0, shift to the right.

[0028] When m yi When m > 0, shift upwards; when m yi When <0, shift downwards.

[0029] Preferably, in step 5), based on the building height value corresponding to the top patch of each building, and combined with the attitude information acquired by the satellite, the position of the top patch of each building is corrected for top-to-bottom offset to obtain the bottom patch of each building, specifically:

[0030] 51) Obtain vector data of the top patches of each building, wherein the vector data includes a set of geographic coordinates of the top patch area of ​​each building;

[0031] 52) Calculate the distance and direction of the offset of each building's top patch based on the building height value corresponding to the top patch and the satellite-acquired attitude information;

[0032] 53) Convert the offset distance of the top patches of each building into geographic coordinate units, and offset the geographic coordinate set of the top patch area of ​​each building in combination with the offset direction. The vector data of the top patches of each building after offset is the vector data of the bottom patches of each building, thereby obtaining the bottom patches of each building.

[0033] Preferably, in step 52), the offset distance of each building's top patch is calculated based on the building height value corresponding to each building's top patch and the satellite-acquired attitude information. Specifically:

[0034]

[0035]

[0036] In the formula, dx k dy is the lateral offset distance of patch k on the top of the building. k h is the longitudinal offset distance of patch k on the top of the building. k This represents the building height value corresponding to the top patch k.

[0037] Preferably, in step 52), the direction of the offset of each building's top patch is calculated based on the building height value corresponding to each building's top patch and the satellite-acquired attitude information, specifically as follows:

[0038] When dx k When dx > 0, shift to the right. k When <0, shift to the left.

[0039] When dy k When dy > 0, shift downwards. k When <0, shift upwards.

[0040] Preferably, in step 4), the location of the building top patch image is matched with the reconstructed nDSM image to obtain the building height value corresponding to each building top patch, specifically:

[0041] The building height data set is obtained from the location area of ​​the top patch of the building based on the reconstructed nDSM image, and the median value in the building height data set is taken as the building height value corresponding to the top patch of the building.

[0042] The advantages of this invention compared to existing technologies are as follows: This invention acquires a difference image containing building height data from DSM and DEM images, and reconstructs the difference image by considering the influence of satellite acquisition attitude on image acquisition. Based on this, it obtains the building height data corresponding to the top patch area of ​​the building based on position matching. Furthermore, it performs top-to-bottom offset correction on the building top patch using satellite attitude to obtain the actual bottom patch outline of the building. This invention effectively improves the geographical location accuracy of buildings, eliminates the height offset of buildings caused by satellite acquisition attitude, avoids repeatedly extracting building patches from remote sensing images with different satellite acquisition attitudes, effectively reduces workload and improves work efficiency. Simultaneously, by combining information such as building outline position and height, it also provides a foundation for subsequent high-precision modeling and positioning of buildings, facilitating the subsequent application of building outlines and expanding the scope of application. Attached Figure Description

[0043] Figure 1 This is a flowchart of the automated top-bottom offset correction method for extracting building patches from remote sensing images according to the present invention;

[0044] Figure 2 This is a schematic diagram of the satellite acquisition attitude angle of the present invention;

[0045] Figure 3 This is a schematic diagram of nDSM image offset reconstruction according to an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram illustrating the top and bottom offset correction effect of a building patch according to an embodiment of the present invention. Detailed Implementation

[0047] The features and advantages of the present invention will become clearer and more apparent from the following detailed description.

[0048] An automated method for correcting top and bottom offsets in building extraction patches from remote sensing images is shown in the flowchart below. Figure 1 As shown, it includes the following steps:

[0049] (1) Obtain the original remote sensing image of the target area and extract the building patch image of the target area from it. In the building patch image, the patch outline of the building is its top outline.

[0050] Specifically, the remote sensing images are captured by the sensors of remote sensing satellites. Due to the changes in the flight attitude of the satellite during the on-orbit shooting process, the shooting angle of the sensor is often side-view, which causes image distortion. As a result, the outline of the building patch with height extracted from the remote sensing image is usually the top outline, which is offset from the actual geographical location of the bottom of the building.

[0051] Furthermore, the satellite acquisition attitude information during the acquisition of the original remote sensing images is obtained, including the satellite's orbital altitude h. sat The satellite side tilt angle α and the image azimuth angle β, wherein the image azimuth angle is the angle at which the image is rotated horizontally clockwise from due north during image imaging.

[0052] (2) Obtain DSM and DEM images of the target area with the same resolution as the building patch images of the target area, and perform a difference operation on the DSM and DEM images based on the pixel values ​​to obtain the nDSM image of the building height of the target area.

[0053] Specifically, the DSM image is a Digital Surface Model (DSM) image, automatically generated based on stereo image pairs; the DEM image is a Digital Elevation Model (DEM) image, which is publicly available. The nDSM image is a Normalized Digital Surface Model, a surface model that includes the true height of ground features to eliminate the influence of terrain.

[0054] Furthermore, if the acquired DSM and DEM images have different ranges and resolutions, the range of the images is cropped and the resolution is resampled to make the ranges and resolutions of the two images uniform. Then, the pixel values ​​at corresponding positions in the two images are subtracted to obtain the building height image of the target area. It is then determined whether the resolution of the building height image of the target area is the same as that of the building patch image. If they are not the same, resampling is performed to make its resolution the same as that of the building patch image of the target area.

[0055] Furthermore, the DSM image, DEM image, and nDSM image of the target area building height obtained through interpolation are raster images, with resolution represented by pixel values, such as 30*30 meters, and the pixel value of each pixel is the height value of that pixel.

[0056] (3) Reconstruct the building height image based on the satellite attitude information acquired during remote sensing image acquisition, so that it contains building offset information.

[0057] Specifically, the calculation will determine the offset caused by the satellite in the direction of building height during the image capture:

[0058] The angle of incidence at the time of image capture was calculated based on the orbital altitude and lateral tilt angle of the original remote sensing image. like Figure 2 As shown,

[0059]

[0060] R is the average radius of the Earth;

[0061] The number of rows and columns of the ground feature height image is obtained, denoted by row and col;

[0062] Obtain the maximum value Hmax from the pixel values ​​of the ground feature height image, based on the incident angle. And with resolution r, the maximum pixel distance of the overall offset of the ground feature height image is estimated:

[0063]

[0064] Resolution r is expressed in pixel values. For example, for an image with a pixel value of 30*30m, r = 30m.

[0065] Create a new elevation image of ground features with the following number of rows and columns: (row+2d, col+2d), which means that relative to the original elevation image, it expands outward by d pixels in all directions.

[0066] Based on the resolution, the height value of each pixel, and the attitude information acquired by the satellite, the lateral offset distance, vertical offset distance, and direction of movement of each pixel are obtained:

[0067]

[0068]

[0069] When m xi When m > 0, shift to the left; when m xi When <0, shift to the right.

[0070] When m yi When m > 0, shift upwards; when m yi When <0, shift downwards.

[0071] In the formula, m xi m is the horizontal offset distance of pixel i. yi β is the vertical offset distance of pixel i, β is the image azimuth angle, and h is the vertical offset distance of pixel i. i This is the height value of pixel i.

[0072] Figure 3 The image shows a comparison of the building height before and after reconstruction in a specific embodiment of the present invention. It can be seen that the reconstructed building height image has shifted to the lower right.

[0073] (4) Overlay the reconstructed building height image with the building patch image based on geographical location to obtain the building height information of each building patch area.

[0074] Specifically, the location and extent of the mask of the top patch of each building in the building patch image are obtained. Based on the location information of the mask of the top patch of the building, the building height dataset within the mask area is obtained in the building height image. The median algorithm is used to take the median height within the mask area of ​​the building as the height value of the building patch.

[0075] (5) Calculate the offset of the top of each building patch relative to the bottom based on the satellite acquisition attitude information and the height information of each building patch area during remote sensing image acquisition.

[0076] Specifically,

[0077]

[0078]

[0079] In the formula, dx k dy is the lateral offset distance of patch k on the top of the building. k h is the longitudinal offset distance of patch k on the top of the building. k This represents the building height value corresponding to the top patch k.

[0080] When dx k When dx > 0, shift to the right. k When <0, shift to the left.

[0081] When dy k When dy > 0, shift downwards. k When <0, shift upwards.

[0082] (6) Using the offset of the top of each building patch relative to the bottom, the building patch image is corrected to obtain the bottom outline of the building patch.

[0083] Specifically, converting the offset to geographic coordinate units yields the geographic coordinate offset of the building outline vector, d′. x ,d′ y The direction of movement: when d′ x When d′ is positive, move to the right; when d′ is negative, move to the left. y When the value is positive, the map moves downwards; when the value is negative, the map moves upwards, thus obtaining the corrected building outline, which is the actual bottom outline of the building.

[0084] like Figure 4 As shown, in a specific embodiment of the present invention, the corrected outline of the actual bottom patch of the building is shifted to the upper left relative to the top outline directly extracted from the remote sensing image.

[0085] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. An automatic top-bottom offset correction method for building extraction polygon of remote sensing image, characterized in that, The method comprises the following steps: 1) acquiring satellite remote sensing images of a target area, acquiring satellite collection posture information at the time of remote sensing image collection, including satellite orbit height, satellite side swing angle and image azimuth; extracting building top polygon image of the target area from the satellite remote sensing images; 2) acquiring DSM images and DEM images of the target area with the same resolution as the building top polygon image of the target area, performing difference operation on the DSM images and DEM images based on pixel values, and acquiring nDSM images containing building height data of the target area; 3) performing offset reconstruction on the nDSM images according to the satellite collection posture information and the building height data, so that the reconstructed nDSM images have top-bottom offset information of buildings caused by the satellite collection posture; 4) performing position matching on the building top polygon image and the reconstructed nDSM images, and acquiring building height values corresponding to each building top polygon; 5) performing top-bottom offset correction on the position of each building top polygon according to the building height values corresponding to each building top polygon and combining the satellite collection posture information, and acquiring the bottom polygon of each building. In the step 3), the offset reconstruction on the nDSM images according to the satellite collection posture information and the building height data is specifically as follows: 31) The maximum offset distance of the image from the maximum height value of each pixel in the nDSM image, the satellite orbit height, and the satellite side-slip angle is estimated as one pixel; 32) Extend nDSM image up, down, left, right respectively outwards pixels; 33) calculating the offset distance and offset direction of each pixel according to the height value of each pixel and the satellite collection posture information, and performing offset reconstruction on the expanded nDSM images; In the step 31), the parameters satisfy the following expression: wherein, is the maximum value of the height values of the pixels in the nDSM image, is the resolution of the nDSM image, is the incidence angle, is the rounding function, is the mean radius of the earth, is the satellite orbit height, is the satellite side-slip angle.

2. The automatic top-bottom offset correction method for building extraction polygon of remote sensing image according to claim 1, characterized in that, In the step 33), the offset distance of each pixel is calculated according to the height value of each pixel and the satellite collection posture information, and is specifically as follows: The offset distance of each pixel is calculated by the following expression: wherein is a pixel is a distance of the lateral offset, is a pixel is a distance of the longitudinal offset, is an image azimuth angle, is a pixel is a height value.

3. The method of claim 2, wherein the method further comprises: determining a top-bottom offset of the building footprint; and correcting the top-bottom offset of the building footprint. In the step 33), the offset direction of each pixel is calculated according to the height value of each pixel and the satellite collection posture information, and is specifically as follows: When > 0, offset to the left, when < 0, offset to the right, When > 0, offset up, when < 0, offset down.

4. The automatic top-bottom offset correction method for building extraction polygon of remote sensing image according to claim 3, characterized in that, In the step 5), the bottom polygon of each building is acquired by performing top-bottom offset correction on the position of each building top polygon according to the building height values corresponding to each building top polygon and combining the satellite collection posture information, and is specifically as follows: 51) acquiring vector data of each building top polygon, wherein the vector data comprises a geographical coordinate set of each building top polygon area; 52) calculating the offset distance and direction of each building top polygon according to the building height values corresponding to each building top polygon and the satellite collection posture information; 53) converting the offset distance of each building top polygon into geographical coordinate units, and offsetting the geographical coordinate set of each building top polygon area in combination with the offset direction, so that the vector data of the offset building top polygon is the vector data of the bottom polygon of each building, thereby acquiring the bottom polygon of each building.

5. The automatic top-bottom offset correction method for building extraction polygon of remote sensing image according to claim 4, characterized in that, In the step 52), the offset distance of each building top polygon is calculated according to the building height values corresponding to each building top polygon and the satellite collection posture information, and is specifically as follows: wherein is the lateral offset distance of the building top footprint is the longitudinal offset distance of the building top footprint is the lateral offset distance of the building top footprint is the longitudinal offset distance of the building top footprint is the lateral offset distance of the building top footprint is the corresponding building height value.

6. The automatic top-bottom offset correction method for building extraction polygon of remote sensing image according to claim 5, characterized in that, In the step 52), the offset direction of each building top polygon is calculated according to the building height values corresponding to each building top polygon and the satellite collection posture information, and is specifically as follows: When > 0, offset to the right, when < 0, offset to the left, When > 0, offset down, when < 0, offset up.

7. An automated top-bottom offset correction method for extracting building patches from remote sensing images according to any one of claims 1 to 6, characterized in that, In the step 4), the building top patch image is positionally matched with the reconstructed nDSM image to obtain a building height value corresponding to each building top patch, specifically as follows: According to the reconstructed nDSM image, a building height data set in a position region of a building top patch is obtained, and a median value in the building height data set is taken as a building height value corresponding to the building top patch.

Images