Address correction method and system based on computer vision and remote sensing image

Abstract

The invention relates to the technical field of computer vision, in particular to an address correction method and system based on computer vision and remote sensing images. The method comprises the following steps: acquiring a remote sensing image of a target area, and performing WGS84 coordinate system conversion; according to the invention, gradients of pixels in horizontal and vertical directions are respectively calculated through a preprocessed remote sensing image to obtain edge information of the remote sensing image after non-maximum suppression, and the remote sensing image after non-maximum suppression is used to determine whether the pixel points are strong edge points or weak edge points to form an image contour set. Then the real-time building contour and the image building contour are judged to belong to the same building, and the text address of the corresponding building is matched by utilizing the building and the address text library, so that the text address can correspond to the actual building more accurately, and the matching error caused by the position drift of the building is avoided; the problem that the character address and the image position are disjointed is effectively solved, and therefore the accuracy of address matching is improved.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and more specifically, to an address correction method and system based on computer vision and remote sensing images. Background Technology

[0002] In today's society, accurate address location plays a crucial role in the efficient operation of many fields, especially in food delivery and emergency rescue scenarios. Accurate address information can significantly improve service efficiency and protect life and property safety. However, the address location problem in urban-rural fringe areas has become a major obstacle to achieving this goal.

[0003] The urban-rural fringe areas have a serious problem of "house number drift". Due to the large number of illegal buildings and temporary structures in the area, the deviation between the physical house number and the building location in the remote sensing image exceeds 5 meters. This phenomenon makes the correspondence between the text address and the actual building extremely confusing, which brings great difficulties to address positioning.

[0004] Traditional address matching technologies are inadequate in addressing this problem. They rely solely on the simple overlay of text address databases and low-precision images, lacking dynamic correlation with the actual building outlines. This two-dimensional matching mode cannot accurately reflect the actual location and shape of buildings, resulting in an address positioning accuracy rate of less than 60% in the complex building environments of urban-rural fringe areas. In practical applications, food delivery drivers often waste a lot of time searching for customers due to inaccurate address positioning, and emergency rescue personnel may also delay rescue opportunities due to their inability to reach the scene in a timely and accurate manner, causing many inconveniences and potential risks to people's lives and safety. Therefore, we provide an address correction method and system based on computer vision and remote sensing imagery. Summary of the Invention

[0005] The purpose of this invention is to provide an address correction method and system based on computer vision and remote sensing images to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, one objective of this invention is to provide an address correction method based on computer vision and remote sensing imagery, comprising the following steps:

[0007] S1. Acquire remote sensing images, address text library, and building outline data of the target area, perform WGS84 coordinate system transformation, integrate the building outline data into a real-time outline set, and obtain the pre-processed remote sensing images.

[0008] S2. The pre-processed remote sensing image is smoothed to obtain the denoised remote sensing image data. The gradient of the pixel in the horizontal and vertical directions is calculated to obtain the edge information of the remote sensing image after non-maximum suppression.

[0009] The remote sensing image after non-maximum suppression is used to determine whether a pixel is a strong edge point or a weak edge point. The strong edge points or weak edge points form an image contour set. Then, it is determined whether the real-time building contour in the real-time contour set and the image building contour in the image contour set belong to the same building.

[0010] Then, the text address of the corresponding building is matched using the building and address text database, and the matching results are integrated into a preliminary matching result set.

[0011] S3. Calculate the Euclidean distance based on the preliminary matching result set, use the Euclidean distance to determine whether there is a door number drift defect in the text address, generate the corrected text address coordinates, calculate the corresponding door number, and integrate the corrected text address coordinates and the corresponding door number into the corrected text address information.

[0012] As a further improvement to this technical solution, step S1 includes the following method steps:

[0013] Remote sensing images and address text databases of the target area are obtained through a GIS platform. Then, computer vision technology is used to collect real-time outline data of buildings in the target area, obtain the collected building outline data, record the building outline coordinates, and use the building outline coordinates as the actual coordinates.

[0014] The collected building outline data is integrated into a real-time outline set, and the coordinates of the building outline are extracted using an edge detection algorithm.

[0015] As a further improvement to this technical solution, step S1 also includes the following method steps:

[0016] The remote sensing images and collected building outline data of the target area are converted to the WGS84 coordinate system. Then the address text library of the target area is processed to remove duplicate and invalid records, resulting in a duplicate-free address text library.

[0017] The transformed remote sensing imagery, building outlines in the WGS84 coordinate system, and a duplicate-removed address text library are associated to form a dataset containing image-text-outline data. In this dataset, the remote sensing imagery data that has undergone coordinate system transformation is extracted from the image-text-outline dataset, which is the preprocessed remote sensing imagery.

[0018] As a further improvement to this technical solution, step S2 includes the following method steps:

[0019] S2.1. Obtain pixel 1 from the preprocessed remote sensing image at any time, establish a local coordinate system with pixel 1 as the center, take pixel 1 as the center pixel, and then obtain the offset of the surrounding pixels relative to the center pixel to smooth the preprocessed remote sensing image and obtain the denoised remote sensing image data.

[0020] S2.2 Set low and high thresholds, extract pixel gray values ​​from the edge information of the remote sensing image after non-maximum suppression, and use the pixel gray values ​​and high thresholds to determine whether the pixel is a strong edge point or a weak edge point.

[0021] When the gray value of a pixel is greater than the high threshold, the pixel is a strong edge point; when the gray value of a pixel is between the high threshold and the low threshold, the pixel is a weak edge point.

[0022] An image contour set is formed by using strong edge points and weak edge points.

[0023] As a further improvement to this technical solution, S2.1 also includes the following method steps:

[0024] Extract the coordinates of pixel 2 from the denoised remote sensing image data and calculate the gradient of pixel 2 in the horizontal and vertical directions respectively;

[0025] Based on the gradient of pixel 2 in the horizontal and vertical directions, the local maximum pixel is retained, and non-edge pixels are removed to obtain the edge information of the remote sensing image after non-maximum suppression.

[0026] As a further improvement to this technical solution, step S2.2 also includes the following method steps:

[0027] Extract real-time building contours and image building contours from the real-time contour set and the image contour set respectively, calculate the intersection-union ratio between the contours, and use the intersection-union ratio between the contours to determine whether the real-time building contour and the image building contour belong to the same building.

[0028] The text address is bound to the real-time building outline and the image building outline to obtain the binding result. The matching result and the binding result are integrated into a preliminary matching result set.

[0029] As a further improvement to this technical solution, step S3 includes the following method steps:

[0030] Semantic parsing is performed on the text addresses in the preliminary matching result set to extract key semantic information such as road names and house number ranges from the text addresses;

[0031] Road centerline data is extracted from the database in the GIS platform, and the theoretical location of the text address is calculated based on the house number range and road centerline data.

[0032] Based on the theoretical location of the text address, coordinates are linearly assigned in order of house number. The coordinates are taken as theoretical coordinates, and Euclidean distance is calculated to determine whether there is a house number drift defect in the text address.

[0033] As a further improvement to this technical solution, step S3 also includes the following method steps:

[0034] The actual coordinates are used to determine the projected position of the actual coordinates on the road centerline data.

[0035] The spatial interpolation correction algorithm is used to generate the corrected text address coordinates based on the projection position, and then the corresponding house number is calculated based on the projection position.

[0036] The corrected text address coordinates and corresponding house numbers are integrated into the corrected text address information, and then updated based on the corrected text address information.

[0037] The second objective of this invention is to provide a system for operating an address correction method based on computer vision and remote sensing images, including any one of the above, comprising a conversion processing unit, an edge matching unit, and a drift update unit.

[0038] The conversion processing unit acquires remote sensing images, address text library, and building outline data of the target area, performs WGS84 coordinate system conversion, and then integrates the building outline data into a real-time outline set to obtain the pre-processed remote sensing image.

[0039] The edge matching unit is used to receive the pre-processed remote sensing image from the conversion processing unit, perform smoothing processing, and then calculate the gradient of the pixel in the horizontal and vertical directions to obtain the edge information of the remote sensing image after non-maximum suppression.

[0040] The remote sensing images after non-maximum suppression are used to form an image contour set, and then it is determined that the real-time building contour in the real-time contour set and the image building contour in the image contour set belong to the same building.

[0041] By using a building and address text database to match the text address of the corresponding building, the matching results are integrated into a preliminary matching result set.

[0042] The drift update unit is used to receive the preliminary matching result set from the edge matching unit, calculate the Euclidean distance, use the Euclidean distance to determine whether there is a door number drift defect in the text address, generate the corrected text address coordinates, calculate the corresponding door number, and integrate the corrected text address coordinates and the corresponding door number into the corrected text address information.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This address correction method and system based on computer vision and remote sensing imagery involves smoothing pre-processed remote sensing images to obtain denoised remote sensing image data. Gradients of pixels in the horizontal and vertical directions are calculated to obtain edge information of the remote sensing image after non-maximum suppression (NMS). The NMS-suppressed remote sensing image is used to determine whether pixels are strong or weak edge points. Image contour sets are formed using these strong or weak edge points. It is then determined that the real-time building contours in the real-time contour set and the image building contours in the image contour set belong to the same building. The corresponding building's text address is then matched using a building and address text library. The matching results are integrated into a preliminary matching result set. This preliminary matching result set, by introducing physical features of the building contours, establishes a three-dimensional association between text address, building contour, and remote sensing imagery. This breaks through the traditional two-dimensional text-image matching mode, enabling more accurate mapping of text addresses to actual buildings and avoiding matching errors caused by building location drift. It effectively solves the problem of text address and image location disconnection, thereby improving the accuracy of address matching. Attached Figure Description

[0045] Figure 1 This is an overall block diagram of the present invention;

[0046] Figure 2 This is a block diagram of the present invention.

[0047] The meanings of the labels in the diagram are as follows:

[0048] 1. Conversion processing unit; 2. Edge matching unit; 3. Drift update unit. Detailed Implementation

[0049] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Example 1

[0051] Please see Figures 1-2 As shown, one of the objectives of this embodiment is to provide an address correction method based on computer vision and remote sensing imagery, including the following method steps:

[0052] S1. Acquire remote sensing images, address text library, and building outline data of the target area, perform WGS84 coordinate system transformation, integrate the building outline data into a real-time outline set, and obtain the pre-processed remote sensing images.

[0053] S1 includes the following method steps:

[0054] Remote sensing imagery and address database (including house numbers, road names, and administrative division codes) of the target area were acquired from a GIS platform. The remote sensing imagery had a resolution of 0.5 meters and was updated every 15 days. Computer vision technology was then used to collect real-time building contour data within the target area at a sampling density of one sampling point per 100 square meters. The collected building contour data was then recorded, and the building contour coordinates were used as the actual coordinates P. real The collected building outline data is integrated into a real-time outline set R, and the coordinates of the building frame are extracted using an edge detection algorithm.

[0055] The remote sensing imagery and acquired building outline data of the target area are converted to the WGS84 coordinate system. The coordinate conversion operation includes the following steps:

[0056] First, determine the coordinate system used for the remote sensing images and collected building outline data of the target area. Then, use ArcGIS software as a conversion tool to import the remote sensing images and collected building outline data of the target area into ArcGIS software.

[0057] In ArcGIS software, find the function option to set the coordinate system, set the target coordinate system to the WGS84 coordinate system, execute the coordinate system transformation command, and after the transformation operation is completed, the transformed remote sensing image and building outline in WGS84 coordinate system will be obtained. Save the transformed remote sensing image and building outline in WGS84 coordinate system to the specified location, and confirm whether the coordinate system has been successfully converted to WGS84.

[0058] When the converted remote sensing image and building outline WGS84 coordinate system are obtained, the address text library of the target area is processed to remove duplicate and invalid records, such as "No house number on XX Road", resulting in a duplicate-free address text library. The converted remote sensing image, building outline WGS84 coordinate system, and duplicate-free address text library are then associated to form a dataset containing image-text-outline. In this dataset, the remote sensing image of the target area has been preprocessed. The preprocessed (coordinate system transformed) remote sensing image data is extracted from the image-text-outline dataset, which is the final preprocessed remote sensing image required.

[0059] S2. The pre-processed remote sensing image is smoothed to obtain the denoised remote sensing image data. The gradient of the pixel in the horizontal and vertical directions is calculated to obtain the edge information of the remote sensing image after non-maximum suppression.

[0060] The remote sensing image after non-maximum suppression is used to determine whether a pixel is a strong edge point or a weak edge point. The strong edge points or weak edge points form an image contour set. Then, it is determined whether the real-time building contour in the real-time contour set and the image building contour in the image contour set belong to the same building.

[0061] Then, the text address of the corresponding building is matched using the building and address text database, and the matching results are integrated into a preliminary matching result set.

[0062] S2 includes the following method steps:

[0063] S2.1. Acquire pixel 1 from the preprocessed remote sensing image at any time, establish a local coordinate system with pixel 1 as the center pixel, and then obtain the offset x of the surrounding pixels relative to the center pixel. p and y p Here, p has no special physical meaning and is only used for differentiation. Gaussian filtering is used to smooth the preprocessed remote sensing image based on the offsets x and y of surrounding pixels relative to the center pixel, resulting in the denoised remote sensing image data G. The specific algorithm formula is as follows: This formula is optimized for the blurred edge characteristics of buildings in urban-rural fringe areas. It can effectively smooth the noise in pre-processed remote sensing images while preserving the detailed information of building edges.

[0064] When the denoised remote sensing image data is obtained, the coordinates (x, y) of pixel 2 are extracted from the denoised remote sensing image data, and the gradient Td of pixel 2 in the horizontal direction is calculated using image functions based on the coordinates (x, y) of pixel 2. x = f(x+1,y)-f(x-1,y) and the gradient Td of pixel 2 in the vertical direction y =f(x,y+1)-f(x,y-1), by calculating the gradient, the rate of change of pixels in the denoised image data can be determined, thereby determining the direction of the edge;

[0065] Based on the gradient of pixel 2 in the horizontal and vertical directions, local maximum pixels are retained and non-edge pixels are removed to obtain the edge information of the remote sensing image after non-maximum suppression. Through gradient edge refinement, only pixels that truly represent the edge are retained to form fine edges.

[0066] S2.2 Set a low threshold T1 = 50 and a high threshold T2 = 150. The high and low thresholds will be used to distinguish between strong edges and weak edges. Pixels above the high threshold T2 are considered strong edge points, pixels below the low threshold T2 are considered non-edge points, and pixels between the high and low thresholds are weak edge points. Then, extract the pixel gray values ​​from the edge information of the remote sensing image after non-maximum suppression, and use the pixel gray values ​​and the high threshold to determine whether the pixel is a strong edge point or a weak edge point.

[0067] When the gray value of a pixel is greater than the high threshold, the pixel is a strong edge point. Connectivity analysis methods (such as 4-connectivity or 8-connectivity) are used to connect strong edge points into continuous strong edges. Here, 4-connectivity means that a pixel is connected to its four adjacent pixels above, below, left, and right. 8-connectivity means that a pixel is connected to its eight adjacent pixels above, below, left, right, upper left, upper right, lower left, and lower right.

[0068] When the gray value of a pixel is between the high threshold and the low threshold, the pixel is a weak edge point. Breadth-first search is used to check whether the weak edge point is connected to the strong edge point of the continuous strong edge. When the weak edge point is connected to the strong edge point of the continuous strong edge, the connected strong edge and the weak edge point connected to the strong edge point of the continuous strong edge are combined together to form the image contour set S.

[0069] Extract the i-th real-time building contour R from the real-time contour set R. i Then, extract the i-th image building outline S from the image outline set S. i Through real-time building outline R i And the architectural outline of the image S i Calculate the crossover ratio between contours Here, "area" refers to the area. The intersection-union ratio (IUU) threshold is set to 0.7 (the IUU threshold is determined through training with 1000 sets of urban-rural fringe building samples). The IUU and the IUU threshold are used to determine whether the real-time building outline and the image building outline belong to the same building. When the IUU between the outlines is greater than or equal to the IUU threshold, it is determined that the real-time building outline and the image building outline belong to the same building. Then, the building and address text library are matched to match the text address of the corresponding building and obtain the matching result. The text address is bound to the real-time building outline and the image building outline to obtain the binding result. The matching result and the binding result are integrated into a preliminary matching result set.

[0070] S3. Calculate the Euclidean distance based on the preliminary matching result set, use the Euclidean distance to determine whether there is a door number drift defect in the text address, then generate the corrected text address coordinates, calculate the corresponding door number, and integrate the corrected text address coordinates and the corresponding door number into the corrected text address information.

[0071] S3 includes the following method steps:

[0072] The pre-trained language model of Transformer, combined with a bidirectional long short-term memory network, is used to perform semantic parsing on the text addresses in the preliminary matching result set. Key semantic information such as road name, house number range, and building number in the text address is extracted. For example, XX Road No. 100-120 Building 3 is parsed as road name XX Road, house number range [100,120], and building number 3.

[0073] Road centerline data is extracted from the database in the GIS platform. Based on the house number range and the road centerline data, the theoretical location of the text address is calculated. For example, 100-120 on XX Road corresponds to 5 meters next to the road centerline. Then, coordinates are linearly assigned according to the house number order based on the theoretical location of the text address, and the coordinates are marked as theoretical coordinates P. theo Through actual coordinates P real and theoretical coordinates P theo Calculate Euclidean distance Then set a distance threshold and use Euclidean distance and distance threshold to determine whether there is a door number drift defect in the text address. When the Euclidean distance is greater than the distance threshold, it is determined that there is a door number drift defect in the text address.

[0074] When a address card drift defect exists, the actual coordinates are projected onto the road centerline using data on the actual coordinates. A spatial interpolation correction algorithm is then used to generate the corrected text address coordinates P based on this projection position. corr Then, the corresponding house number is calculated based on the projection position. For example, if the original match for XX Road is No. 110, but the actual location is closer to No. 115, it is corrected to No. 115 on XX Road. The corrected text address coordinates and the corresponding house number are integrated into the corrected text address information, and the information is updated based on the corrected text address information.

[0075] The second objective of this invention is to provide a system for operating the above-mentioned address correction method based on computer vision and remote sensing images, including a conversion processing unit 1, an edge matching unit 2, and a drift update unit 3.

[0076] The conversion processing unit 1 acquires remote sensing images, address text library, and building outline data of the target area, performs WGS84 coordinate system conversion, and then integrates the building outline data into a real-time outline set to obtain the pre-processed remote sensing image.

[0077] The edge matching unit 2 is used to receive the pre-processed remote sensing image in the conversion processing unit 1, perform smoothing processing, and then calculate the gradient of the pixel in the horizontal and vertical directions to obtain the edge information of the remote sensing image after non-maximum suppression.

[0078] The remote sensing images after non-maximum suppression are used to form an image contour set, and then it is determined that the real-time building contour in the real-time contour set and the image building contour in the image contour set belong to the same building.

[0079] By using a building and address text database to match the text address of the corresponding building, the matching results are integrated into a preliminary matching result set.

[0080] The drift update unit 3 is used to receive the preliminary matching result set from the edge matching unit 2, calculate the Euclidean distance, use the Euclidean distance to determine whether there is a door number drift defect in the text address, generate the corrected text address coordinates, calculate the corresponding door number, and integrate the corrected text address coordinates and the corresponding door number into the corrected text address information.

[0081] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for address correction based on computer vision and remote sensing imagery, characterized in that: The methods and steps include the following: S1. Acquire remote sensing images, address text library, and building outline data of the target area, perform WGS84 coordinate system transformation, integrate the building outline data into a real-time outline set, and obtain the pre-processed remote sensing images. S2. The pre-processed remote sensing image is smoothed to obtain the denoised remote sensing image data. The gradient of the pixel in the horizontal and vertical directions is calculated to obtain the edge information of the remote sensing image after non-maximum suppression. The remote sensing image after non-maximum suppression is used to determine whether a pixel is a strong edge point or a weak edge point. The strong edge points or weak edge points form an image contour set. Then, it is determined whether the real-time building contour in the real-time contour set and the image building contour in the image contour set belong to the same building. Then, the text address of the corresponding building is matched using the building and address text database, and the matching results are integrated into a preliminary matching result set. S3. Calculate the Euclidean distance based on the preliminary matching result set, use the Euclidean distance to determine whether there is a door number drift defect in the text address, generate the corrected text address coordinates, calculate the corresponding door number, and integrate the corrected text address coordinates and the corresponding door number into the corrected text address information.

2. The address correction method based on computer vision and remote sensing imagery according to claim 1, characterized in that: S1 includes the following method steps: Remote sensing images and address text databases of the target area are obtained through a GIS platform. Then, computer vision technology is used to collect real-time outline data of buildings in the target area, obtain the collected building outline data, record the building outline coordinates, and use the building outline coordinates as the actual coordinates. The collected building outline data is integrated into a real-time outline set, and the coordinates of the building outline are extracted using an edge detection algorithm.

3. The address correction method based on computer vision and remote sensing imagery according to claim 2, characterized in that: S1 further includes the following method steps: The remote sensing images and collected building outline data of the target area are converted to the WGS84 coordinate system. Then the address text library of the target area is processed to remove duplicate and invalid records, resulting in a duplicate-free address text library. The transformed remote sensing imagery, building outlines in the WGS84 coordinate system, and a duplicate-removed address text library are associated to form a dataset containing image-text-outline data. In this dataset, the remote sensing imagery data that has undergone coordinate system transformation is extracted from the image-text-outline dataset, which is the preprocessed remote sensing imagery.

4. The address correction method based on computer vision and remote sensing imagery according to claim 3, characterized in that: S2 includes the following method steps: S2.

1. Obtain pixel 1 from the preprocessed remote sensing image at any time, establish a local coordinate system with pixel 1 as the center, take pixel 1 as the center pixel, and then obtain the offset of the surrounding pixels relative to the center pixel to smooth the preprocessed remote sensing image and obtain the denoised remote sensing image data. S2.2 Set low and high thresholds, extract pixel gray values ​​from the edge information of the remote sensing image after non-maximum suppression, and use the pixel gray values ​​and high thresholds to determine whether the pixel is a strong edge point or a weak edge point. When the gray value of a pixel is greater than the high threshold, the pixel is a strong edge point; when the gray value of a pixel is between the high threshold and the low threshold, the pixel is a weak edge point. An image contour set is formed by using strong edge points and weak edge points.

5. The address correction method based on computer vision and remote sensing imagery according to claim 4, characterized in that: S2.1 further includes the following method steps: Extract the coordinates of pixel 2 from the denoised remote sensing image data and calculate the gradient of pixel 2 in the horizontal and vertical directions respectively; Based on the gradient of pixel 2 in the horizontal and vertical directions, the local maximum pixel is retained, and non-edge pixels are removed to obtain the edge information of the remote sensing image after non-maximum suppression.

6. The address correction method based on computer vision and remote sensing imagery according to claim 5, characterized in that: S2.2 further includes the following method steps: Extract real-time building contours and image building contours from the real-time contour set and the image contour set respectively, calculate the intersection-union ratio between the contours, and use the intersection-union ratio between the contours to determine whether the real-time building contour and the image building contour belong to the same building. The text address is bound to the real-time building outline and the image building outline to obtain the binding result. The matching result and the binding result are integrated into a preliminary matching result set.

7. The address correction method based on computer vision and remote sensing imagery according to claim 6, characterized in that: S3 includes the following method steps: Semantic parsing is performed on the text addresses in the preliminary matching result set to extract key semantic information such as road names and house number ranges from the text addresses; Road centerline data is extracted from the database in the GIS platform, and the theoretical location of the text address is calculated based on the house number range and road centerline data. Based on the theoretical location of the text address, coordinates are linearly assigned in order of house number. The coordinates are taken as theoretical coordinates, and Euclidean distance is calculated to determine whether there is a house number drift defect in the text address.

8. The address correction method based on computer vision and remote sensing imagery according to claim 7, characterized in that: S3 further includes the following method steps: The actual coordinates are used to determine the projected position of the actual coordinates on the road centerline data. The spatial interpolation correction algorithm is used to generate the corrected text address coordinates based on the projection position, and then the corresponding house number is calculated based on the projection position. The corrected text address coordinates and corresponding house numbers are integrated into the corrected text address information, and then updated based on the corrected text address information.

9. A system for operating the address correction method based on computer vision and remote sensing imagery according to any one of claims 1-8, characterized in that: It includes a conversion processing unit (1), an edge matching unit (2), and a drift update unit (3); The conversion processing unit (1) acquires remote sensing images, address text library and building outline data of the target area, performs WGS84 coordinate system conversion, and then integrates the building outline data into a real-time outline set to obtain the pre-processed remote sensing image. The edge matching unit (2) is used to receive the pre-processed remote sensing image in the conversion processing unit (1), perform smoothing processing, and then calculate the gradient of the pixel in the horizontal and vertical directions to obtain the edge information of the remote sensing image after non-maximum suppression. The remote sensing images after non-maximum suppression are used to form an image contour set, and then it is determined that the real-time building contour in the real-time contour set and the image building contour in the image contour set belong to the same building. By using a building and address text database to match the text address of the corresponding building, the matching results are integrated into a preliminary matching result set. The drift update unit (3) is used to receive the preliminary matching result set in the edge matching unit (2) to calculate the Euclidean distance, use the Euclidean distance to determine whether there is a door number drift defect in the text address, generate the corrected text address coordinates, calculate the corresponding door number, and integrate the corrected text address coordinates and the corresponding door number into the corrected text address information.

Images