A high-precision remote sensing mapping image processing method, system and device
By clustering and extracting features from remote sensing images, and combining edge, line, corner, and elevation information, the problem of identification in the surveying of complex buildings was solved, and high-precision real estate surveying was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MAOMAO (NANTONG) INTELLIGENT TECH CO LTD
- Filing Date
- 2025-12-17
- Publication Date
- 2026-08-04
AI Technical Summary
Existing remote sensing technology suffers from insufficient efficiency and accuracy in real estate surveying, especially when surveying complex buildings. This is mainly due to the differences in features caused by different building materials and occlusion phenomena, which lead to blurred textures. Existing geometric shape and texture feature recognition methods are unable to accurately identify building areas.
By clustering remote sensing images, edge features and line information are extracted. Combined with corner and elevation information, surface feature coefficients and feature optimization coefficients are calculated. Clustering algorithms are used to identify building clusters, and the mapping results are obtained by combining the location distribution of image blocks.
It improves the accuracy of building identification under the influence of different materials and background contrast, enhances the robustness and precision of surveying and mapping, and realizes high-precision surveying and mapping of real estate buildings.
Smart Images
Figure CN121708481B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of real estate surveying and mapping technology, specifically to a method, system, and device for processing high-precision remote sensing surveying and mapping images. Background Technology
[0002] With the acceleration of urbanization, the demand for real estate information is increasing, and the requirements for real estate surveying are gradually moving towards higher efficiency and accuracy. Remote sensing technology, with its advantages of wide coverage and convenient data acquisition, has been widely used in real estate surveying. By acquiring surface images through sensors mounted on satellites or drones and using image processing techniques to extract features of buildings and other land features, it provides crucial data support for urban planning, land management, and environmental protection. However, as application scenarios continue to expand, existing technologies based on geometric shape or texture features are gradually revealing their limitations, especially when dealing with the more complex surveying of buildings in real estate, where their surveying efficiency and accuracy are further challenged.
[0003] Existing methods primarily rely on the geometry and surface texture of buildings in remote sensing images for identification and analysis. However, due to the wide variety of building materials and the significant differences in their appearance in remote sensing images, different parts of the same building may exhibit different characteristics, increasing the difficulty of identification. When buildings of various materials are mixed, occlusion between buildings and large contrast differences between different materials can lead to blurred textures in some buildings. In such cases, using geometry and texture to identify building areas can easily result in the failure to identify some buildings, thus affecting the accuracy of building mapping. Summary of the Invention
[0004] In view of the above, it is necessary to provide a method, system and device for processing high-precision remote sensing images to solve the above problems.
[0005] According to one aspect of this application, a method for processing high-precision remote sensing mapping images is provided, the method comprising: Acquire remote sensing images of real estate surveying; Cluster all pixels in the remote sensing image, and obtain all image patches for each cluster based on the distribution of pixels in the cluster; obtain several regular graphic regions based on the edge features in the remote sensing image; extract the straight line information in the remote sensing image; determine the surface feature coefficients of each image patch based on the distribution of edge pixels and the number of straight lines in each image patch, combined with the overlap between the regular image regions and each image patch. Corner points are extracted from remote sensing images. Based on the position coordinates of corner points in each image patch, corner point sequences in various dimensions are obtained. Based on the randomness of the distribution of all corner points in each image patch, combined with the elevation information in the image patch and the corner point sequences in various dimensions, feature optimization coefficients for each image patch are obtained. Based on the surface feature coefficients and the feature optimization coefficients, all image blocks are clustered to obtain building clusters, non-building clusters, and suspected building clusters. Combined with the location distribution of the image blocks, the mapping results of each image block are obtained.
[0006] Specifically, obtaining all image patches for each cluster involves: Extract the binary mask image of each cluster, and use each connected component in each binary mask image as an image block of the corresponding cluster.
[0007] The regular graphic region is determined by a chain code algorithm.
[0008] Specifically, determining the surface feature coefficients of each image block involves: Calculate the percentage of the area of each image block that overlaps with all regular image regions; Each edge pixel and non-edge pixel in each image block is labeled with a different value, and the randomness of the label values of all pixels in each image block is measured. The surface feature coefficient is positively correlated with the area ratio and the number of straight lines in each image block, and negatively correlated with the randomness measure.
[0009] Specifically, obtaining the corner point sequence in each dimension involves: The x-coordinates of all corner points in each image patch form the x-dimensional corner point sequence; the y-coordinates of all corner points in each image patch form the y-dimensional corner point sequence.
[0010] The specific process for obtaining the feature optimization coefficients for each image block is as follows: The elevation information of all pixels in each image block is used to form the elevation sequence of each image block; the average similarity measure between the elevation sequences of each image block and other image blocks in the same cluster is calculated. For each image patch, the minimum skewness of the corner point sequence across all dimensions is calculated and positively fused with the average similarity metric to obtain the first result; The absolute value of the Moran index of the elevation values at all corner points in each image patch is positively fused with the dispersion of the elements in the elevation sequence of the corresponding image patch to obtain the second result; The negative correlation mapping of the first result is positively fused with the second result to obtain the feature optimization coefficients for each image patch.
[0011] Specifically, the building cluster is the cluster with the largest sum of the normalized mean of the surface feature coefficients and the normalized mean of the feature optimization coefficients of all image blocks; the non-building cluster is the cluster with the smallest sum of the normalized mean of the surface feature coefficients and the normalized mean of the feature optimization coefficients of all image blocks; and the suspected building cluster is all clusters other than the building cluster and the non-building cluster.
[0012] The process of obtaining the mapping results for each image patch includes: All image patches in the building cluster are identified as building image patches; all image patches in the non-building image patch are identified as non-building image patches. For image patches in a suspected building cluster, extract the image patch with the most mapped images in its neighborhood and denote it as the target image patch. If all mapped images in the target image patch's neighborhood are building image patches, then the target image patch is a building image patch. If all mapped images in the target image patch's neighborhood are non-building image patches, then the target image patch is a non-building image patch. Otherwise, use the distance metric between the target image patch and the feature vectors of all building image patches in its neighborhood as the first metric; use the distance metric between the target image patch and the feature vectors of all non-building image patches in its neighborhood as the second metric. If the first metric is less than the second metric, the target image patch is a building image patch; if the first metric is greater than or equal to the second metric, the target image patch is a non-building image patch. The feature vector of the image patch is determined by the surface feature coefficient and the feature optimization coefficient.
[0013] According to another aspect of this application, a high-precision remote sensing image processing apparatus is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.
[0014] According to another aspect of this application, a high-precision remote sensing mapping image processing system is provided, wherein the system stores a computer program that, when executed by a processor, implements any of the methods described above.
[0015] This application has at least the following beneficial effects: This application considers improving mapping accuracy under the influence of different building materials and background contrast. It goes beyond relying solely on geometric and textural features within a region; it requires a comprehensive judgment combining relevant features with global comparison across the entire remote sensing image, thereby enhancing building identification. First, this application calculates surface feature coefficients based on the geometric and textural features within each image patch in the remote sensing image. Combining geometric, textural, and feature distributions enhances the ability to identify building surface features. Then, based on the local and global distribution of corner features and elevation information within the image patch, feature optimization coefficients are calculated to further enhance building feature identification. Simultaneously, considering both local and global distributions in the image also improves robustness to environmental changes, increasing the accuracy of building mapping. This approach helps to accurately identify buildings despite the influence of building materials, background contrast, and environmental conditions, achieving high-precision mapping of building areas in real estate. Attached Figure Description
[0016] Figure 1 A flowchart illustrating the steps of a high-precision remote sensing image processing method provided in this application; Figure 2 The flowchart provided in this application illustrates the process of identifying buildings and non-buildings in an image block. Detailed Implementation
[0017] In the description of the embodiments in this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.
[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.
[0019] It should also be noted that the terms "first" and "second" in this application and its accompanying drawings are used to distinguish similar objects, rather than to describe a specific order or sequence. The methods disclosed in the embodiments of this application or the methods shown in the flowcharts include one or more steps for implementing the method. Without departing from the scope of this application, the execution order of multiple steps can be interchanged, and some steps can also be deleted.
[0020] Please see Figure 1 The diagram illustrates a flowchart of a high-precision remote sensing image processing method according to an embodiment of this application. The method includes the following steps: Step 1: Obtain remote sensing images of real estate surveying.
[0021] The method involves using a drone equipped with a lidar sensor to acquire remote sensing images of real estate surveying and the corresponding digital surface model (DSM) data. The acquired remote sensing images are used as input, and a median filtering algorithm is applied for image denoising to output the denoised remote sensing image. Other implementations may also employ existing known image denoising techniques such as mean filtering and Gaussian filtering.
[0022] Step 2: Cluster all pixels in the remote sensing image, and obtain all image patches for each cluster based on the distribution of pixels in the cluster; obtain several regular graphic regions based on the edge features in the remote sensing image; extract the straight line information in the remote sensing image; determine the surface feature coefficients of each image patch based on the distribution of edge pixels and the number of straight lines in each image patch, combined with the overlap between the regular image regions and each image patch.
[0023] For buildings, the texture often exhibits regular geometric features and contains a large number of straight lines. The top or sides of a building may have relatively complex texture patterns, and all complex textures are surrounded by regular geometric edges. Therefore, the texture features of building areas exhibit regular geometric edges and a large distribution of straight lines. For other objects, the regularity of their edge geometry is relatively poor, and the texture within the area is more chaotic, meaning the distribution of edge lines is more random and disordered.
[0024] First, the color values of all pixels and the number of clusters in the denoised remote sensing image are analyzed. As input, the k-means clustering algorithm is used, and the output is the number of clusters. In this embodiment, the number of clusters is... Choosing 10, it should be noted that when the number of clusters is too small, it affects the fineness of image segmentation, which in turn affects subsequent building mapping. Conversely, when the value is too large, it may damage the texture features in the image blocks, making accurate building mapping impossible. Therefore, the range of the number of clusters in this application is [insert range here]. .
[0025] Next, a binary mask image is generated for each cluster. Specifically, the pixel values belonging to that cluster are set to 1, and the values of the remaining pixels are set to 0. A connected component analysis algorithm is used to identify all connected regions in each binary image, and morphological operations are performed on each connected region to obtain the final image patch, thus completing the segmentation of the remote sensing image. It should be noted that the k-means clustering algorithm, connected component analysis algorithm, and morphological operations are all well-known techniques and will not be elaborated further. Additionally, other methods based on image texture and color distribution can be used for image segmentation in other implementations, including but not limited to mean shift, graph cut, and superpixel segmentation.
[0026] Furthermore, using the denoised remote sensing image as input, Canny edge detection is employed to output edge images. Based on the region of each image patch, edge lines for each patch are obtained. When an edge line spans multiple patches, only the edge line portion within the corresponding patch is considered. Using the edge images as input, Freeman chain code is used to trace boundaries clockwise, with a chain code simplification threshold set to 0.05, resulting in several regular regions. The bounding box information includes the corner coordinates and the length, width, and height of the bounding box. Hough line detection is then performed on the edge images of each patch, outputting the corresponding line information. Canny edge detection, Freeman chain code, and Hough line detection are well-known techniques and will not be elaborated further.
[0027] Based on the distribution of edge pixels and line information in each image block, and combined with the overlap between regular image regions and each image block, the surface feature coefficient of each image block is determined to measure the geometric and texture features within each region: the percentage of the area of the overlap between all regular image regions and each image block in the corresponding image block is calculated; edge pixels and non-edge pixels in each image block are marked with different values to obtain a measure of the randomness of the marked values of all pixels in each image block; the surface feature coefficient is positively correlated with the area percentage and the number of lines in each image block, and negatively correlated with the randomness measure.
[0028] In this embodiment, in the edge image of each image block, pixels located on the edge line are marked as 1, and otherwise marked as 0. Following a top-to-bottom and left-to-right order, the pixel values within each image block are used to construct the edge sequence of that image block. The formula for the surface feature coefficient is: ;in, It is the first Surface feature coefficients of each image patch; It is all regular graphic regions and the first The overlapping area of the image patches in the th... The area percentage of each image patch; It is the first The number of straight lines within an image block. It should be noted that as long as a straight line exists in any part of the image block, it is considered to be a straight line in that image block. It is the first Each image patch corresponds to a randomness measure of the edge sequence, which is calculated using entropy values; This represents a preset parameter adjustment factor, used to prevent the formula for the surface characteristic coefficient from being meaningless when the denominator is 0. In this embodiment, the value is 0.01.
[0029] It is understandable that the edges of buildings have relatively regular geometric features, and the interior contains a large number of straight-line texture features. At the same time, the texture distribution has a strong periodicity, that is, the distribution has a certain regularity. Therefore, the surface feature coefficient is relatively large in this case. Conversely, the surface feature coefficient of other objects is relatively small.
[0030] Step 3: Extract corner points from remote sensing images. Based on the position coordinates of corner points in each image block, obtain the corner point sequence in each dimension. Based on the randomness of the distribution of all corner points in each image block, and combined with the elevation information in the image block and the corner point sequence in each dimension, obtain the feature optimization coefficients of each image block.
[0031] While surface feature coefficients enhance the analysis of geometric and edge distribution features, and do improve recognition accuracy to some extent compared to the simple processing of geometry and texture in existing technologies, relying solely on surface features cannot completely eliminate errors caused by differences in materials and background contrast. Therefore, to further improve the sensitivity and accuracy of building recognition, feature extraction methods still need to be optimized. Among building-related features, corner points and elevation information are highly unique and suitable for distinguishing different objects.
[0032] Because buildings are typically regular geometric shapes, their corner points tend to be concentrated at the vertices of the shape, and the distribution of corner points along the same straight line exhibits a relatively obvious symmetry. Furthermore, the texture of buildings is also relatively regular, resulting in a high degree of regularity in the distribution of their corner points. In contrast, other objects exhibit relatively poor regularity, and due to the complexity of their textures, their corner point distribution often displays randomness.
[0033] Secondly, the elevation changes of buildings differ significantly from those of other features. Due to the varying heights of buildings, there are noticeable elevation fluctuations between image patches containing different buildings, and even within the same image patch, elevation data can show significant variations. For non-building features (such as vegetation and roads), elevation changes are typically more gradual. For example, the elevation changes of vegetation are mainly concentrated near the ground surface, with small differences between adjacent points. Similarly, the elevation changes of man-made features such as roads are relatively uniform, without significant fluctuations. Therefore, within the same image patch, the elevation data of these features usually do not show significant fluctuations, and the elevation data between different image patches with high similarity will not differ significantly.
[0034] The denoised remote sensing image is used as input, and a corner detection algorithm is employed to output corner coordinates. This embodiment uses Harris corner detection; the corner detection algorithm is a well-known technique, and its specific details are not elaborated here. Then, the corners located in each image patch are counted, and the x-coordinates and y-coordinates of all corners in an image patch are arranged in ascending order to form the x-dimensional and y-dimensional corner sequence of that image patch. Following a top-to-bottom and left-to-right order, the maximum elevation data of each pixel in an image patch are used to form the elevation sequence of that image patch.
[0035] Based on the randomness of the distribution of all corner points in each image patch, and combining the elevation information and corner point sequences in each dimension of the image patch, feature optimization coefficients for each image patch are obtained. This further enhances the recognition features of buildings and improves the accuracy of building recognition. Specifically: the average similarity measure between each image patch and the high-order sequence of other image patches in the same cluster is calculated; for each image patch, the minimum skewness of the corner point sequence in all dimensions is calculated and positively fused with the average similarity measure to obtain a first result; the absolute value of the Moran's index of the elevation values at all corner points in each image patch is positively fused with the dispersion of the elements in the high-order sequence of the corresponding image patch to obtain a second result; the negative correlation mapping of the first result is positively fused with the second result to obtain the feature optimization coefficients for each image patch.
[0036] In this embodiment, multiple variables are fused using a multiplication method. Specifically, : It is the first Feature optimization coefficients for each image patch; It is the first The absolute value of the Moran's index of the elevation values at all corner points in an image patch is used to measure the randomness of the corner point distribution. It is the first The dispersion of elements in a sequence of high-order segments of image blocks is calculated using variance; , They are the first The normalized values of the absolute values of the skewness of the corner point sequences in the x and y dimensions of an image patch. The normalization method uses the sigmoid function. The normalized values of the absolute values of the skewness are used to measure the symmetry of the corner point distribution. It is the first The average similarity measure between a high-order sequence of image patches and high-order sequences of other image patches in the same cluster. It should be noted that, in this embodiment, to ensure consistency of dimensions, variance and skewness need to be normalized; the similarity measure is obtained using the Pearson similarity coefficient method. Indicates the first result; This represents the negative correlation mapping of the first result; This indicates the second result.
[0037] It should be understood that although buildings belong to the same category, differences in floor height and top-floor structure result in low similarity in elevation between individuals; whereas non-buildings (such as ground and water surfaces) have uniform elevation and high similarity. Therefore, the average similarity measure... The smaller it is, the more likely it is to be a building.
[0038] Understandably, the corner points of buildings exhibit symmetrical characteristics and a certain regularity in their distribution, resulting in lower skewness and higher Moran's index. Secondly, elevation data within the same image patch may fluctuate significantly, and the elevation data corresponding to different image patches (i.e., different buildings) within the same cluster will also show significant differences, leading to larger feature optimization coefficients. Conversely, the feature optimization coefficients for image patches corresponding to other objects are smaller.
[0039] Step 4: Based on the surface feature coefficients and the feature optimization coefficients, cluster all image blocks to obtain building clusters, non-building clusters, and suspected building clusters. Combine the location distribution of the image blocks to obtain the mapping results of each image block.
[0040] The surface feature coefficients and feature optimization coefficients of each image patch in the remote sensing image are used to construct the feature vector of each image patch. The feature vectors of all image patches and the number of clusters are then analyzed. As input, the k-means clustering algorithm is used to output each cluster. In this embodiment, the number of clusters is... The value is 4. The k-means clustering algorithm is a well-known technique, and its details will not be elaborated further. Then, the mean surface feature coefficients and the mean feature optimization coefficients of all image patches in each cluster are processed using the minimum-maximum normalization method. The cluster with the largest sum of these normalization results is labeled as a building cluster, the cluster with the smallest sum is labeled as a non-building cluster, and the remaining clusters are labeled as suspected building clusters. All image patches in the building clusters are then classified as building image patches; all image patches in the non-building image patches are then classified as non-building image patches.
[0041] If the geometric and texture features, or corner and elevation features, of a target image patch closely resemble building features, and it is surrounded by other building image patches, then the image patch is likely blurred due to the building's material or the influence of other buildings. For example, when a relatively short building is occluded by other buildings, the features of the short building or its contrast with the background will decrease. Specifically, the number of image patches with known mapping results among the adjacent image patches is obtained, and the image patch with the largest number is taken as the target image patch. If all adjacent image patches with known mapping results are building image patches, then the target image patch is a building image patch.
[0042] Since the features of buildings are relatively obvious compared to other objects, it is relatively simple to identify buildings around other objects, and the features are also more obvious. The surface feature coefficient and feature optimization coefficient should be large. If the known mapping result image blocks adjacent to the target image block all belong to the non-building cluster, it means that the feature distinction between the target image block and the adjacent image blocks is not obvious. In this case, the target image block is a non-building area.
[0043] When a target image patch is located between non-building and building areas, the difference between buildings and non-building areas is more obvious, and the error caused by building materials and background contrast on the non-building side is less affected. Therefore, if the target image patch belongs to a building area, the difference between it and non-building areas is obvious, but the feature may be less obvious due to the influence of buildings on the other side. Thus, the distance between the feature vectors of the target image patch and all building image patches in its neighborhood is denoted as the first metric; the distance between the feature vectors of the target image patch and all non-building image patches in its neighborhood is denoted as the second metric. If the first metric is less than the second metric, the target image patch is a building image patch; if the first metric is greater than or equal to the second metric, the target image patch is a non-building image patch. In this embodiment, Euclidean distance is used for calculation.
[0044] The flowchart for identifying buildings and non-buildings in image blocks is as follows: Figure 2 As shown.
[0045] Through the above steps, each image patch is classified, and connected building image patches are marked as belonging to the same building. Using remote sensing imagery and the corresponding image patch for each building (i.e., control point data used for georegistration) as input, remote sensing image processing and Geographic Information System (GIS) technologies are employed to output the building's mapping information, such as its location, shape, size, and height. Remote sensing image processing and GIS technologies are well-known in the field of remote sensing mapping and will not be elaborated upon further.
[0046] Based on the same concept as the method embodiments of this application, a high-precision remote sensing image processing device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.
[0047] Based on the same concept as the method embodiments of this application, a high-precision remote sensing mapping image processing system is provided. The system stores a computer program, which, when executed by a processor, implements any of the methods described above.
[0048] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of the systems, methods, and computer program products according to embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0049] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A method for processing high-precision remote sensing images, characterized in that, The method includes the following steps: Acquire remote sensing images of real estate surveying; Cluster all pixels in the remote sensing image, and obtain all image patches for each cluster based on the distribution of pixels in the cluster; obtain several regular image regions based on the edge features in the remote sensing image; extract the straight line information in the remote sensing image; determine the surface feature coefficients of each image patch based on the distribution of edge pixels and the number of straight lines in each image patch, combined with the overlap between the regular image regions and each image patch. Corner points are extracted from remote sensing images. Based on the position coordinates of corner points in each image patch, corner point sequences in various dimensions are obtained. Based on the randomness of the distribution of all corner points in each image patch, combined with the elevation information in the image patch and the corner point sequences in various dimensions, feature optimization coefficients for each image patch are obtained. Based on the surface feature coefficients and the feature optimization coefficients, all image blocks are clustered to obtain building clusters, non-building clusters, and suspected building clusters. Combined with the location distribution of the image blocks, the mapping results of each image block are obtained. The determination of the surface feature coefficients for each image block is specifically as follows: Calculate the percentage of the area of each image block that overlaps with all regular image regions; Each edge pixel and non-edge pixel in each image block is labeled with a different value, and the randomness of the label values of all pixels in each image block is measured. The surface feature coefficient is positively correlated with the area ratio and the number of straight lines in each image block, and negatively correlated with the randomness measure. The specific steps for obtaining the corner point sequence in each dimension are as follows: The x-coordinates of all corner points in each image patch form the x-dimensional corner point sequence; the y-coordinates of all corner points in each image patch form the y-dimensional corner point sequence. The specific process for obtaining the feature optimization coefficients for each image block is as follows: The elevation information of all pixels in each image block is used to form the elevation sequence of each image block; the average similarity measure between the elevation sequences of each image block and other image blocks in the same cluster is calculated. For each image patch, the minimum skewness of the corner point sequence across all dimensions is calculated and positively fused with the average similarity metric to obtain the first result; The absolute value of the Moran index of the elevation values at all corner points in each image patch is positively fused with the dispersion of the elements in the elevation sequence of the corresponding image patch to obtain the second result; The negative correlation mapping of the first result is positively fused with the second result to obtain the feature optimization coefficients for each image patch.
2. The method for processing high-precision remote sensing images as described in claim 1, characterized in that, The specific steps for obtaining all image patches for each cluster are as follows: Extract the binary mask image of each cluster, and use each connected component in each binary mask image as an image block of the corresponding cluster.
3. The method for processing high-precision remote sensing images as described in claim 1, characterized in that, The regular image regions are determined using a chain code algorithm.
4. The method for processing high-precision remote sensing images as described in claim 1, characterized in that, The building cluster is specifically the cluster with the largest sum of the normalized mean of the surface feature coefficients and the normalized mean of the feature optimization coefficients of all image blocks; the non-building cluster is specifically the cluster with the smallest sum of the normalized mean of the surface feature coefficients and the normalized mean of the feature optimization coefficients of all image blocks; the suspected building cluster is specifically all clusters other than the building cluster and the non-building cluster.
5. The method for processing high-precision remote sensing images as described in claim 1, characterized in that, The process of obtaining the mapping results for each image patch includes: All image patches in building clusters are identified as building image patches; all image patches in non-building clusters are identified as non-building image patches. For image patches in a suspected building cluster, extract the image patch with the most mapped images in its neighborhood and denote it as the target image patch. If all mapped images in the target image patch's neighborhood are building image patches, then the target image patch is a building image patch. If all mapped images in the target image patch's neighborhood are non-building image patches, then the target image patch is a non-building image patch. Otherwise, use the distance metric between the target image patch and the feature vectors of all building image patches in its neighborhood as the first metric; use the distance metric between the target image patch and the feature vectors of all non-building image patches in its neighborhood as the second metric. If the first metric is less than the second metric, the target image patch is a building image patch; if the first metric is greater than or equal to the second metric, the target image patch is a non-building image patch. The feature vector of the image patch is determined by the surface feature coefficient and the feature optimization coefficient.
6. A high-precision remote sensing image processing device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-5.
7. A high-precision remote sensing image processing system, wherein the system stores a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-5.