Parameter iteration-based automatic modeling and updating method for dynamic evolution of urban morphology

Abstract

A parameter iteration-based automatic modeling and updating method for dynamic evolution of urban morphology, comprising six steps: data acquisition and three-dimensional model library construction, building height information calibration, building change area identification, high-precision modeling of change areas, building function information identification, and abnormal area identification, detection and updating. The method achieves dynamic monitoring and precise modeling of urban morphology by means of acquisition and processing of high-resolution remote sensing images, urban three-dimensional models, and land-use functional data, in conjunction with neural network models and image processing algorithms, thereby providing effective technical support for urban planning and management.

Description

Automatic Modeling and Update Method for Dynamic Evolution of Urban Morphology Based on Parameter Iteration Technical Field

[0001] This invention relates to the fields of urban planning, geographic information systems (GIS), remote sensing technology, computer vision, machine learning, 3D modeling, and urban planning. Specifically, it combines remote sensing image processing, neural network model applications, image recognition and segmentation technology, and lidar scanning technology to achieve dynamic monitoring, automatic modeling, and updating of urban morphology. Background Technology

[0002] With the acceleration of urbanization, the dynamic evolution monitoring and modeling of urban morphology has become an important requirement for urban planning and management. Traditional methods of urban morphology monitoring and modeling mainly rely on manual field surveys and two-dimensional map analysis. These methods are not only time-consuming and labor-intensive, but also difficult to capture subtle changes in the three-dimensional form of the city. In recent years, the development of remote sensing technology and geographic information systems (GIS) has provided new technical means for the dynamic monitoring of urban morphology. High-resolution remote sensing images can provide information on a wide range of urban surfaces, while three-dimensional modeling technology can intuitively display the urban spatial structure. However, how to efficiently and automatically extract information on urban morphological changes from remote sensing data and update the three-dimensional model in real time remains a technical challenge. Among the existing technologies, although some methods can realize the construction and updating of urban three-dimensional models, these methods often have the following shortcomings: (1) Low data processing efficiency: Traditional data processing methods are difficult to cope with the high-efficiency processing requirements of large amounts of remote sensing data. (2) Low accuracy: Existing automatic modeling technologies are difficult to meet the requirements of high-precision urban morphology modeling. (3) Insufficient functional information identification: Existing technologies are difficult to accurately identify and distinguish the functional information of buildings. (4) Weak dynamic update capability: Existing systems often lack effective dynamic monitoring and automatic update mechanisms. Summary of the Invention

[0003] Purpose of the invention: To address the shortcomings in the aforementioned background technology, this invention proposes an automatic modeling and updating system for the dynamic evolution of urban morphology based on parameter iteration. This system aims to improve data processing efficiency, enhance modeling accuracy, strengthen functional information recognition capabilities, and achieve dynamic monitoring and automatic updating of urban morphology. It can be widely applied in fields such as urban planning, urban change analysis, and urban electronic map production.

[0004] The technical solution adopted in this invention is: an automatic modeling and updating method for the dynamic evolution of urban morphology based on parameter iteration, comprising the following steps:

[0005] S1: Implement data acquisition and construct a 3D urban morphology model database; for a specified study period and region, collect dual-temporal high-resolution remote sensing images, urban 3D models, and urban land use function data; the image database integrates historical dual-temporal high-resolution remote sensing images of multiple cities; the urban morphology 3D model database stores shapefile files of urban 3D models that match the remote sensing images.

[0006] S2: Building height information calibration; The UNet neural network model is used to extract remote sensing information of buildings and shadows from the image database, and the building height is estimated by using the shadows extracted from the images and combining them with the imaging time of the remote sensing images.

[0007] S3: Building Change Area Identification; Using a region growing algorithm to compare changes in satellite images, areas where the building's outer contour changes beyond a threshold are marked as change areas;

[0008] S4: High-precision modeling of dynamic regions; edge detection of satellite images using the Sobel algorithm to accurately capture the edges, corners, and texture features of buildings; region segmentation based on image topology using the watershed algorithm to separate individual buildings; contour tracking of segmented regions using the Marching Squares algorithm to form precise building contours; contour optimization to remove noise and small irrelevant details; finally, further refinement of contour shape and continuity through morphological dilation and erosion operations.

[0009] S5: Building function information identification; high-precision modeling of the change area extracted in step S4 of land use function division to obtain building function features, and use different color blocks to distinguish building functions; construct a building geometric feature description index system, use the entropy weight method to assign weights to the indexes, calculate the high-precision modeling of the change area extracted in step S4 to obtain building geometric features, and then segment the measured values ​​and assign corresponding colors to each segment of the values.

[0010] Further, step S2 uses a UNet neural network model to extract remote sensing information about buildings and shadows from the image database, including: extracting urban building information using the original UNet convolutional neural network model; firstly, annotating the remote sensing images in the image database to generate label data for training and validation; the annotations include the boundaries of buildings and shadows, represented in pixel-level binary mask form; cropping the remote sensing image data to 256 pixels * 256 pixels, and then normalizing the image data to improve the stability and convergence speed of network training; expanding the dataset through operations such as rotation, flipping, cropping, and scaling to enhance the obtained sample data; dividing the sample data into a training set and a validation set at a ratio of 4:1; when the accuracy and loss rate of the validation set tend to stabilize and the model converges during model training, saving the optimal model for extracting building and shadow information.

[0011] Further, step S2 estimates the building height using the shadow extracted from the image and combined with the remote sensing image imaging time, including: estimating the building height using the directional relationship between the building and its shadow during image imaging; generating several parallel straight lines along the solar azimuth direction and calculating the average length of the straight lines within the building's shadow to estimate the building's shadow length; there are two geometric relationships between the solar altitude angle and the building's shadow, namely, the sun and the satellite being on the same side of the building and opposite sides; when the sun and the satellite are on the same side of the building, the building height H is estimated using the following formula: In the formula, ED is the shadow length on the image, EC is the actual shadow length of the building, CD is the shadow length that cannot be displayed on the image due to the building's obstruction, α is the solar altitude angle, and β is the satellite altitude angle. When the sun and the satellite are on opposite sides of the building, the formula for estimating the building height is H = EC × tanα = ED × tanα.

[0012] Furthermore, step S3, the region growing algorithm, compares the changes in the satellite image, including:

[0013] Select a seed point located where the building's functional and geometric features change; then set growth criteria, including similarity criteria and a change threshold; starting from the seed point, judge adjacent pixels or regions according to the growth criteria. If the similarity criteria are met, include them in the change region and perform region growth; during the growth process, merge adjacent regions that meet the conditions to form a larger change region; when the outer contour change of the change region exceeds a preset threshold, mark it as a change region; finally, when no new pixels or regions meet the growth criteria, the region growth process terminates, thereby achieving a comparison of changes in the building's functional and geometric features and effectively marking regions with significant outer contour changes.

[0014] Furthermore, the Sobel algorithm in step S4 for edge detection of the satellite image includes:

[0015] Step S4-1-1 Satellite image preprocessing: Perform Gaussian filtering preprocessing on the input satellite image of the variable area, use a 5×5 convolution kernel for noise suppression, and set the standard deviation σ = 1.5 to balance the denoising effect and edge preservation;

[0016] Step S4-1-2: Gradient calculation; Constructing the horizontal Sobel operator G x and the vertical Sobel operator G y Two-dimensional convolution operations are performed with the preprocessed image to obtain the horizontal gradient matrix I. x and the vertical gradient matrix I y The G x and G y for:

[0017] I x and I y for:

[0018] Where A is the original satellite image; the pixel-level gradient magnitude G and gradient direction θ are calculated based on the gradient matrix to establish a gradient feature field, wherein G and θ are:

[0019] Step S4-1-3 Edge Detection: A non-maximum suppression algorithm is used to refine the edges along the gradient direction, retaining pixels with local gradient maximum values ​​to eliminate edge blurring effects; a dynamic double threshold T is set. low and T high The pixels are divided into three categories: strong edge, weak edge, and non-edge. The T... low and T high For: T low =0.05×G max ,T high =0.15×G max

[0020] Among them, G max This represents the maximum value of the gradient magnitude G;

[0021] Pixel classification rules

[0022] Step S4-1-4 Edge connectivity optimization: Perform eight-neighbor connectivity detection on weak edge pixels. If a weak edge pixel is adjacent to a strong edge pixel, it is retained as a valid edge; otherwise, it is discarded. Output a binarized edge feature map and record the gradient direction matrix for subsequent building corner detection. The corner determination criterion is that the gradient direction changes by more than 45° within a 3×3 window.

[0023] Furthermore, the process of performing region segmentation based on image topology using the watershed algorithm in step S4 includes:

[0024] Step S4-2-1 Gradient magnitude normalization and image enhancement: Normalize the gradient magnitude matrix G output by Sobel edge detection to the [0,255] interval to generate the input gradient image for the watershed algorithm. The specific method is as follows:

[0025] Among them, G norm (i,j) is the value of the normalized gradient magnitude matrix at position (i,j), and G(i,j) is the value of the original gradient magnitude matrix at position (i,j). min and G max These are the global minimum and maximum values ​​of the gradient magnitude, respectively; histogram equalization is performed on the normalized image to enhance the edge features of low-contrast regions;

[0026] Step S4-2-2: Range field map generation and seed label extraction; binarize the gradient image and threshold T. binary =0.1×G max A building base mask is generated, resulting in an Euclidean distance transformation map D(x,y) representing the geometric distance of each pixel to the nearest background. Local maxima are searched in the distance field map as initial seeds, satisfying the following conditions:

[0027] D(x,y)>D(x±1,y±1) and D(x,y)≥5 pixels

[0028] The spacing between marker points must be greater than the minimum building spacing to avoid overly dense seeds; where D(x±1,y±1) represents the distance value of pixels within the 8-neighborhood of a pixel.

[0029] Step S4-2-3 Seed label optimization: Perform morphological opening operation on the seed label image using a 3×3 circular structuring element to eliminate isolated noise labels with an area less than 10 pixels, and retain connected regions with an area S ≥ 20 pixels. 2 The core of the candidate buildings;

[0030] Step S4-2-4 Watershed segmentation; Overlay the optimized seed-labeled image with the normalized gradient image to generate the label-constrained input image I. marked The seed region is forcibly set to the minimum gray value of 0; the water level is initialized to 0; pixels are traversed in ascending gray value order; the seed region is used as the initial catchment basin; when the water level rises to the watershed line (adjacent basins meet), the barrier position is recorded; the flooding termination condition is that the boundaries of all basins are extended to all basins to complete the boundary division.

[0031] Step S4-2-5 Region Merging Optimization: Merging over-segmented regions requires the following conditions to be met simultaneously:

[0032] (a) Mean gradient magnitude of the watershed line between adjacent regions satisfy:

[0033] (b) Area ratio:

[0034] For regions that meet the conditions, the watershed barrier is removed and the region labels are updated. The segmentation result map with topological labels is output, where each independent labeled region corresponds to a single building entity. The region adjacency matrix is ​​recorded for subsequent contour tracing. Here, min(S1,S2) and max(S1,S2) represent the smaller and larger values ​​of the two region areas, respectively.

[0035] Furthermore, the construction of the building geometric feature description index system in step S5, and the assignment of weights to the indicators using the entropy weight method, includes:

[0036] Geometric feature description index system

[0037] The beneficial effects of this invention are:

[0038] 1. An automatic modeling and updating method for the dynamic evolution of urban morphology based on parameter iteration was constructed. Compared with traditional methods, the recognition efficiency was improved by at least 80%, manual intervention was reduced, and rapid and automated monitoring of urban morphological changes was achieved.

[0039] 2. Reduced false recognition rate: This invention significantly reduces the false recognition rate in building change detection by optimizing the threshold setting and growth criteria of the region growing algorithm. Compared with traditional methods, the false recognition rate is reduced by at least 30%, thereby improving the accuracy and reliability of change area marking and reducing the workload of subsequent manual review. Attached Figure Description

[0040] Figure 1 is a flowchart of the automatic modeling and updating method for dynamic evolution of urban morphology based on parameter iteration according to the present invention. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0042] As shown in Figure 1, an automatic modeling and updating method for the dynamic evolution of urban morphology based on parameter iteration includes the following steps:

[0043] S1: Implement data acquisition and construct a 3D urban morphology model database; for a specified study period and region, collect dual-temporal high-resolution remote sensing images, urban 3D models, and urban land use function data; the image database integrates historical dual-temporal high-resolution remote sensing images of multiple cities; the urban morphology 3D model database stores shapefile files of urban 3D models that match the remote sensing images.

[0044] S2: Building Height Information Calibration; The UNet neural network model is used to extract remote sensing information of buildings and shadows from the image database. The extracted shadows are used in conjunction with the remote sensing image imaging time to estimate building height. Step S2, using the UNet neural network model to extract remote sensing information of buildings and shadows from the image database, includes: extracting urban building information using the original UNet convolutional neural network model; firstly, the remote sensing images in the image database are labeled to generate label data for training and validation. The labeled content includes the boundaries of buildings and shadows, represented in pixel-level binary mask form; the remote sensing image data is cropped to 256 pixels * 256 pixels, and then the image data is normalized to improve the stability and convergence speed of network training; the dataset is expanded through operations such as rotation, flipping, cropping, and scaling to enhance the obtained sample data; the sample data is divided into a training set and a validation set at a 4:1 ratio. When the accuracy and loss rate of the validation set tend to stabilize and the model converges during model training, the optimal model is saved for extracting building and shadow information.

[0045] Step S2 estimates the building height by extracting the shadow from the image and combining it with the remote sensing image imaging time. This includes: estimating the building height based on the directional relationship between the building and its shadow during image imaging; generating several parallel straight lines along the solar azimuth direction and calculating the average length of the line within the building's shadow to estimate the shadow length; recognizing two geometric relationships between the solar altitude angle and the building's shadow, namely, when the sun and satellite are on the same side of the building and when they are on opposite sides; and using the formula to estimate the building height H when the sun and satellite are on the same side of the building. In the formula, ED is the shadow length on the image, EC is the actual shadow length of the building, CD is the shadow length that cannot be displayed on the image due to the building's obstruction, α is the solar altitude angle, and β is the satellite altitude angle. When the sun and the satellite are on opposite sides of the building, the formula for estimating the building height is H = EC × tanα = ED × tanα.

[0046] S3: Building Change Area Identification; Using a region growing algorithm to compare changes in satellite images, areas where the building's outer contour changes beyond a threshold are marked as change areas;

[0047] The S3 region growing algorithm compares changes in satellite images, including:

[0048] Select a seed point located where the building's functional and geometric features change; then set growth criteria, including similarity criteria and a change threshold; starting from the seed point, judge adjacent pixels or regions according to the growth criteria. If the similarity criteria are met, include them in the change region and perform region growth; during the growth process, merge adjacent regions that meet the conditions to form a larger change region; when the outer contour change of the change region exceeds a preset threshold, mark it as a change region; finally, when no new pixels or regions meet the growth criteria, the region growth process terminates, thereby achieving a comparison of changes in building functional and geometric features and effectively marking regions with significant outer contour changes.

[0049] S4: High-precision modeling of dynamic regions; edge detection of satellite images using the Sobel algorithm to accurately capture the edges, corners, and texture features of buildings; region segmentation based on image topology using the watershed algorithm to separate individual buildings; contour tracking of segmented regions using the Marching Squares algorithm to form precise building contours; contour optimization to remove noise and small irrelevant details; finally, further refinement of contour shape and continuity through morphological dilation and erosion operations.

[0050] The Sobel algorithm in step S4 performs edge detection on the satellite image, including:

[0051] Step S4-1-1 Satellite image preprocessing: Perform Gaussian filtering preprocessing on the input satellite image of the variable area, use a 5×5 convolution kernel for noise suppression, and set the standard deviation σ = 1.5 to balance the denoising effect and edge preservation;

[0052] Step S4-1-2: Gradient calculation; Constructing the horizontal Sobel operator G x and the vertical Sobel operator G y Two-dimensional convolution operations are performed with the preprocessed image to obtain the horizontal gradient matrix I. x and the vertical gradient matrix I y The G x and G y for:

[0053] I x and I y for:

[0054] Where A is the original satellite image; the pixel-level gradient magnitude G and gradient direction θ are calculated based on the gradient matrix to establish a gradient feature field, wherein G and θ are:

[0055] Step S4-1-3 Edge Detection: A non-maximum suppression algorithm is used to refine the edges along the gradient direction, retaining pixels with local gradient maximum values ​​to eliminate edge blurring effects; a dynamic double threshold T is set. low and T high The pixels are divided into three categories: strong edge, weak edge, and non-edge. The T... low and T high For: T low =0.05×G max ,T high =0.15×G max

[0056] Among them, G max This represents the maximum value of the gradient magnitude G;

[0057] Pixel classification rules

[0058] Step S4-1-4 Edge connectivity optimization: Perform eight-neighbor connectivity detection on weak edge pixels. If a weak edge pixel is adjacent to a strong edge pixel, it is retained as a valid edge; otherwise, it is discarded. Output a binarized edge feature map and record the gradient direction matrix for subsequent building corner detection. The corner determination criterion is that the gradient direction changes by more than 45° within a 3×3 window.

[0059] Step S4, which involves using the watershed algorithm to perform region segmentation based on image topology, includes:

[0060] Step S4-2-1 Gradient magnitude normalization and image enhancement: Normalize the gradient magnitude matrix G output by Sobel edge detection to the [0,255] interval to generate the input gradient image for the watershed algorithm. The specific method is as follows:

[0061] Among them, G norm (i,j) is the value of the normalized gradient magnitude matrix at position (i,j), and G(i,j) is the value of the original gradient magnitude matrix at position (i,j). min and G max These are the global minimum and maximum values ​​of the gradient magnitude, respectively; histogram equalization is performed on the normalized image to enhance the edge features of low-contrast regions;

[0062] Step S4-2-2: Range field map generation and seed label extraction; binarize the gradient image and threshold T. binary =0.1×G max A building base mask is generated, resulting in an Euclidean distance transformation map D(x,y) representing the geometric distance of each pixel to the nearest background. Local maxima are searched in the distance field map as initial seeds, satisfying the following conditions:

[0063] D(x,y)>D(x±1,y±1) and D(x,y)≥5 pixels

[0064] The spacing between marker points must be greater than the minimum building spacing to avoid overly dense seeds; where D(x±1,y±1) represents the distance value of pixels within the 8-neighborhood of a pixel.

[0065] Step S4-2-3 Seed label optimization: Perform morphological opening operation on the seed label image using a 3×3 circular structuring element to eliminate isolated noise labels with an area less than 10 pixels, and retain connected regions with an area S ≥ 20 pixels. 2 The core of the candidate buildings;

[0066] Step S4-2-4 Watershed segmentation; Overlay the optimized seed-labeled image with the normalized gradient image to generate the label-constrained input image I. marked The seed region is forcibly set to the minimum gray value of 0; the water level is initialized to 0; pixels are traversed in ascending gray value order; the seed region is used as the initial catchment basin; when the water level rises to the watershed line (adjacent basins meet), the barrier position is recorded; the flooding termination condition is that the boundaries of all basins are extended to all basins to complete the boundary division.

[0067] Step S4-2-5 Region Merging Optimization: Merging over-segmented regions requires the following conditions to be met simultaneously:

[0068] (a) Mean gradient magnitude of the watershed line between adjacent regions satisfy:

[0069] (b) Area ratio:

[0070] For regions that meet the conditions, the watershed barrier is removed and the region labels are updated. The segmentation result map with topological labels is output, where each independent labeled region corresponds to a single building entity. The region adjacency matrix is ​​recorded for subsequent contour tracing. Here, min(S1,S2) and max(S1,S2) represent the smaller and larger values ​​of the two region areas, respectively.

[0071] S5: Building function information identification; high-precision modeling of the change area extracted in step S4 of land use function division to obtain building function features, and use different color blocks to distinguish building functions; construct a building geometric feature description index system, use the entropy weight method to assign weights to the indexes, calculate the high-precision modeling of the change area extracted in step S4 to obtain building geometric features, and then divide the measured values ​​into segments and assign corresponding colors to each segment of the values.

[0072] Step S5, which involves constructing a building geometric feature description index system and assigning weights to the indexes using the entropy weight method, includes:

[0073] Geometric feature description index system

[0074] 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 illustrative of the principles of 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 claimed invention.

Claims

1. An automatic modeling and updating method for the dynamic evolution of urban morphology based on parameter iteration, characterized in that, Includes the following steps: S1: Implement data acquisition and construct a 3D urban morphology model database; for a specified study period and region, collect dual-temporal high-resolution remote sensing images, urban 3D models, and urban land use function data; the image database integrates historical dual-temporal high-resolution remote sensing images of multiple cities; the urban morphology 3D model database stores shapefile files of urban 3D models that match the remote sensing images. S2: Building height information calibration; The UNet neural network model is used to extract remote sensing information of buildings and shadows from the image database, and the building height is estimated by using the shadows extracted from the images and combining them with the imaging time of the remote sensing images. S3: Building Change Area Identification; Using a region growing algorithm to compare changes in satellite images, areas where the building's outer contour changes beyond a threshold are marked as change areas; S4: High-precision modeling of dynamic regions; edge detection of satellite images using the Sobel algorithm to accurately capture the edges, corners, and texture features of buildings; region segmentation based on image topology using the watershed algorithm to separate individual buildings; contour tracking of segmented regions using the Marching Squares algorithm to form precise building contours; contour optimization to remove noise and small irrelevant details; finally, further refinement of contour shape and continuity through morphological dilation and erosion operations. S5: Building function information identification; Based on the high-precision modeling of the change area extracted in step S4 of land use function division, the building function features are obtained, and different color blocks are used to distinguish the building functions. A building geometric feature description index system is constructed, and the index is assigned weights using the entropy weight method. The variable area extracted in step S4 is modeled with high precision to obtain the building geometric features. Subsequently, the measured values ​​are segmented, and each segment is assigned a corresponding color.

2. The automatic modeling and updating method for dynamic evolution of urban morphology based on parameter iteration according to claim 1, characterized in that, Step S2 uses a UNet neural network model to extract remote sensing information about buildings and shadows from the image database. This includes: extracting urban building information using the original UNet convolutional neural network model; firstly, annotating the remote sensing images in the image database to generate label data for training and validation; the annotations include the boundaries of buildings and shadows, represented as pixel-level binary masks; cropping the remote sensing image data to 256 pixels * 256 pixels, and then normalizing the image data to improve the stability and convergence speed of network training; expanding the dataset through operations such as rotation, flipping, cropping, and scaling to enhance the obtained sample data; dividing the sample data into a training set and a validation set at a 4:1 ratio; and saving the optimal model for extracting building and shadow information after the accuracy and loss rate of the validation set stabilize and the model converges during training.

3. The automatic modeling and updating method for dynamic evolution of urban morphology based on parameter iteration according to claim 2, characterized in that, Step S2 estimates the building height by extracting the shadow from the image and combining it with the remote sensing image imaging time. This includes: estimating the building height based on the directional relationship between the building and its shadow during image imaging; generating several parallel straight lines along the solar azimuth direction and calculating the average length of the line within the building's shadow to estimate the shadow length; recognizing two geometric relationships between the solar altitude angle and the building's shadow, namely, when the sun and satellite are on the same side of the building and when they are on opposite sides; and using the formula to estimate the building height H when the sun and satellite are on the same side of the building. In the formula, ED is the shadow length on the image, EC is the actual shadow length of the building, CD is the shadow length that cannot be displayed on the image due to the building's obstruction, α is the solar altitude angle, and β is the satellite altitude angle. When the sun and the satellite are on opposite sides of the building, the formula for estimating the building height is H = EC × tanα = ED × tanα.

4. The automatic modeling and updating method for dynamic evolution of urban morphology based on parameter iteration according to claim 3, characterized in that, The S3 region growing algorithm compares changes in satellite images, including: Select a seed point located where the building's functional and geometric features change; then set growth criteria, including similarity criteria and a change threshold; starting from the seed point, judge adjacent pixels or regions according to the growth criteria. If the similarity criteria are met, include them in the change region and perform region growth; during the growth process, merge adjacent regions that meet the conditions to form a larger change region; when the outer contour change of the change region exceeds a preset threshold, mark it as a change region; finally, when no new pixels or regions meet the growth criteria, the region growth process terminates, thereby achieving a comparison of changes in the building's functional and geometric features and effectively marking regions with significant outer contour changes.

5. The automatic modeling and updating method for dynamic evolution of urban morphology based on parameter iteration according to claim 4, characterized in that, The Sobel algorithm in step S4 performs edge detection on the satellite image, including: Step S4-1-1 Satellite image preprocessing: Perform Gaussian filtering preprocessing on the input satellite image of the variable area, use a 5×5 convolution kernel for noise suppression, and set the standard deviation σ = 1.5 to balance the denoising effect and edge preservation; Step S4-1-2: Gradient calculation; Constructing the horizontal Sobel operator G x and the vertical Sobel operator G y Two-dimensional convolution operations are performed with the preprocessed image to obtain the horizontal gradient matrix I. x and the vertical gradient matrix I y The G x and G y for: I x Japanese I y Main: Where A is the original satellite image; the pixel-level gradient magnitude G and gradient direction θ are calculated based on the gradient matrix to establish a gradient feature field, wherein G and θ are: Step S4-1-3 Edge Detection: A non-maximum suppression algorithm is used to refine the edges along the gradient direction, retaining pixels with local gradient maximum values ​​to eliminate edge blurring effects; a dynamic double threshold T is set. low and T high The pixels are divided into three categories: strong edge, weak edge, and non-edge. The T... low and T high for: T low =0.05×G max ,T high =0.15×G max Among them, G max This represents the maximum value of the gradient magnitude G; Pixel classification rules Step S4-1-4 Edge connectivity optimization: Perform eight-neighbor connectivity detection on weak edge pixels. If a weak edge pixel is adjacent to a strong edge pixel, it is retained as a valid edge; otherwise, it is discarded. Output a binarized edge feature map and record the gradient direction matrix for subsequent building corner detection. The corner determination criterion is that the gradient direction changes by more than 45° within a 3×3 window.

6. The automatic modeling and updating method for dynamic evolution of urban morphology based on parameter iteration according to claim 5, characterized in that, Step S4, which involves using the watershed algorithm to perform region segmentation based on image topology, includes: Step S4-2-1 Gradient magnitude normalization and image enhancement: Normalize the gradient magnitude matrix G output by Sobel edge detection to the [0,255] interval to generate the input gradient image for the watershed algorithm. The specific method is as follows: Among them, G norm (i,j) is the value of the normalized gradient magnitude matrix at position (i,j), and G(i,j) is the value of the original gradient magnitude matrix at position (i,j). min and G max These are the global minimum and maximum values ​​of the gradient magnitude, respectively; histogram equalization is performed on the normalized image to enhance the edge features of low-contrast regions; Step S4-2-2: Range field map generation and seed label extraction; binarize the gradient image and threshold T. binary =0.1×G max A building base mask is generated, resulting in an Euclidean distance transformation map D(x,y) representing the geometric distance of each pixel to the nearest background. Local maxima are searched in the distance field map as initial seeds, satisfying the following conditions: D(x,y)>D(x±1,y±1) and D(x,y)≥5 pixels The spacing between marker points must be greater than the minimum building spacing to avoid overly dense seeds; where D(x±1,y±1) represents the distance value of pixels within the 8-neighborhood of a pixel. Step S4-2-3 Seed label optimization: Perform morphological opening operation on the seed label image using a 3×3 circular structuring element to eliminate isolated noise labels with an area less than 10 pixels, and retain connected regions with an area S ≥ 20 pixels. 2 The core of the candidate buildings; Step S4-2-4 Watershed segmentation; Overlay the optimized seed-labeled image with the normalized gradient image to generate the label-constrained input image I. marked The seed region is forcibly set to the minimum gray value of 0; the water level is initialized to 0; pixels are traversed in ascending gray value order; the seed region is used as the initial water catchment basin; when the water level rises to the watershed line, the barrier position is recorded; the flooding termination condition is that all basin boundaries are extended to all basins to complete the boundary division. Step S4-2-5 Region Merging Optimization: Merging over-segmented regions requires the following conditions to be met simultaneously: (a) Mean gradient magnitude of the watershed line between adjacent regions satisfy: (b) Area ratio: For regions that meet the conditions, the watershed barrier is removed and the region labels are updated. The segmentation result map with topological labels is output, where each independent labeled region corresponds to a single building entity. The region adjacency matrix is ​​recorded for subsequent contour tracing. Here, min(S1,S2) and max(S1,S2) represent the smaller and larger values ​​of the two region areas, respectively.

7. The automatic modeling and updating method for dynamic evolution of urban morphology based on parameter iteration according to claim 6, characterized in that, Step S5, which involves constructing a building geometric feature description index system and assigning weights to the indexes using the entropy weight method, includes: Geometric feature description index system

Images