Sunken green space system and method under the design of sponge city garden combined with rainwater management

Abstract

The application discloses a sunken green land system and method for sponge city garden design combined with rainwater management. Global high-precision terrain point cloud data and surface runoff monitoring data of the sunken green land are collected, and a terrain gradient mean difference value between an initial confluence area and an initial non-confluence area is counted. When the terrain gradient mean difference value is lower than a complex terrain catchment area division threshold value, a typical depression terrain feature reference and a typical slope terrain feature reference are extracted, a depression feature similarity grid map is constructed, and a confluence cumulative amount map is weighted and corrected for a depression storage effect to obtain a confluence cumulative amount correction map. The confluence cumulative amount correction map, a terrain slope map and a depression feature similarity grid map are superimposed to generate a multi-feature composite map, the multi-feature composite map is input into a pre-trained semantic segmentation network model, a binary catchment area mask is output, and a catchment area boundary coordinate sequence is generated. The technical scheme can realize accurate division of a sunken green land catchment area boundary under complex garden micro-terrain conditions.

Description

Technical Field

[0001] This application relates to the field of sponge city landscape design technology, and more specifically, to a sunken green space system and method for sponge city landscape design that incorporates rainwater management. Background Technology

[0002] In sponge city construction, sunken green spaces are typically used to receive surface runoff from surrounding paved areas, road edges, and garden micro-topography areas, reducing peak stormwater runoff through short-term retention, infiltration, and purification. To ensure that sunken green spaces can receive water as designed and function as stormwater storage facilities, the corresponding catchment area must be determined during the landscape design phase. This allows for further determination of the sunken green space's location, service area, water depth, overflow location, and connection to surrounding drainage facilities.

[0003] Current methods for delineating the catchment areas of sunken green spaces typically rely on topographic elevation data to generate slope aspect, slope gradient, and runoff accumulation, and then identify the catchment boundaries based on the magnitude of the runoff accumulation or manually set runoff thresholds. While this approach is applicable to scenarios with significant topographic relief and clear runoff directions, in sponge city landscaping scenarios, sunken green spaces often exhibit complex micro-topographic features such as gentle slopes, shallow depressions, gradual slope transitions, raised paving edges, and weakly undulating green space boundaries. The topographic gradient differences between the catchment and non-catchment areas are not significant. Traditional delineation methods based on slope aspect and runoff accumulation are easily affected by minor elevation errors and threshold settings, leading to breaks, expansions, or omissions in the catchment area boundaries, making it difficult to accurately reflect the actual spatial range of runoff received by the sunken green space. In particular, the retention effect of shallow depressions can cause surface runoff to be interrupted at these depressions. Traditional methods cannot accurately identify the boundaries of such interruptions and often mistakenly classify the downstream areas of shallow depressions as catchment areas. Therefore, how to improve the accuracy of the boundary delineation of the catchment area of ​​sunken green spaces under complex garden micro-topography conditions where the topographic gradient is not obvious has become an urgent problem to be solved in this field. Summary of the Invention

[0004] This application provides a sunken green space system and method for sponge city landscaping design that integrates rainwater management, which can improve the accuracy of the boundary delineation of the water catchment area of ​​sunken green space under complex micro-topographic conditions where the topographic gradient difference is not obvious.

[0005] Firstly, this application provides a method for delineating the catchment area of ​​sunken green spaces in sponge city landscaping design, comprising the following steps: High-precision topographic point cloud data and surface runoff monitoring data of the entire sunken green space were collected to calculate the topographic slope map and runoff accumulation map, and the mean difference of topographic gradient between the initial runoff area and the initial non-runoff area in the runoff accumulation map was calculated. When the mean difference of the terrain gradient is lower than the threshold for dividing the water catchment area of ​​complex terrain, typical depression terrain feature benchmarks and typical slope terrain feature benchmarks are extracted from the terrain point cloud data. Based on the typical depression terrain feature benchmarks and the typical slope terrain feature benchmarks, the terrain feature similarity normalization calculation is performed on the raster digital elevation model generated by the terrain point cloud to obtain the depression feature similarity of each raster unit. A depression feature similarity raster map is constructed based on the similarity of all depression features. The original confluence accumulation map is then weighted and corrected for depression stagnation effect using the depression feature similarity raster map to obtain a corrected confluence accumulation map. The multi-feature composite map generated by overlaying the confluence accumulation correction map, the terrain slope map, and the depression feature similarity raster map is input into the pre-trained semantic segmentation network model, and the output is a binarized catchment area mask. The outline of the catchment area mask is used as the catchment area boundary coordinate sequence.

[0006] In some embodiments, the difference in the mean topographic gradient between the initial confluence region and the initial non-confluence region in the statistical confluence accumulation map specifically includes: The confluence accumulation map is divided into initial confluence regions and initial non-confluence regions according to a predetermined confluence accumulation threshold. Based on the terrain slope map, the slope values ​​of each grid cell in the initial confluence area and the slope values ​​of each grid cell in the initial non-confluence area are read respectively; The average slope values ​​within the initial confluence area are statistically analyzed to obtain the average topographic gradient of the confluence area. The average slope values ​​within the initial non-convergence area are statistically analyzed to obtain the average topographic gradient of the non-convergence area. The absolute value of the difference between the mean topographic gradient of the confluence region and the mean topographic gradient of the non-confluence region is defined as the difference in mean topographic gradient.

[0007] In some embodiments, extracting typical depression terrain feature benchmarks and typical slope terrain feature benchmarks from the terrain point cloud data specifically includes: Based on the terrain point cloud data, generate a local relative elevation map, a terrain slope map, and a terrain curvature map; Extract the local relative elevation value, slope value, curvature value and confluence accumulation value corresponding to each grid cell from the local relative elevation map, the terrain slope map, the terrain curvature map and the confluence accumulation map; Grid cells whose local relative elevation value is lower than the predetermined depression elevation threshold, whose curvature value meets the predetermined concave curvature range, and whose cumulative runoff value is higher than the predetermined cumulative runoff threshold are identified as depression candidate cells. Raster cells with local relative elevation values ​​higher than a predetermined slope elevation threshold, slope values ​​higher than a predetermined slope threshold, and runoff accumulation values ​​lower than a predetermined runoff accumulation threshold are identified as candidate slope cells. Spatial connectivity screening is performed on the candidate depression units to obtain typical depression terrain feature benchmarks; Spatial connectivity screening is performed on the candidate slope units to obtain typical slope terrain feature benchmarks.

[0008] In some embodiments, the terrain feature similarity normalization calculation of the raster digital elevation model generated from the terrain point cloud is performed based on the typical depression terrain feature benchmark and the typical slope terrain feature benchmark to obtain the depression feature similarity of each raster cell, specifically including: The local relative elevation, slope, curvature, and cumulative runoff values ​​of the typical depression topographic feature benchmark are extracted to obtain the depression feature benchmark vector; The local relative elevation, slope, curvature, and cumulative runoff values ​​of the typical slope topographic feature benchmark are extracted to obtain the slope feature benchmark vector; Extract the local relative elevation, slope, curvature and confluence accumulation values ​​of each grid cell in the grid digital elevation model to obtain the feature vector of the grid cell to be calculated; Calculate the Euclidean distance between the feature vector of the grid cell to be calculated and the reference vector of the depression feature to obtain the depression feature distance; Calculate the Euclidean distance between the feature vector of the grid cell to be calculated and the reference vector of the slope feature to obtain the slope feature distance; The depression feature similarity of each grid cell is determined based on the proportion of the slope feature distance in the depression feature distance and the sum of the slope feature distances.

[0009] Specifically, if the local relative elevation, curvature, and cumulative runoff values ​​of a certain grid cell are closer to the depression feature reference vector, then the depression feature distance of that grid cell is smaller, the slope feature distance is larger, and its depression feature similarity is also higher. If the slope value and local relative elevation value of a certain grid cell are closer to the slope feature reference vector, then the slope feature distance of that grid cell is smaller, the depression feature distance is larger, and its depression feature similarity is lower. In this application, by using this continuous similarity expression, the weak runoff tendency in shallow depressions, gentle slopes, and transition edges can be preserved.

[0010] In some embodiments, the original confluence accumulation map is weighted and corrected for the confluence stagnation effect using the depression feature similarity raster map to obtain the corrected confluence accumulation map. Specifically, this includes: Based on the similarity of depression features of each grid cell in the depression feature similarity grid map, each grid cell is divided into high depression similarity cells, medium depression similarity cells, and low depression similarity cells. A first confluence correction coefficient is configured for the high-low similarity unit; A second confluence correction coefficient is configured for the mid-depression similarity unit; A third confluence correction coefficient is configured for the low-lying area similarity unit; Wherein, the first bus correction coefficient is less than the second bus correction coefficient, the second bus correction coefficient is less than the third bus correction coefficient, and the value range of all bus correction coefficients is from 0.1 to 1.0; Based on the first, second, and third confluence correction coefficients, the confluence accumulation values ​​of the corresponding grid cells in the original confluence accumulation map are weighted to obtain the confluence accumulation correction map.

[0011] Among them, high depression similarity units refer to grid units with depression feature similarity in a predetermined high range. They have more obvious depression storage and retention characteristics and can intercept more upstream runoff. Therefore, a smaller confluence correction coefficient can be configured. Medium depression similarity units refer to grid units with depression feature similarity in a middle range. They are located at the edge of shallow depressions or in the transition area of ​​gentle slopes. Their storage and retention capacity is weak. Therefore, a medium confluence correction coefficient can be configured. Low depression similarity units refer to grid units with depression feature similarity in a predetermined low range. They are closer to slopes or non-confluence areas and have almost no storage and retention capacity. Therefore, a confluence correction coefficient close to 1.0 is configured. In this application, by configuring different confluence correction coefficients, the problem of overestimation of confluence accumulation caused by depression storage and retention effect in the original confluence accumulation map can be corrected.

[0012] In some embodiments, the pre-trained semantic segmentation network model is trained through the following steps: Obtain the corrected map of runoff accumulation, topographic slope map, raster map of depression feature similarity, and artificial catchment area mask based on surface runoff monitoring data corresponding to the historical sunken green space sample area; The confluence accumulation correction map, the terrain slope map, and the depression feature similarity raster map are overlaid in the same raster position to obtain the training input multi-feature composite map. The training input multi-feature composite map is input into the semantic segmentation network model to obtain the predicted catchment area mask; The predicted catchment area mask is compared with the artificial catchment area mask verified based on surface runoff monitoring data at the pixel level to obtain the mask difference results. Update the network parameters of the semantic segmentation network model based on the mask difference results; When the mask difference results meet the predetermined convergence conditions, a pre-trained semantic segmentation network model is obtained.

[0013] The artificial catchment mask, verified based on surface runoff monitoring data, can be obtained by designers based on on-site runoff observation data, post-rainwater accumulation traces, road and paving slopes, storm drain locations, green space edge elevations, and historical maintenance records. Surface runoff monitoring data can be used to verify the actual flow direction and inlet location of the runoff, improving the accuracy of the manual verification. The runoff accumulation correction map in the training input multi-feature composite map is used to represent the runoff characteristics after correction for depression retention effects; the topographic slope map is used to represent surface slope changes; and the depression feature similarity raster map is used to represent shallow depressions or depression tendencies. After these three types of layers are overlaid with channels at the same raster position, the semantic segmentation network model can simultaneously read runoff, slope, and depression feature similarity information at the same spatial location.

[0014] In some embodiments, using the contour of the catchment area mask as a sequence of catchment area boundary coordinates specifically includes: The water catchment area mask is subjected to a hole filling process to obtain a filled water catchment area mask; The isolated patch removal process is performed on the filled catchment area mask to obtain a connected catchment area mask; Extract the outer contour pixels of the connected catchment area mask to obtain the catchment area contour pixel set; Based on the spatial reference information of the raster digital elevation model, the set of pixels representing the catchment area is converted into a set of planar coordinate points; The set of planar coordinate points is sorted according to the contour tracing order to obtain the boundary coordinate sequence of the catchment area.

[0015] The process includes: hole filling to repair small blank areas inside the catchment area mask; isolated patch removal to eliminate scattered misidentified areas not connected to the main catchment area; and outer contour pixels to represent the boundary between the catchment area mask and non-catchment areas. In this application, after converting the outer contour pixels into a set of planar coordinate points, they can be further arranged according to the contour tracking order to form a catchment area boundary coordinate sequence. This catchment area boundary coordinate sequence can be imported into landscape design software or stormwater management platforms to calculate the catchment area, verify the water storage capacity of sunken green spaces, determine the location of overflow outlets, and assist in stormwater organization design.

[0016] Secondly, this application provides a sunken green space system for sponge city landscaping design that integrates rainwater management, which includes a water catchment area division unit for the sunken green space. The water catchment area division unit of the sunken green space specifically includes: The data acquisition module is used to collect high-precision topographic point cloud data and surface runoff monitoring data of the entire sunken green space, calculate the topographic slope map and runoff accumulation map, and calculate the mean difference of topographic gradient between the initial runoff area and the initial non-runoff area in the runoff accumulation map. The processing module is used to extract typical depression terrain feature benchmarks and typical slope terrain feature benchmarks from the terrain point cloud data when the mean difference of the terrain gradient is lower than the threshold for dividing the water catchment area of ​​complex terrain. Based on the typical depression terrain feature benchmarks and the typical slope terrain feature benchmarks, the module performs terrain feature similarity normalization calculation on the raster digital elevation model generated from the terrain point cloud to obtain the depression feature similarity of each raster unit. The processing module is also used to construct a depression feature similarity raster map based on all depression feature similarities, and to perform a depression stagnation effect weighted correction on the original confluence accumulation map using the depression feature similarity raster map to obtain a confluence accumulation correction map; The execution module is used to input a multi-feature composite map generated by overlaying the confluence accumulation correction map, the terrain slope map, and the depression feature similarity raster map into a pre-trained semantic segmentation network model, and output a binarized catchment area mask, using the contour of the catchment area mask as the catchment area boundary coordinate sequence.

[0017] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described method for dividing the catchment area of ​​sunken green space in sponge city landscape design.

[0018] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-mentioned method for dividing the catchment area of ​​sunken green space in sponge city landscape design.

[0019] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In the process of delineating the catchment area of ​​sunken green spaces in sponge city landscape design, this application first collects high-precision topographic point cloud data and surface runoff monitoring data for the entire sunken green space, and calculates topographic slope maps and runoff accumulation maps, thereby obtaining micro-topographic undulation information and surface runoff collection information within the entire sunken green space. Secondly, by statistically analyzing the difference in the mean topographic gradient between the initial catchment area and the initial non-catchment area, it determines whether the traditional delineation process relying on differences in topographic slope and runoff accumulation is suitable for the current micro-topographic conditions of the landscape. When the mean difference in topographic gradient is lower than the threshold for delineating catchment areas in complex terrain, this application no longer directly relies on the original runoff accumulation map to determine the catchment area boundary. Instead, it extracts typical depression topographic feature benchmarks and typical slope topographic feature benchmarks from the topographic point cloud data, and utilizes the typical depression... The terrain feature benchmark and the typical slope terrain feature benchmark are used to perform terrain feature similarity normalization calculation on the raster digital elevation model to obtain the depression feature similarity of each raster unit. Next, this application constructs a depression feature similarity raster map based on all depression feature similarities, and uses the depression feature similarity raster map to perform depression stagnation effect weighted correction on the original runoff accumulation map, correcting the overestimation of runoff accumulation caused by depression stagnation effect in the original runoff accumulation map, so that the runoff accumulation is more consistent with the actual surface runoff process. Finally, the multi-feature composite map generated by overlaying the runoff accumulation correction map, terrain slope map and depression feature similarity raster map is input into a pre-trained semantic segmentation network model, outputting a binarized catchment area mask, and the contour of the catchment area mask is used as the catchment area boundary coordinate sequence. Therefore, this application can improve the accuracy of the boundary delineation of the water catchment area of ​​sunken green space under complex garden micro-topography conditions where the topographic gradient difference is not obvious, so that the output water catchment area boundary is more in line with the actual spatial range of runoff received by the sunken green space. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the main process of the method for dividing the catchment area of ​​sunken green space in sponge city landscape design according to some embodiments of this application; Figure 2 This is a schematic diagram illustrating the statistical analysis of the difference in the mean of terrain gradients and the determination of complex micro-terrain, based on some embodiments of this application. Figure 3 This is a schematic diagram illustrating the extraction of typical depression terrain feature benchmarks and typical slope terrain feature benchmarks, as well as the calculation process of terrain feature similarity normalization, according to some embodiments of this application. Figure 4 This is a structural schematic diagram of the water catchment area division unit of a sunken green space according to some embodiments of this application. Detailed Implementation

[0021] To facilitate understanding of the technical solutions of the embodiments of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. The following embodiments are only used to explain this application and are not intended to limit this application.

[0022] This application provides a method for delineating the catchment area of ​​sunken green spaces in sponge city landscape design. This method can be applied to the sponge city landscape design stage, as well as to the renovation, review, and stormwater management evaluation stages of existing sunken green spaces. The sunken green space refers to a green space unit that is lower than the surrounding ground surface, capable of receiving surface runoff from the surrounding area and storing, infiltrating, or purifying rainwater. The catchment area refers to the spatial range that can generate surface runoff flowing into the sunken green space under rainfall conditions. For sunken green spaces, the accuracy of the catchment area boundary directly affects the determination of the service area, the verification of water storage capacity, the layout of overflow outlets, and the organization of stormwater runoff.

[0023] Existing methods based on slope aspect and runoff accumulation are applicable to scenarios with significant topographic relief and clear runoff directions. However, in sponge city landscape scenarios, green space edges, paved boundaries, road edges, and transitional areas with gentle slopes often involve shallow depressions, weak slopes, and gentle slopes. The topographic gradient difference between runoff and non-runoff areas is not significant. Directly determining the catchment area boundary based on the original runoff accumulation map can easily lead to boundary breaks, boundary expansion, or local omissions. In particular, the retention effect in shallow depressions can cause surface runoff to be interrupted at these depressions. Traditional methods cannot accurately identify such interruption boundaries, often incorrectly classifying the downstream area of ​​the depression into the catchment area. This application uses the difference in the mean topographic gradient to determine complex micro-topographic conditions. Under complex micro-topographic conditions, it introduces typical depression topographic feature benchmarks and typical slope topographic feature benchmarks. By using depression feature similarity to weight and correct the original runoff accumulation map for depression retention effects, it obtains a catchment area boundary that more accurately reflects the actual runoff carrying capacity of sunken green spaces.

[0024] refer to Figure 1 , Figure 1 This is a schematic flowchart illustrating the method for dividing the catchment area of ​​sunken green space in sponge city landscaping design according to some embodiments of this application. The method includes steps 101 to 104.

[0025] In step 101, high-precision topographic point cloud data and surface runoff monitoring data of the entire sunken green space are collected, topographic slope map and runoff accumulation map are calculated, and the mean difference of topographic gradient between the initial runoff area and the initial non-runoff area in the runoff accumulation map is calculated.

[0026] Among them, the high-precision topographic point cloud data of the entire sunken green space refers to the set of three-dimensional spatial points covering the sunken green space itself and the surrounding areas where runoff may flow into the sunken green space. Each point in this set has both planar and elevation coordinates, which can be used to represent the micro-topographic undulations around the sunken green space. Specifically, data can be collected using ground-based three-dimensional laser scanning equipment, mobile surveying equipment, UAV oblique photogrammetry equipment, or other equipment capable of acquiring high-precision surface elevation information. Surface runoff monitoring data refers to data used to characterize the direction of surface runoff after rainfall, the location of runoff inlets, the extent of shallow water accumulation, or the boundaries of watermarks. This data can be obtained through runoff monitoring instruments deployed at key locations, high-definition cameras, or manual on-site observation.

[0027] In some implementations, the acquired high-precision terrain point cloud data can first undergo coordinate unification, outlier removal, ground point classification, and elevation normalization to obtain ground point cloud data suitable for rasterization representation. Coordinate unification ensures that data from different acquisition batches are in the same spatial reference system; outlier removal eliminates abnormal points caused by reflection, occlusion, or equipment vibration; ground point classification excludes points from vegetation canopy, building facades, and temporary objects; and elevation normalization reduces the impact of overall terrain changes on the assessment of local micro-topography.

[0028] Subsequently, the ground point cloud data is converted into a raster digital elevation model (DEM). A raster DEM is a digital elevation model that uses a regular raster to represent the surface elevation. For the micro-topography surrounding sunken green spaces, the raster resolution should not be too coarse, otherwise shallow depressions, gentle slopes, and edge uplift features will be smoothed out; conversely, the raster resolution should not be too fine, otherwise point cloud noise will be amplified. Therefore, a predetermined raster resolution can be set according to the point cloud density, the precision of the landscape design, and the scale of the sunken green space, typically between 0.1 meters and 0.5 meters.

[0029] Based on a raster DEM, the slope value of each raster cell can be calculated to obtain a topographic slope map. The topographic slope map is used to represent the topographic gradient changes of each raster cell. Furthermore, the runoff direction and the number of upstream inflow cells for each raster cell can be determined based on the raster DEM to obtain a runoff accumulation map. The runoff accumulation map is used to represent the degree to which each raster cell receives upstream runoff. Generally, areas with higher runoff accumulation are more likely to be located along the runoff path or in areas of concentrated catchment.

[0030] refer to Figure 2 , Figure 2 This is a schematic diagram illustrating the statistical analysis of the mean difference of terrain gradient and the determination of complex micro-terrain, based on some embodiments of this application. Figure 2In this context, the cumulative flow map can be divided into an initial cumulative region and an initial non-cumulative region based on a predetermined cumulative flow threshold. The initial cumulative region refers to the area composed of grid cells whose cumulative flow values ​​have reached the predetermined cumulative flow threshold; the initial non-cumulative region refers to the area composed of grid cells whose cumulative flow values ​​have not reached the predetermined cumulative flow threshold.

[0031] In some embodiments, calculating the difference in the mean topographic gradient between the initial confluence region and the initial non-confluence region in the confluence accumulation map specifically includes: dividing the confluence accumulation map into initial confluence regions and initial non-confluence regions according to a predetermined confluence accumulation threshold; reading the slope values ​​of each grid cell in the initial confluence region and the slope values ​​of each grid cell in the initial non-confluence region according to the topographic slope map; calculating the mean of the slope values ​​in the initial confluence region to obtain the mean topographic gradient of the confluence region; calculating the mean of the slope values ​​in the initial non-confluence region to obtain the mean topographic gradient of the non-confluence region; and determining the absolute value of the difference between the mean topographic gradient of the confluence region and the mean topographic gradient of the non-confluence region as the mean difference in topographic gradient.

[0032] It should be noted that the difference in the mean topographic gradient is not used to directly output the catchment area boundary, but rather to determine whether the entire sunken green space belongs to complex micro-topographic conditions. When the difference between the mean topographic gradient of the catchment area and the mean topographic gradient of the non-catchment area is large, it indicates that the catchment area and the non-catchment area are clearly distinguishable in terms of slope; when the difference is small, it indicates that the slope transition between the catchment area and the non-catchment area is not obvious, and the boundary representation of the original catchment accumulation map is prone to instability. Therefore, when the difference in the mean topographic gradient is lower than the threshold for dividing the catchment area of ​​complex terrain, this application proceeds to the subsequent feature benchmark extraction and feature similarity calculation process.

[0033] In step 102, when the mean difference of the terrain gradient is lower than the threshold for dividing the water catchment area of ​​complex terrain, typical depression terrain feature benchmarks and typical slope terrain feature benchmarks are extracted from the terrain point cloud data. Based on the typical depression terrain feature benchmarks and the typical slope terrain feature benchmarks, the terrain feature similarity normalization calculation is performed on the raster DEM generated by the terrain point cloud to obtain the depression feature similarity of each raster unit.

[0034] The threshold for defining complex terrain catchment areas is a pre-set judgment threshold used to determine whether the original runoff accumulation map is sufficient to stably represent the catchment area boundary. This threshold can be determined based on the average slope differences of the local garden micro-topography: if the local garden is mainly composed of gentle slopes and shallow depressions with an average slope of less than 3 degrees, it can be set to 0.5 to 1 degree; if the local garden has many micro-slope transitions and paved edge elevations with an average slope between 3 and 5 degrees, it can be set to 1 to 2 degrees. It should be noted that setting the threshold too high can lead to some simple terrains being misjudged as complex terrain, increasing unnecessary computation; setting the threshold too low can result in complex terrains not being identified, leading to division errors using traditional methods. The typical depression terrain feature benchmark refers to a raster unit or connected raster area that can represent the catchment characteristics of sunken green spaces and surrounding shallow depressions. The typical slope terrain feature benchmark refers to a raster unit or connected raster area that can represent the slope runoff transition or non-catchment slope characteristics. In this application, the feature benchmark is used to express the feature vectors of two types of typical terrain morphology.

[0035] refer to Figure 3 , Figure 3 This is a schematic diagram illustrating the extraction of typical depression terrain feature benchmarks and typical slope terrain feature benchmarks, as well as the normalization calculation process of terrain feature similarity, based on some embodiments of this application. Figure 3 In the process, local relative elevation maps, terrain slope maps, and terrain curvature maps can be generated first based on terrain point cloud data. Local relative elevation maps express the elevation relationship of a raster cell relative to the surface within its neighborhood, and are obtained by calculating the difference between the elevation value of a raster cell and the average elevation of all raster cells in its 3×3 neighborhood. Terrain curvature maps express the concavity and convexity of the landform, and are obtained by calculating the second derivative of the terrain surface. For sunken green space scenarios, areas with lower local relative elevations and concave curvature are more likely to form shallow depressions or catchment areas; areas with higher local relative elevations and steeper slopes are more likely to form slope runoff input areas or non-catchment boundary areas.

[0036] In some embodiments, extracting typical depression terrain feature benchmarks and typical slope terrain feature benchmarks from the terrain point cloud data can be done in the following ways: Based on the topographic point cloud data, a local relative elevation map, a topographic slope map, and a topographic curvature map are generated. From the local relative elevation map, the topographic slope map, the topographic curvature map, and the confluence accumulation map, the local relative elevation value, slope value, curvature value, and confluence accumulation value corresponding to each grid cell are extracted. Grid cells with local relative elevation values ​​lower than a predetermined depression elevation threshold, curvature values ​​satisfying a predetermined concave curvature range, and confluence accumulation values ​​higher than a predetermined confluence accumulation threshold are identified as depression candidate cells. Grid cells with local relative elevation values ​​higher than a predetermined slope elevation threshold, slope values ​​higher than a predetermined slope threshold, and confluence accumulation values ​​lower than a predetermined confluence accumulation threshold are identified as slope candidate cells. Spatial connectivity filtering is performed on the depression candidate cells, retaining the connected regions with the largest area and satisfying the continuity condition as typical depression topographic feature benchmarks. Spatial connectivity filtering is performed on the slope candidate cells, retaining the connected regions with the largest area and satisfying the continuity condition as typical slope topographic feature benchmarks.

[0037] Spatial connectivity filtering refers to identifying continuously distributed candidate cells as candidate connected regions based on the adjacency relationships between grid cells, and retaining candidate connected regions that meet predetermined area and continuity conditions. This avoids misidentification caused by a single abnormal grid cell, local noise elevation points, or temporary obstacles. For typical depression terrain feature benchmarks, they should have continuous concave terrain features in space, and an area of ​​not less than 1 square meter; for typical slope terrain feature benchmarks, they should have continuous slope transition features in space, and an area of ​​not less than 1 square meter.

[0038] After obtaining the typical depression terrain feature benchmark and the typical slope terrain feature benchmark, the terrain feature similarity normalization calculation can be further performed on the raster DEM. In this application, the terrain feature similarity normalization calculation is used to determine whether each raster cell is closer to the typical depression terrain feature benchmark or the typical slope terrain feature benchmark, and outputs the depression feature similarity.

[0039] It should be noted that many areas surrounding the sunken green space are not purely depressions or purely slopes, but rather a mixed terrain state between the edge of a depression, a gentle slope transition, and a rising boundary. Direct binary partitioning would cause abrupt changes in the catchment boundary. Therefore, this application uses depression feature similarity to continuously express the depression tendency of raster cells.

[0040] In some embodiments, the terrain feature similarity normalization calculation of the raster DEM generated from the terrain point cloud is performed based on the typical depression terrain feature benchmark and the typical slope terrain feature benchmark to obtain the depression feature similarity of each raster cell. Specifically, the following methods can be used: Extract the local relative elevation, slope, curvature, and cumulative runoff values ​​of the typical depression topographic feature benchmark to obtain the depression feature benchmark vector; extract the local relative elevation, slope, curvature, and cumulative runoff values ​​of the typical slope topographic feature benchmark to obtain the slope feature benchmark vector; extract the local relative elevation, slope, curvature, and cumulative runoff values ​​of each grid cell in the raster DEM to obtain the grid cell feature vector to be calculated; calculate the Euclidean distance between the grid cell feature vector to be calculated and the depression feature benchmark vector to obtain the depression feature distance; calculate the Euclidean distance between the grid cell feature vector to be calculated and the slope feature benchmark vector to obtain the slope feature distance; determine the depression feature similarity of each grid cell based on the proportion of the slope feature distance in the sum of the depression feature distance and the slope feature distance; if the sum of the depression feature distance and the slope feature distance is 0, the depression feature similarity of the grid cell is set to 0.5.

[0041] Specifically, the formula for calculating Euclidean distance is as follows: Where d is the Euclidean distance between two feature vectors, and n is the dimension of the feature vectors. Let i be the i-th component of the eigenvector of the raster cell to be calculated. It is the i-th component of the feature reference vector.

[0042] If the local relative elevation, curvature, and runoff accumulation values ​​of a grid cell are closer to the depression feature reference vector, then the depression feature distance of that grid cell is small, the slope feature distance is large, and its depression feature similarity is also high. Conversely, if the slope value and local relative elevation value of a grid cell are closer to the slope feature reference vector, then the slope feature distance of that grid cell is small, the depression feature distance is large, and its depression feature similarity is low. This continuous similarity representation can preserve the weak runoff tendency in shallow depressions, gentle slopes, and transitional edges.

[0043] In step 103, a depression feature similarity raster map is constructed based on the similarity of all depression features. The original confluence accumulation map is then weighted and corrected for depression stagnation effect using the depression feature similarity raster map to obtain a corrected confluence accumulation map.

[0044] Specifically, the depression feature similarity of each grid cell can be written into the corresponding grid location according to the grid location of the raster DEM, forming a depression feature similarity raster map. The depression feature similarity raster map can express the spatial continuous distribution of depression tendency across the entire area. Compared with the original runoff accumulation map, the depression feature similarity raster map places greater emphasis on the influence of micro-topographic morphology itself on the catchment boundary.

[0045] In some embodiments, the process of applying a weighted correction for the depression stagnation effect to the original confluence accumulation map using the depression feature similarity raster map to obtain a corrected confluence accumulation map specifically includes: dividing each raster unit into high depression similarity units, medium depression similarity units, and low depression similarity units according to the depression feature similarity of each raster unit in the depression feature similarity raster map; configuring a first confluence correction coefficient for the high depression similarity units; configuring a second confluence correction coefficient for the medium depression similarity units; configuring a third confluence correction coefficient for the low depression similarity units; wherein the first confluence correction coefficient is less than the second confluence correction coefficient, the second confluence correction coefficient is less than the third confluence correction coefficient, and the value range of all confluence correction coefficients is from 0.1 to 1.0; and weighting the confluence accumulation values ​​of the corresponding raster units in the original confluence accumulation map according to the first confluence correction coefficient, the second confluence correction coefficient, and the third confluence correction coefficient to obtain the corrected confluence accumulation map.

[0046] Among them, the high depression similarity unit refers to the grid unit with a depression feature similarity greater than or equal to 0.7. It usually has more obvious depression storage characteristics and can intercept more upstream runoff. Therefore, a first runoff correction coefficient of 0.1 to 0.3 is configured. The medium depression similarity unit refers to the grid unit with a depression feature similarity greater than or equal to 0.3 and less than 0.7. It is usually located at the edge of shallow depressions or in the transition area of ​​gentle slopes. Its storage capacity is weak. Therefore, a second runoff correction coefficient of 0.3 to 0.7 is configured. The low depression similarity unit refers to the grid unit with a depression feature similarity less than 0.3. It is usually closer to the slope or non-runoff area and has almost no storage capacity. Therefore, a third runoff correction coefficient of 0.7 to 1.0 is configured. By configuring different runoff correction coefficients, this application can correct the overestimation of runoff accumulation caused by the depression storage effect in the original runoff accumulation map.

[0047] In step 104, the multi-feature composite map generated by overlaying the confluence accumulation correction map, the terrain slope map, and the depression feature similarity raster map is input into the pre-trained semantic segmentation network model, and a binarized catchment area mask is output. The contour of the catchment area mask is used as the catchment area boundary coordinate sequence.

[0048] In some embodiments, the pre-trained semantic segmentation network model is trained through the following steps: Obtain the historical runoff accumulation correction map, topographic slope map, depression feature similarity raster map, and artificial catchment area mask verified based on surface runoff monitoring data corresponding to the sample area of ​​the sunken green space. Overlay the runoff accumulation correction map, topographic slope map, and depression feature similarity raster map at the same raster position to obtain a training input multi-feature composite map. Input the training input multi-feature composite map into a semantic segmentation network model to obtain a predicted catchment area mask. Perform pixel-level segmentation between the predicted catchment area mask and the artificial catchment area mask verified based on surface runoff monitoring data. The differences are compared to obtain the mask difference results; the network parameters of the semantic segmentation network model are updated according to the mask difference results; when the mask difference results meet any of the following predetermined convergence conditions, training can be stopped and a pre-trained semantic segmentation network model can be obtained: 1. The intersection-over-union (IoU) ratio of the predicted catchment area mask and the artificial catchment area mask is greater than or equal to 0.92; 2. The average loss function value decreases by less than 0.001 over 10 consecutive training rounds; or 3. The number of training rounds reaches a preset upper limit of 200 rounds. This is only an example and is not intended to limit the specific scope of this application.

[0049] The artificial catchment area mask, verified based on surface runoff monitoring data, can be obtained by designers based on on-site runoff observation data, post-rainwater accumulation traces, road and paving slopes, storm drain locations, green space edge elevations, and historical maintenance records. Surface runoff monitoring data is used to verify the actual flow direction and inlet location of the runoff, improving the accuracy of the manual verification. The runoff accumulation correction map in the training input multi-feature composite map is used to represent the runoff characteristics after correction for depression retention effects; the topographic slope map is used to represent surface slope changes; and the depression feature similarity raster map is used to represent shallow depressions or depression tendencies. After these three types of layers are overlaid with channels at the same raster position, the semantic segmentation network model can simultaneously read runoff, slope, and depression feature similarity information at the same spatial location. The semantic segmentation network model can adopt the U-Net network structure, which has good performance in small sample image segmentation tasks.

[0050] In some embodiments, the contour of the catchment area mask can be used as the catchment area boundary coordinate sequence in the following ways: The catchment area mask is subjected to hole filling processing to obtain a filled catchment area mask; the filled catchment area mask is subjected to isolated patch removal processing: all scattered patches not connected to the main catchment area are identified, for example, isolated patches with an area of ​​less than 0.5 square meters are removed, and independent catchment area patches with an area of ​​greater than or equal to 0.5 square meters are retained to obtain a connected catchment area mask; the outer contour pixels of the connected catchment area mask are extracted to obtain a catchment area contour pixel set; according to the spatial reference information of the raster DEM, the catchment area contour pixel set is converted into a planar coordinate point set; the planar coordinate point set is sorted according to the contour tracking order to obtain the catchment area boundary coordinate sequence.

[0051] The process includes several steps: Hole filling to repair small blank areas within the catchment area mask; isolated patch removal to eliminate scattered, misidentified areas not connected to the main catchment area; and outer contour pixels to represent the boundary between the catchment area mask and non-catchment areas. After converting the outer contour pixels into a set of planar coordinate points, they can be further arranged according to the contour tracing sequence to form a catchment area boundary coordinate sequence. This catchment area boundary coordinate sequence can be imported into landscape design software or stormwater management platforms to calculate catchment area, verify the water storage capacity of sunken green spaces, determine overflow outlet locations, and assist in stormwater organization design.

[0052] This application also provides a sunken green space system for sponge city landscaping design, which includes water catchment area division units for sunken green spaces, as referenced. Figure 4 , Figure 4 This is a structural schematic diagram of a water catchment area division unit for a sunken green space according to some embodiments of this application. The water catchment area division unit for the sunken green space includes a data acquisition module 401, a processing module 402, and an execution module 403.

[0053] The acquisition module 401 is used to acquire high-precision topographic point cloud data and surface runoff monitoring data of the entire sunken green space, calculate the topographic slope map and runoff accumulation map, and calculate the mean difference of topographic gradient between the initial runoff area and the initial non-runoff area in the runoff accumulation map.

[0054] The processing module 402 is used to extract typical depression terrain feature benchmarks and typical slope terrain feature benchmarks from the terrain point cloud data when the mean difference of the terrain gradient is lower than the threshold for dividing the water catchment area of ​​complex terrain. Based on the typical depression terrain feature benchmarks and the typical slope terrain feature benchmarks, the module performs terrain feature similarity normalization calculation on the raster DEM generated from the terrain point cloud to obtain the depression feature similarity of each raster unit.

[0055] The processing module 402 is also used to construct a depression feature similarity raster map based on the similarity of all depression features, and to perform a depression stagnation effect weighted correction on the original confluence accumulation map through the depression feature similarity raster map to obtain a confluence accumulation correction map.

[0056] The execution module 403 is used to input a multi-feature composite map generated by overlaying the confluence accumulation correction map, the terrain slope map and the depression feature similarity raster map into a pre-trained semantic segmentation network model, output a binarized catchment area mask, and use the contour of the catchment area mask as the catchment area boundary coordinate sequence.

[0057] It should be noted that, Figure 1 The method for dividing the catchment area of ​​sunken green spaces in sponge city landscaping design, as shown, can be derived from... Figure 4 The method described above is the processing procedure for a functional unit within the sunken green space's catchment area division, which is part of the sponge city landscape design that incorporates rainwater management. After obtaining the catchment area boundary coordinate sequence, this system can use it to determine the service area of ​​the sunken green space, design rainwater runoff organization, verify water storage capacity, and arrange overflow facilities.

[0058] In some implementations, the acquisition module 401 can be connected to a point cloud acquisition device, a surface runoff monitoring device, or a geographic information data interface to obtain high-precision topographic point cloud data and surface runoff monitoring data. The processing module 402 can be configured as a processing unit with raster data processing capabilities and topographic feature extraction capabilities, used to generate local relative elevation maps, topographic slope maps, topographic curvature maps, runoff accumulation maps, and depression feature similarity raster maps. The execution module 403 can be configured as a processing unit with image segmentation capabilities, used to output a catchment area mask and generate a catchment area boundary coordinate sequence.

[0059] This application also provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the method for dividing the catchment area of ​​sunken green space in sponge city landscape design as described in any of the above embodiments.

[0060] The memory can be used to store high-precision topographic point cloud data, surface runoff monitoring data, raster DEMs, topographic slope maps, runoff accumulation maps, local relative elevation maps, topographic curvature maps, depression feature similarity raster maps, runoff accumulation correction maps, catchment area masks, and catchment area boundary coordinate sequences. The processor can be used to perform topographic slope calculation, runoff accumulation calculation, topographic gradient mean difference statistics, feature benchmark extraction, topographic feature similarity normalization calculation, depression stagnation effect weighted correction, semantic segmentation, and boundary coordinate sequence output.

[0061] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for dividing the catchment area of ​​sunken green space in sponge city landscape design as described in any of the above embodiments.

[0062] As can be seen from the above embodiments, this application identifies complex garden micro-topographic conditions by using the difference in the mean of topographic gradients. Under these conditions, it further extracts typical depression topographic feature benchmarks and typical slope topographic feature benchmarks, and performs a weighted correction on the original runoff accumulation map based on the depression feature similarity to compensate for the depression retention effect. Finally, it outputs the catchment area mask and the catchment area boundary coordinate sequence through a pre-trained semantic segmentation network model. Therefore, it can improve the accuracy of boundary delineation for concave green space catchment areas under complex garden micro-topographic conditions where topographic gradient differences are not significant.

[0063] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0064] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for dividing the catchment area of ​​sunken green space in sponge city landscape design, characterized in that, Includes the following steps: High-precision topographic point cloud data and surface runoff monitoring data of the entire sunken green space were collected to calculate the topographic slope map and runoff accumulation map, and the mean difference of topographic gradient between the initial runoff area and the initial non-runoff area in the runoff accumulation map was calculated. When the mean difference of the terrain gradient is lower than the threshold for dividing the water catchment area of ​​complex terrain, typical depression terrain feature benchmarks and typical slope terrain feature benchmarks are extracted from the terrain point cloud data. Based on the typical depression terrain feature benchmarks and the typical slope terrain feature benchmarks, the terrain feature similarity normalization calculation is performed on the raster digital elevation model generated by the terrain point cloud to obtain the depression feature similarity of each raster unit. A depression feature similarity raster map is constructed based on the similarity of all depression features. The original confluence accumulation map is then weighted and corrected for depression stagnation effect using the depression feature similarity raster map to obtain a corrected confluence accumulation map. The multi-feature composite map generated by overlaying the confluence accumulation correction map, the terrain slope map, and the depression feature similarity raster map is input into the pre-trained semantic segmentation network model, and the output is a binarized catchment area mask. The outline of the catchment area mask is used as the catchment area boundary coordinate sequence.

2. The method for dividing the catchment area of ​​sunken green space in sponge city landscape design according to claim 1, characterized in that, The difference in the mean topographic gradient between the initial confluence region and the initial non-confluence region in the statistical confluence accumulation map specifically includes: The confluence accumulation map is divided into initial confluence regions and initial non-confluence regions according to a predetermined confluence accumulation threshold. Based on the terrain slope map, the slope values ​​of each grid cell in the initial confluence area and the slope values ​​of each grid cell in the initial non-confluence area are read respectively; The average slope values ​​within the initial confluence area are statistically analyzed to obtain the average topographic gradient of the confluence area. The average slope values ​​within the initial non-convergence area are statistically analyzed to obtain the average topographic gradient of the non-convergence area. The absolute value of the difference between the mean topographic gradient of the confluence region and the mean topographic gradient of the non-confluence region is defined as the difference in mean topographic gradient.

3. The method for dividing the catchment area of ​​sunken green space in sponge city landscape design according to claim 1, characterized in that, Extracting typical depression terrain feature benchmarks and typical slope terrain feature benchmarks from the terrain point cloud data specifically includes: Based on the terrain point cloud data, generate a local relative elevation map, a terrain slope map, and a terrain curvature map; Extract the local relative elevation value, slope value, curvature value and confluence accumulation value corresponding to each grid cell from the local relative elevation map, the terrain slope map, the terrain curvature map and the confluence accumulation map; Grid cells whose local relative elevation value is lower than the predetermined depression elevation threshold, whose curvature value meets the predetermined concave curvature range, and whose cumulative runoff value is higher than the predetermined cumulative runoff threshold are identified as depression candidate cells. Raster cells with local relative elevation values ​​higher than a predetermined slope elevation threshold, slope values ​​higher than a predetermined slope threshold, and runoff accumulation values ​​lower than a predetermined runoff accumulation threshold are identified as candidate slope cells. Spatial connectivity screening is performed on the candidate depression units to obtain typical depression terrain feature benchmarks; Spatial connectivity screening is performed on the candidate slope units to obtain typical slope terrain feature benchmarks.

4. The method for dividing the catchment area of ​​sunken green space in sponge city landscape design according to claim 3, characterized in that, Based on the typical depression terrain feature benchmark and the typical slope terrain feature benchmark, the raster digital elevation model generated from the terrain point cloud is subjected to terrain feature similarity normalization calculation, and the depression feature similarity of each raster cell is obtained, specifically including: The local relative elevation, slope, curvature, and cumulative runoff values ​​of the typical depression topographic feature benchmark are extracted to obtain the depression feature benchmark vector; The local relative elevation, slope, curvature, and cumulative runoff values ​​of the typical slope topographic feature benchmark are extracted to obtain the slope feature benchmark vector; Extract the local relative elevation, slope, curvature and confluence accumulation values ​​of each grid cell in the grid digital elevation model to obtain the feature vector of the grid cell to be calculated; Calculate the Euclidean distance between the feature vector of the grid cell to be calculated and the reference vector of the depression feature to obtain the depression feature distance; Calculate the Euclidean distance between the feature vector of the grid cell to be calculated and the reference vector of the slope feature to obtain the slope feature distance; The depression feature similarity of each grid cell is determined based on the proportion of the slope feature distance in the depression feature distance and the sum of the slope feature distances.

5. The method for dividing the catchment area of ​​sunken green space in sponge city landscape design according to claim 1, characterized in that, The original confluence accumulation map is weighted and corrected for the confluence stagnation effect using the raster map of depression feature similarity to obtain the corrected confluence accumulation map, specifically including: Based on the similarity of depression features of each grid cell in the depression feature similarity grid map, each grid cell is divided into high depression similarity cells, medium depression similarity cells, and low depression similarity cells. A first confluence correction coefficient is configured for the high-low similarity unit; A second confluence correction coefficient is configured for the mid-depression similarity unit; A third confluence correction coefficient is configured for the low-lying area similarity unit; Wherein, the first bus correction coefficient is less than the second bus correction coefficient, the second bus correction coefficient is less than the third bus correction coefficient, and the value range of all bus correction coefficients is from 0.1 to 1.0; Based on the first, second, and third confluence correction coefficients, the confluence accumulation values ​​of the corresponding grid cells in the original confluence accumulation map are weighted to obtain the confluence accumulation correction map.

6. The method for dividing the catchment area of ​​sunken green space in sponge city landscape design according to claim 1, characterized in that, The pre-trained semantic segmentation network model is obtained through the following steps: Obtain the corrected map of runoff accumulation, topographic slope map, raster map of depression feature similarity, and artificial catchment area mask based on surface runoff monitoring data corresponding to the historical sunken green space sample area; The confluence accumulation correction map, the terrain slope map, and the depression feature similarity raster map are overlaid in the same raster position to obtain the training input multi-feature composite map. The training input multi-feature composite map is input into the semantic segmentation network model to obtain the predicted catchment area mask; The predicted catchment area mask is compared with the artificial catchment area mask verified based on surface runoff monitoring data at the pixel level to obtain the mask difference results. Update the network parameters of the semantic segmentation network model based on the mask difference results; When the mask difference results meet the predetermined convergence conditions, a pre-trained semantic segmentation network model is obtained.

7. The method for dividing the catchment area of ​​sunken green space in sponge city landscape design according to claim 1, characterized in that, Using the contour of the catchment area mask as the catchment area boundary coordinate sequence specifically includes: The water catchment area mask is subjected to a hole filling process to obtain a filled water catchment area mask; The isolated patch removal process is performed on the filled catchment area mask to obtain a connected catchment area mask; Extract the outer contour pixels of the connected catchment area mask to obtain the catchment area contour pixel set; Based on the spatial reference information of the raster digital elevation model, the set of pixels representing the catchment area is converted into a set of planar coordinate points; The set of planar coordinate points is sorted according to the contour tracing order to obtain the boundary coordinate sequence of the catchment area.

8. A sunken green space system for sponge city landscaping design that integrates rainwater management, comprising a water catchment area division unit for the sunken green space, characterized in that, The water catchment area division unit of the sunken green space specifically includes: The data acquisition module is used to collect high-precision topographic point cloud data and surface runoff monitoring data of the entire sunken green space, calculate the topographic slope map and runoff accumulation map, and calculate the mean difference of topographic gradient between the initial runoff area and the initial non-runoff area in the runoff accumulation map. The processing module is used to extract typical depression terrain feature benchmarks and typical slope terrain feature benchmarks from the terrain point cloud data when the mean difference of the terrain gradient is lower than the threshold for dividing the water catchment area of ​​complex terrain. Based on the typical depression terrain feature benchmarks and the typical slope terrain feature benchmarks, the module performs terrain feature similarity normalization calculation on the raster digital elevation model generated from the terrain point cloud to obtain the depression feature similarity of each raster unit. The processing module is also used to construct a depression feature similarity raster map based on all depression feature similarities, and to perform a depression stagnation effect weighted correction on the original confluence accumulation map using the depression feature similarity raster map to obtain a confluence accumulation correction map; The execution module is used to input a multi-feature composite map generated by overlaying the confluence accumulation correction map, the terrain slope map, and the depression feature similarity raster map into a pre-trained semantic segmentation network model, and output a binarized catchment area mask, using the contour of the catchment area mask as the catchment area boundary coordinate sequence.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the method for dividing the catchment area of ​​sunken green space in sponge city landscape design as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for dividing the catchment area of ​​sunken green space in sponge city landscape design as described in any one of claims 1 to 7.

Images