Post-processing method and system for converting natural resource factor mask raster data into high-quality vector polygons
By preprocessing, simplifying, and smoothing the mask raster data output by the deep learning model, and combining it with vector topology verification, the problems of irregular boundaries and topological errors in remote sensing image data conversion are solved, and the generation and automated processing of high-quality vector polygons are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGXI ZHUANG AUTONOMOUS REGION NATURAL RESOURCES SURVEY & MONITORING INST
- Filing Date
- 2026-04-21
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies face problems such as irregular boundary undulations, jagged edges, redundant polygons, and topological errors when converting remote sensing image mask raster data output by deep learning models into vector polygon data, resulting in low data quality.
Masked raster data is obtained through a deep learning model, and then preprocessed, simplified by improving the Douglas-Peucker algorithm, Gaussian smoothing and vector topology verification are performed to remove redundant elements, optimize boundaries and correct topological errors, and generate high-quality vector polygons.
It significantly improves the geometric accuracy and topology of vector polygons, increases boundary smoothness by 80%, reduces the number of polygon vertices by 60%, significantly improves data quality, adapts to scenarios with different accuracy requirements, and supports automated pipeline processing.
Smart Images
Figure CN122156221A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of remote sensing image data processing and geographic information systems, specifically relating to a post-processing method and system for converting natural resource element mask raster data into high-quality vector polygons. Background Technology
[0002] With the rapid development of artificial intelligence technology, deep learning models, especially semantic segmentation models based on encoder-decoder architectures, have become the mainstream technology in the automated identification and extraction of natural resource elements (such as arable land, forest land, construction land, and buildings). These models can perform pixel-level classification of high-resolution remote sensing images and output binary mask raster data representing the range of target elements, greatly improving the efficiency of natural resource surveys, monitoring, and supervision.
[0003] However, converting the raster mask data output by the model into vector polygon data required for actual geographic information analysis presents a series of pressing technical challenges. First, due to sensor noise, uneven illumination, ground feature shadows, and mixed pixel interference inherent in remote sensing imagery, coupled with the over-responsiveness of deep learning models to local image textures during segmentation, the generated mask boundaries exhibit numerous pixel-level irregularities. Second, when using traditional image processing methods (such as contour tracking algorithms) for raster-to-vector conversion, these boundary noises are directly inherited and amplified, resulting in severely jagged edges on the initial vector polygon boundaries. Furthermore, the conversion process also generates a large number of small redundant polygons ("holes" or "islands"), as well as topological errors such as boundary discontinuities and polygon self-intersections. Summary of the Invention
[0004] This invention aims to address the shortcomings of existing technologies and provides the following solutions: A post-processing method for converting natural resource element masked raster data into high-quality vector polygons includes the following steps: Obtain a mask grid of natural resource elements using a deep learning model; The natural resource element mask grid is preprocessed to remove redundant elements, resulting in a preliminary outline set. The preliminary contours in the preliminary contour set are subjected to coordinate transformation, and the range contours are simplified using the improved Douglas-Peucker algorithm to obtain a simplified contour set. Gaussian smoothing is used to smooth the boundaries of the simplified contours in the simplified contour set to obtain an optimized contour set. Perform vector topology compliance verification on the optimized contours in the optimized contour set, then convert the verified and repaired contours into GeoDataFrames and save them as SHP format vector files, and process them after completion.
[0005] Preferably, the method for obtaining the preliminary contour set includes: Read the natural resource element mask grid and fill the tiny holes inside the target area through morphological operations; Use OpenCV's cv2.findContours function to extract all contours in the filled mask raster and record the pixel coordinates and topological level of each contour; The pixel resolution of the contour is obtained by using a geographic transformation matrix, then the pixel area of a single pixel is calculated, and the actual geographic area of each contour is obtained based on the pixel area. A minimum effective area threshold is set, and isolated contours whose actual geographical area is smaller than the minimum effective area threshold are removed. At the same time, contours that are repeatedly identified are removed by contour coordinate relative velocity analysis, thus obtaining the preliminary contour set.
[0006] Preferably, the method for obtaining the simplified contour set includes: The preliminary contours in the preliminary contour set are converted into geographic coordinates using a geographic transformation matrix, and a simplification tolerance is set according to the image resolution. The existing Douglas-Peucker algorithm is improved, and the improved Douglas-Peucker algorithm is used to simplify the range contour to obtain a preliminary simplified contour. For the initially simplified outline, adjacent short sides with a length of less than 0.3 meters are merged to obtain the simplified outline set.
[0007] Preferably, the method for obtaining the optimized contour set includes: For the simplified contours in the simplified contour set, calculate the slope change of the line segment formed by adjacent vertices, and determine whether the contour is a straight line segment or a curved segment based on the obtained slope change rate. When the line segment is determined to be a straight line segment, the original vertex coordinates are directly retained without smoothing, ensuring that the regular shape of the photovoltaic panel is not destroyed. When the segment is identified as a curved segment, B-spline curve fitting is used to supplement the intermediate vertices through interpolation, so that the curved boundary transitions smoothly. At the same time, smoothing constraints are set to ensure smoothness without distortion. At the junction of the straight line segment and the curved segment, the vertex coordinates are adjusted by linear interpolation to obtain the optimized contour set.
[0008] Preferably, the method for performing the vector topology compliance verification includes: For the optimized contours in the optimized contour set, the Polygon.is_simple method of the Shapely library is used to determine whether there is self-intersection; If self-intersections exist, the self-intersection regions are split using the shapely.ops.unary_union function, retaining the valid parts and removing the invalid intersection regions; For the outlines of multiple natural resource elements, calculate the overlapping area of any two outlines. If the overlapping area accounts for more than 5% of the area of the smaller outline, it is determined to be an overlap anomaly. Then, an area-weighted fusion strategy is used to merge the overlapping area into the larger outline. Calculate the minimum distance between adjacent contours. If the minimum distance is less than 0.3 meters, it is determined to be an abnormal gap. The small gap is eliminated by generating a buffer and merging adjacent contours. Check if all contours are closed. If they are not closed, automatically fill in the coordinates of the last vertex to match the first vertex, and obtain the verified and repaired contour.
[0009] The present invention also provides a post-processing system for converting natural resource element mask raster data into high-quality vector polygons. The system applies the above-mentioned method and includes: a mask raster acquisition module, a preprocessing module, a simplification module, a smoothing module, and a verification and repair module. The mask grid acquisition module obtains the mask grid of natural resource elements through a deep learning model; The preprocessing module is used to preprocess the natural resource element mask grid, remove redundant elements, and obtain a preliminary outline set. The simplification module is used to perform coordinate transformation on the preliminary contours in the preliminary contour set, and to simplify the range contours using the improved Douglas-Peucker algorithm to obtain a simplified contour set. The smoothing module uses Gaussian smoothing to smooth the boundaries of the simplified contours in the simplified contour set, thereby obtaining an optimized contour set. The verification and repair module is used to perform vector topology compliance verification on the optimized contours in the optimized contour set, then convert the verified and repaired contours into GeoDataFrames and save them as SHP format vector files, and then process them.
[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Highly targeted and effectively addresses core pain points: The technical solution of this invention is specifically designed for the physical characteristics of natural resource elements (such as arable land, forest land, and construction land), and can accurately solve core problems commonly found when converting deep learning mask raster data into vector polygons, such as jagged edges on the boundary, small-area redundant noise, and abnormal topology. The average geometric deviation between the boundary of the processed vector polygon and the actual ground feature outline can be controlled within 0.3 meters, significantly improving the geometric accuracy and fidelity of the vector data. (2) Strong adaptability and wide applicability: The key processing parameters of this method (such as the simplification threshold and smoothing coefficient) can be adaptively adjusted according to the spatial resolution of the input remote sensing image and the scale of the target natural resource elements, without the need for repeated manual intervention and parameter adjustment. This adaptability enables it to flexibly adapt to masked data processing tasks with different accuracy requirements and operational scenarios, such as land change surveys and dynamic monitoring, thereby improving the robustness and practicality of the method. (3) The processed data quality is significantly improved, achieving dual optimization of spatial representation and data structure, which can fully meet the business needs of intelligent identification of large-scale natural resource elements and rapid conversion of monitoring patches. After processing by the process of this invention, the quality of vector data is significantly improved: the number of polygon vertices is reduced by more than 60% on average, effectively reducing data redundancy and optimizing data volume; the boundary smoothness is improved by more than 80%, and the spatial shape is more regular and smooth; at the same time, the data strictly passes the topological compliance verification, with no overlap, no gaps or other topological anomalies. The final output does not require secondary manual correction and can be directly applied to scenarios such as comprehensive monitoring and supervision of natural resources, change surveys and land spatial planning, efficiently supporting the rapid conversion of patches and business implementation.
[0011] (4) High degree of engineering sophistication, improving overall efficiency: The method described in this invention is implemented in Python, with a high degree of algorithm modularity. It can be seamlessly integrated with the inference process of mainstream deep learning models, forming an automated pipeline from intelligent recognition to vector post-processing. It supports parallel processing and automated deployment of large amounts of data, greatly reducing manual intervention and advancing the application of deep learning technology in the extraction of natural resource elements from "experimental output" to "engineering delivery", thus comprehensively improving production efficiency. Attached Figure Description
[0012] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation
[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0016] Example 1 In this embodiment, as Figure 1 As shown, a post-processing method for converting natural resource element masked raster data into high-quality vector polygons includes the following steps: S1. Obtain a mask raster of natural resource elements using a deep learning model. (TIFF format, 1-meter resolution, EPSG3857 coordinate system, target pixel value 1).
[0017] S2. Preprocess the natural resource element mask raster to remove redundant elements and obtain a preliminary contour set.
[0018] The method for obtaining the preliminary contour set includes: reading the natural resource element mask raster and filling the tiny holes inside the target area through morphological operations; using OpenCV's cv2.findContours function to extract all contours in the filled mask raster and recording the pixel coordinates and topological level of each contour; obtaining the pixel resolution of the contour through the geographic transformation matrix, then calculating the pixel area of a single pixel, and converting the pixel area to obtain the actual geographic area of each contour; setting a minimum effective area threshold, removing isolated contours whose actual geographic area is less than the minimum effective area threshold, and simultaneously removing duplicate contours through contour coordinate relative velocity analysis to obtain the preliminary contour set.
[0019] In this embodiment, S2 includes: ① Mask preprocessing: Read the binary mask grid output from deep learning inference, and fill the tiny holes inside the target area through morphological closing operation (kernel size is 3×3) to avoid contour splitting caused by holes.
[0020] ② Preliminary contour extraction: The cv2.findContours function of OpenCV is used to extract all contours (including outer contours and inner hole contours) in cv2.RETR_CCOMP mode, and the pixel coordinates and topological level of each contour are recorded.
[0021] ③ Geographic area conversion: Obtain the pixel resolution (unit: meters / pixel) through the geographic transformation matrix of the mask raster, calculate the geographic area of a single pixel (pixel area = pixel width × pixel height); based on the pixel area of the contour (calculated by cv2.contourArea), convert it to obtain the actual geographic area of each contour.
[0022] ④ Redundant contour filtering: Set a minimum effective area threshold (default 1m2, which can be adaptively adjusted according to image resolution) to remove isolated contours with a geographic area lower than the threshold; at the same time, through contour coordinate similarity analysis (Euclidean distance threshold set to 0.5 meters), remove duplicate contours and retain the unique effective contour. ⑤ Output: A preliminary contour set with no small-area noise and no repeated contours is obtained.
[0023] S3. Perform coordinate transformation on the preliminary contours in the preliminary contour set, and use the improved Douglas-Peucker algorithm to simplify the range contours to obtain a simplified contour set.
[0024] The method for obtaining the simplified contour set includes: converting the preliminary contours in the preliminary contour set into geographic coordinates through a geographic transformation matrix, and setting a simplification tolerance according to the image resolution; improving the existing Douglas-Peucker algorithm and using the improved Douglas-Peucker algorithm to simplify the range contours to obtain the preliminary simplified contours; for the preliminary simplified contours, merging adjacent short sides with a length of less than 0.3 meters to obtain the simplified contour set.
[0025] In this embodiment, S3 includes: ① Contour coordinate transformation: The pixel coordinates of the initial contour are converted into geographic coordinates (unit: meters) through the geographic transformation matrix of the mask raster, ensuring that the simplification process is based on real geographic space; ② Adaptive simplified tolerance setting: Set a simplified tolerance (tolerance=R×0.8) based on the image resolution (R, unit: meters / pixel). The higher the resolution, the smaller the tolerance, ensuring that boundary details are not lost in high-resolution images.
[0026] ③ Improved Douglas-Peucker Simplification: The improved Douglas-Peucker algorithm is applied to each contour. The core optimization is: when the angle between the contour line segment and the horizontal / vertical direction is less than 10°, the endpoints of the line segment are forcibly retained to avoid oversimplification of the regular boundaries. Specifically, the algorithm changes pixel-level simplification to geographic unit-level simplification by "retaining key vertices and eliminating redundant vertices," and strengthens the ability to retain straight line segments. This includes the following steps: 1. Let the contour vertex sequence be... ( , k =1,2,.., n (Geographic coordinates), simplified threshold is ϵ (distance threshold, unit: meters), connecting the first and last vertices. v 1 and vn The fitted line segment is obtained. L 2. Calculate all intermediate vertices. vk arrive L Disposal distance Dk : ; 3. Find the maximum distance Dmax = max ( Dk ):like Dmax <ϵ: Remove all intermediate vertices, retain v 1 and vn ;like Dmax ≥ϵ: Preserve the corresponding vertex vm Recursively process subsequences { v 1,..., vm}and{ vm ,..., vn}; Combine the recursive results to obtain the simplified vertex sequence.
[0027] ④ Short side merging: For the simplified outline, adjacent short sides with a length of less than 0.3 meters are merged to further reduce the number of vertices while maintaining the regularity of the outline.
[0028] ⑤ Output: A simplified outline set with reduced vertex count and preserved overall shape.
[0029] S4. Use Gaussian smoothing to smooth the boundaries of the simplified contours in the simplified contour set to obtain the optimized contour set.
[0030] The method for obtaining the optimized contour set includes: calculating the slope change of the line segments formed by adjacent vertices in the simplified contour set, and determining whether the contour is a straight line segment or a curved segment based on the obtained slope change rate; when it is determined to be a straight line segment, the original vertex coordinates are directly retained without smoothing to ensure that the regular shape of the photovoltaic panel is not destroyed; when it is determined to be a curved segment, B-spline curve fitting is used, and intermediate vertices are supplemented by interpolation to make the curved boundary transition smoothly, while setting smoothing constraints to ensure smoothness without distortion; at the junction of the straight line segment and the curved segment, the vertex coordinates are adjusted by linear interpolation to obtain the optimized contour set.
[0031] In this embodiment, S4 includes: ① Contour segmentation identification: For the simplified contour, calculate the slope change rate of the line segment formed by adjacent vertices. When the slope change rate is less than 5° / meter, it is determined to be a straight line segment; when the slope change rate is greater than 5° / meter, it is determined to be a curved segment (such as the arc transition at the edge of a photovoltaic power station).
[0032] ② Line segment preservation: For the determined line segments, the original vertex coordinates are directly retained without smoothing, ensuring that the regular shape of the photovoltaic panel is not destroyed.
[0033] ③ Smoothing of curved sections: For the determined curved sections, B-spline curve fitting (order set to 3) is used to supplement the intermediate vertices through interpolation, so that the curved boundary transitions smoothly; at the same time, a smoothing constraint condition is set: the maximum deviation distance between the fitted curve and the original contour does not exceed 0.2 meters to ensure smoothness without distortion.
[0034] ④ Boundary connection optimization: At the junction of straight and curved segments, the vertex coordinates are adjusted by linear interpolation to avoid connection breaks or abrupt angle changes.
[0035] ⑤ Output: Obtain an optimized contour set with smooth boundaries and no jagged edges.
[0036] S5. Perform vector topology compliance verification on the optimized contours in the optimized contour set, then convert the verified and repaired contours into GeoDataFrames and save them as SHP format vector files, and process them after completion.
[0037] The method for vector topology compliance verification includes: for the optimized contours in the optimized contour set, the `Polygon.is_simple` method of the Shapely library is used to determine whether there is self-intersection; if self-intersection exists, the self-intersecting region is split using the `shapely.ops.unary_union` function, retaining the valid part and removing the invalid intersection region; for contours of multiple natural resource elements, the overlapping area of any two contours is calculated. If the overlapping area accounts for more than 5% of the area of the smaller contour, it is judged as an overlap anomaly, and then an area-weighted fusion strategy is used to merge the overlapping area into the larger contour; the minimum distance between adjacent contours is calculated. If the minimum distance is less than 0.3 meters, it is judged as a gap anomaly. A buffer is generated and adjacent contours are merged to eliminate small gaps; all contours are checked for closure. If they are not closed, the coordinates of the last vertex are automatically supplemented to be consistent with the first vertex to obtain the verified and repaired contour.
[0038] In this embodiment, S5 includes: ① Self-intersection verification and repair: For each optimized contour, the Polygon.is_simple method of the Shapely library is used to determine whether there is self-intersection; if there is self-intersection, the self-intersection region is split by the shapely.ops.unary_union function, retaining the valid part and removing the invalid intersection region.
[0039] ② Overlap verification and repair: For the outlines of multiple natural resource elements, calculate the overlap area of any two outlines. If the overlap area accounts for more than 5% of the area of the smaller outline, it is judged as an overlap anomaly. The "area-weighted fusion" strategy is adopted to merge the overlapping area into the outline with the larger area to ensure the contiguousness of natural resource elements.
[0040] ③ Gap verification and repair: Calculate the minimum distance between adjacent contours. If the minimum distance is less than 0.3 meters (set to 1.5 times the image resolution), it is judged as an abnormal gap. Eliminate the small gap by generating a buffer (buffer distance = minimum distance / 2) and merging adjacent contours.
[0041] ④ Closure check and repair: Check whether all contours are closed (the coordinates of the first vertex and the last vertex are consistent). If they are not closed, automatically supplement the coordinates of the last vertex to be consistent with the first vertex to ensure the topological integrity of the vector polygon.
[0042] ⑤ Output: Convert the verified and repaired contours into GeoDataFrames, save them as SHP format vector files according to the EPSG3857 coordinate system (consistent with the mask raster), and complete the entire post-processing process.
[0043] ⑥ Output results: SHP format vector file, the average deviation between the boundary and the actual natural resource elements is 0.25 meters, the number of vertices is reduced by 68%, which meets the requirements for natural resource element area accounting.
[0044] Example 2 In this embodiment, a post-processing system for converting natural resource element mask raster data into high-quality vector polygons includes: a mask raster acquisition module, a preprocessing module, a simplification module, a smoothing module, and a verification and repair module.
[0045] The mask raster acquisition module obtains natural resource element mask rasters using a deep learning model. The preprocessing module preprocesses the natural resource element mask rasters, removing redundant elements to obtain a preliminary contour set. The simplification module performs coordinate transformation on the preliminary contours in the preliminary contour set and uses an improved Douglas-Peucker algorithm to simplify the contours, resulting in a simplified contour set. The smoothing module uses Gaussian smoothing to smooth the boundaries of the simplified contours in the simplified contour set, resulting in an optimized contour set. The verification and repair module performs vector topology compliance verification on the optimized contours in the optimized contour set, then converts the verified and repaired contours into GeoDataFrames and saves them as SHP format vector files for further processing.
[0046] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
Claims
1. A post-processing method for converting natural resource element masked raster data into high-quality vector polygons, characterized in that, Includes the following steps: Obtain a mask grid of natural resource elements using a deep learning model; The natural resource element mask grid is preprocessed to remove redundant elements, resulting in a preliminary outline set. The preliminary contours in the preliminary contour set are subjected to coordinate transformation, and the range contours are simplified using the improved Douglas-Peucker algorithm to obtain a simplified contour set. Gaussian smoothing is used to smooth the boundaries of the simplified contours in the simplified contour set to obtain an optimized contour set. Perform vector topology compliance verification on the optimized contours in the optimized contour set, then convert the verified and repaired contours into GeoDataFrames and save them as SHP format vector files, and process them after completion.
2. The post-processing method for converting natural resource element mask raster data into high-quality vector polygons according to claim 1, characterized in that, The method for obtaining the preliminary contour set includes: Read the natural resource element mask grid and fill the tiny holes inside the target area through morphological operations; Use OpenCV's cv2.findContours function to extract all contours in the filled mask raster and record the pixel coordinates and topological level of each contour; The pixel resolution of the contour is obtained by using a geographic transformation matrix, then the pixel area of a single pixel is calculated, and the actual geographic area of each contour is obtained based on the pixel area. A minimum effective area threshold is set, and isolated contours whose actual geographical area is smaller than the minimum effective area threshold are removed. At the same time, contours that are repeatedly identified are removed by contour coordinate relative velocity analysis, thus obtaining the preliminary contour set.
3. The post-processing method for converting natural resource element mask raster data into high-quality vector polygons according to claim 1, characterized in that, The method for obtaining the simplified contour set includes: The preliminary contours in the preliminary contour set are converted into geographic coordinates using a geographic transformation matrix, and a simplification tolerance is set according to the image resolution. The existing Douglas-Peucker algorithm is improved, and the improved Douglas-Peucker algorithm is used to simplify the range contour to obtain a preliminary simplified contour. For the initially simplified outline, adjacent short sides with a length of less than 0.3 meters are merged to obtain the simplified outline set.
4. The post-processing method for converting natural resource element mask raster data into high-quality vector polygons according to claim 1, characterized in that, The methods for obtaining the optimized contour set include: For the simplified contours in the simplified contour set, calculate the slope change of the line segment formed by adjacent vertices, and determine whether the contour is a straight line segment or a curved segment based on the obtained slope change rate. When the line segment is determined to be a straight line segment, the original vertex coordinates are directly retained without smoothing, ensuring that the regular shape of the photovoltaic panel is not destroyed. When the segment is identified as a curved segment, B-spline curve fitting is used to supplement the intermediate vertices through interpolation, so that the curved boundary transitions smoothly. At the same time, smoothing constraints are set to ensure smoothness without distortion. At the junction of the straight segment and the curved segment, the vertex coordinates are adjusted by linear interpolation to obtain the optimized contour set.
5. The post-processing method for converting natural resource element mask raster data into high-quality vector polygons according to claim 1, characterized in that, The method for performing the vector topology compliance verification includes: For the optimized contours in the optimized contour set, the Polygon.is_simple method of the Shapely library is used to determine whether there is self-intersection; If self-intersections exist, the self-intersection regions are split using the shapely.ops.unary_union function, retaining the valid parts and removing the invalid intersection regions; For the outlines of multiple natural resource elements, calculate the overlapping area of any two outlines. If the overlapping area accounts for more than 5% of the area of the smaller outline, it is determined to be an overlap anomaly. Then, an area-weighted fusion strategy is used to merge the overlapping area into the larger outline. Calculate the minimum distance between adjacent contours. If the minimum distance is less than 0.3 meters, it is determined to be an abnormal gap. The small gap is eliminated by generating a buffer and merging adjacent contours. Check if all contours are closed. If they are not closed, automatically fill in the coordinates of the last vertex to match the first vertex, and obtain the verified and repaired contour.
6. A post-processing system for converting natural resource element masked raster data into high-quality vector polygons, wherein the system applies the method described in any one of claims 1-5, characterized in that, include: The system includes a mask raster acquisition module, a preprocessing module, a simplification module, a smoothing module, and a verification and repair module. The mask grid acquisition module obtains the mask grid of natural resource elements through a deep learning model; The preprocessing module is used to preprocess the natural resource element mask grid, remove redundant elements, and obtain a preliminary outline set. The simplification module is used to perform coordinate transformation on the preliminary contours in the preliminary contour set, and to simplify the range contours using the improved Douglas-Peucker algorithm to obtain a simplified contour set. The smoothing module uses Gaussian smoothing to smooth the boundaries of the simplified contours in the simplified contour set, thereby obtaining an optimized contour set. The verification and repair module is used to perform vector topology compliance verification on the optimized contours in the optimized contour set, then convert the verified and repaired contours into GeoDataFrames and save them as SHP format vector files, and then process them.