Building simplification algorithm selection method, device and equipment based on graph convolution network
By using a building simplification algorithm based on graph convolutional networks, the problems of slow update speed and high cost in traditional map making techniques are solved. It realizes automated processing and efficient simplification of building data, meeting the needs of rapid updates of multi-scale maps.
Patent Information
- Application Number
- CN202310739574.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-20
- Publication Date
- 2025-12-30
- Estimated Expiration
- 2043-06-20
AI Technical Summary
Traditional mapmaking techniques have long production cycles and involve a lot of manual labor, resulting in slow map updates and high costs, making it difficult to meet the needs of rapid urban development, especially in terms of multi-scale data updates for building elements.
A building simplification algorithm based on graph convolutional networks is adopted. Through high-resolution remote sensing image processing, semantic segmentation, boundary vectorization, graph structure construction and feature extraction, the simplification algorithm model is trained by graph convolutional neural networks to automatically process building data and select appropriate simplification algorithms for building simplification.
It enables batch and automated simplification of building data, improves the accuracy of building shape description and the quality of simplification results, reduces labor costs, and increases map update speed and efficiency.
Smart Images
Figure CN116758424B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cartographic generalization, and in particular to a method, apparatus, and device for selecting building simplification algorithms based on graph convolutional networks. Background Technology
[0002] Maps are essential tools for understanding the world, serving as a visual representation of spatial phenomena and often referred to as the second language of Earth science research. Maps abstract and summarize various natural and man-made elements in the geographical world, showcasing their spatial distribution, structure, and spatiotemporal variations. Scale is a crucial characteristic of maps, reflecting their level of abstraction and generalization. With rapid urbanization, building elements on maps are frequently updated, creating new demands for faster map updates. However, traditional multi-scale data production has a long cycle, causing its production speed to lag behind the pace of urban development. Furthermore, mapmaking is a complex task requiring experienced experts to perform meticulous data processing, image analysis, and information extraction, necessitating significant human involvement and investment, resulting in relatively high production costs. Therefore, the development of rapidly and continuously updated multi-scale maps is of paramount importance.
[0003] With the continuous development of digital technology, maps have expanded from simply recording features and topography and guiding directions to a wide range of applications. As a crucial means of human spatial cognition and information acquisition, maps play an increasingly important role in people's lives. With the widespread application of digital maps, people have placed higher demands on map presentation and visual effects. To meet these demands, modern map services must possess automated functions such as zoom support and adaptive web map browsing. In the current era of big data, massive amounts of geographic information data need to be presented within limited carrier space or scale, which can lead to the inaccurate display of relevant information such as element content and distribution characteristics. Therefore, more efficient and accurate map-making techniques are needed to optimize map data presentation to meet users' requirements for map presentation quality and efficiency. Building data obtained from high-resolution remote sensing imagery can effectively update large-scale maps, but during scale transformation, a similar update speed as large-scale maps cannot be achieved. Simplifying building polygons is one of the fundamental operations for solving scale transformation problems. Therefore, proposing the problem of simplifying building outlines under the constraint of target scale can effectively solve the problem of slow update speed of medium and small scale maps. By using simplification rules under the target scale, the mapping of building elements at the target scale can be realized. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to provide a method, apparatus and device for selecting building simplification algorithms based on graph convolutional networks, in order to address the shortcomings of the prior art.
[0005] According to a first aspect of the present invention, the present invention provides a building simplification algorithm selection method based on graph convolutional networks, comprising the following steps:
[0006] S1: Acquire high-resolution remote sensing images, perform cropping and data augmentation operations on the high-resolution remote sensing images, and use the HRnet method to segment the high-resolution remote sensing images to obtain a dataset of building semantic segmentation results.
[0007] S2: Vectorize the building boundaries of the building semantic segmentation result dataset to obtain the building vector dataset;
[0008] S3: Perform graph structure construction and feature extraction on the building vector dataset obtained in step S2 to obtain the building graph structure dataset;
[0009] S4: Based on the building vector dataset obtained in step S2, a classification sample library is constructed according to the building shape using manual identification, which serves as the classification label dataset;
[0010] S5: Train a simplified algorithm selection model based on graph convolutional network based on the building graph structure dataset obtained in S3 and the classification label dataset obtained in S4. After training, the trained simplified algorithm selection model is obtained.
[0011] S6: Using the simplification algorithm selection model trained in step S5, select a simplification algorithm for a single building to obtain the appropriate simplification algorithm for the building, and use the simplification algorithm to simplify the building to obtain the final building simplification result.
[0012] Further, in step S1, the steps of cropping and data augmentation of the high-resolution remote sensing image include:
[0013] S1.1: Crop the high-resolution remote sensing image to a size suitable for the semantic segmentation method;
[0014] S1.2: Perform random flipping and rotation, and add noise points to the cropped image to increase the diversity of the data.
[0015] Furthermore, step S2 specifically includes:
[0016] The building binary map results in the building semantic segmentation result dataset are vectorized using edge detection to obtain the building vector dataset.
[0017] Furthermore, step S3 specifically includes:
[0018] Construct a graph structure for the shape features of a single building in a building vector dataset, ensuring that the constructed graph structure effectively reflects the shape features of the building.
[0019] Furthermore, the step of constructing a graph structure from the shape features of individual buildings in the building vector dataset includes:
[0020] S3.1: A Deloni triangulation graph structure is generated for the boundary of a single vectorized building to strengthen the connection between nodes;
[0021] S3.2: Using the graph structure obtained in S3.1, feature extraction is performed on the building to obtain an initial feature representation of the building shape, including global features and contour-based multi-scale sequence features, covering global and local, regional and contour cognitive perspectives, which constitute the feature matrix X of each building.
[0022] Furthermore, step S4 specifically includes:
[0023] By utilizing existing simplification knowledge, the optimal simplification algorithm is used to label each building in the semantic segmentation result dataset to obtain the best simplification algorithm suitable for that building, and a classification label dataset Y is created.
[0024] The existing simplification knowledge includes: maintaining the shape and structural features of the building before and after simplification, keeping the center position unchanged before and after simplification, keeping the visual center of the building unchanged as much as possible before and after simplification, and keeping the area of the building unchanged before and after simplification.
[0025] Furthermore, step S5 specifically includes:
[0026] The feature matrix X of each building obtained in step S3 and the classification label dataset Y obtained in step S4 are respectively divided into training set and test set. The training set is input into the graph convolutional neural network to train the network. The training process is as follows:
[0027] S5.1: Input the feature matrix X of each building obtained in step S3 into the graph convolutional neural network, and after convolution calculation and softmax(), obtain...
[0028] S5.2: Through calculation The cross-entropy loss with the classification label dataset Y is used to calculate the difference between the building label dataset and the model prediction results. The calculation method is as follows:
[0029]
[0030] In the formula, y i This represents the true label probability of sample i. This represents the predicted probability, and N represents the sum of the categorical data.
[0031] S5.3: Determine whether the difference value calculated in S5.2 has converged. If yes, stop network training and select the graph convolutional network under the current convergence condition as the building simplification algorithm model; otherwise, return to step S5.1 and continue training.
[0032] Furthermore, step S6 specifically includes:
[0033] S6.1: Input the feature matrix of the building to be simplified into the simplification algorithm selection model trained in step S5 to obtain the classification result of the building;
[0034] S6.2: Select the corresponding simplification algorithm based on the classification result to obtain the final simplified result of the building.
[0035] According to a second aspect of the present invention, the present invention provides a building simplification algorithm selection device based on graph convolutional networks, for implementing the building simplification algorithm selection method, comprising the following modules:
[0036] The semantic segmentation module is used to acquire high-resolution remote sensing images, perform cropping and data augmentation operations on the high-resolution remote sensing images, and segment the high-resolution remote sensing images using the HRnet method to obtain a dataset of semantic segmentation results for buildings.
[0037] The boundary vectorization module is used to vectorize the building boundaries of the building semantic segmentation result dataset to obtain a building vector dataset.
[0038] The graph structure construction module is used to construct the graph structure and extract features from the building vector dataset to obtain the building graph structure dataset.
[0039] The label creation module is used to construct a classification sample library based on the building vector dataset and the building shape, using manual identification, as a classification label dataset;
[0040] The model training module is used to train a simplified algorithm selection model based on a graph convolutional network according to the building graph structure dataset and the classification label dataset. After training, the trained simplified algorithm selection model is obtained.
[0041] The algorithm selection module is used to select a simplification algorithm for a single building using a trained simplification algorithm selection model, obtain the appropriate simplification algorithm for the building, and then use the simplification algorithm to simplify the building to obtain the final building simplification result.
[0042] According to a third aspect of the present invention, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the building simplification algorithm selection method.
[0043] The technical solution provided by this invention has the following beneficial effects:
[0044] 1. This invention applies graph convolutional neural networks to the study of building simplification algorithm selection, which can effectively simplify the semantic segmentation results of buildings in batches and automatically;
[0045] 2. This invention introduces multi-scale features when describing the shape of buildings, making the description of building shapes more accurate. Attached Figure Description
[0046] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:
[0047] Figure 1 This is a flowchart illustrating the overall process of a building simplification algorithm selection method based on graph convolutional networks according to the present invention.
[0048] Figure 2 This invention relates to a schematic diagram of the building structure.
[0049] Figure 3 The model selected for the building simplification algorithm of this invention;
[0050] Figure 4 The diagram shows a comparison of the simplification results of the present invention, where (a) is the original data, (b) is the simplification result obtained by the present invention, (c) is the simplification result obtained by rectangular transformation, (d) is the simplification result obtained by the four nearest neighbor method, (e) is the simplification result obtained by template matching, and (f) is the simplification result obtained by recursive regression.
[0051] Figure 5 This is a schematic diagram of the structure of a building simplification algorithm selection device based on graph convolutional networks according to the present invention;
[0052] Figure 6 This is a schematic diagram of the structure of an electronic device according to the present invention. Detailed Implementation
[0053] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0054] refer to Figure 1 This invention provides a building simplification algorithm selection method based on graph convolutional networks, comprising the following steps:
[0055] S1: Acquire high-resolution remote sensing images, perform cropping and data augmentation operations on the high-resolution remote sensing images, and use the HRnet method to segment the high-resolution remote sensing images to obtain a dataset of building semantic segmentation results.
[0056] S2: Vectorize the building boundaries of the building semantic segmentation result dataset to obtain the building vector dataset;
[0057] S3: Perform graph structure construction and feature extraction on the building vector dataset obtained in step S2 to obtain the building graph structure dataset;
[0058] S4: Based on the building vector dataset obtained in step S2, a classification sample library is constructed according to the building shape using manual identification, which serves as the classification label dataset;
[0059] S5: Train a simplified algorithm selection model based on graph convolutional network based on the building graph structure dataset obtained in S3 and the classification label dataset obtained in S4. After training, the trained simplified algorithm selection model is obtained.
[0060] S6: Using the simplification algorithm selection model trained in step S5, select a simplification algorithm for a single building to obtain the appropriate simplification algorithm for the building, and use the simplification algorithm to simplify the building to obtain the final building simplification result.
[0061] Based on, but not limited to, the above methods, step S1 is implemented as follows:
[0062] S1.1: The high-resolution remote sensing image is cropped to a size suitable for the semantic segmentation method. In this invention, the HRnet method is used for semantic segmentation, so the cropping size is 256×256 pixels.
[0063] S1.2: Perform operations such as random flipping and rotation, and adding noise points on the cropped image to increase the diversity of the data.
[0064] Based on, but not limited to, the above methods, step S2 is implemented as follows:
[0065] The binary images of buildings in the semantic segmentation dataset are vectorized using the Canny operator in edge detection to obtain a vector dataset of buildings.
[0066] Based on, but not limited to, the above methods, in step S3, the graph structure construction and feature extraction of the building vector dataset requires the reasonable and effective construction of a graph structure for the shape features of individual buildings to ensure that the constructed graph structure effectively reflects the shape features of the building. Specifically, this includes:
[0067] S3.1: A Delonix triangulation is used to generate a graph structure for the boundary of a single vectorized building to strengthen the connection between nodes. The graph structure generated in this invention is as follows: Figure 2As shown, the lines connecting adjacent nodes are the boundaries of the building itself, and the edges inside the boundaries are generated by the Deloni triangulation. The main reason for using the Deloni triangulation to construct the graph structure of the building data is that no matter from which node position the triangulation is constructed, the structure it forms is unique, thus ensuring the uniqueness of the graph structure.
[0068] S3.2: Using the graph structure obtained in S3.1, feature extraction is performed on the building to obtain an initial feature representation of the building shape, including global features and contour-based multi-scale sequence features, covering the cognitive perspectives of global and local, region and contour, to obtain the feature matrix X of each building. Each row in the feature matrix represents a vertex of the building, and each column represents the feature corresponding to the vertex. This invention selects 12 indicators, including the building's area, compactness, main direction, flatness, and other indicators used to describe the location of the building's vertices. By training this model, a building graph structure dataset is obtained.
[0069] Based on, but not limited to, the above methods, the specific implementation of step S4 is as follows:
[0070] By utilizing existing simplification knowledge, the optimal simplification method is used to label each building in the semantic segmentation result dataset to obtain the best simplification method suitable for that building, and a classification label dataset Y is created.
[0071] It should be noted that existing simplification techniques, such as preserving the building's shape and structural features before and after simplification (eliminating building outline details that cannot be expressed at the target scale while maintaining consistency in human visual perception, retaining the building's main curvature and convex / concave structures, and ensuring the accuracy of local feature point positions while removing minor curvatures), and keeping the center position unchanged, should be applied as much as possible. The visual center and area of the building should also remain unchanged before and after simplification.
[0072] Based on, but not limited to, the above methods, the specific implementation process of step S5 is as follows:
[0073] The feature matrix X of each building obtained in step S3 and the classification label dataset Y obtained in step S4 are respectively divided into training set and test set. The training set is input into the graph convolutional neural network model to train the network. The model is as follows: Figure 3 As shown, the model is trained by inputting the feature matrix of the building into the network and calculating the difference between the network output and the label value. The specific training process is as follows:
[0074] S5.1: Input the feature matrix X of each building obtained in step S3 into the graph convolutional neural network, and after convolution calculation and softmax(), obtain...
[0075] S5.2: Through calculation The cross-entropy loss with the classification label dataset Y is used to calculate the difference between the building label dataset and the model prediction results. The calculation method is as follows:
[0076]
[0077] In the formula, y i This represents the true label probability of sample i. This represents the predicted probability, and N represents the sum of the categorical data.
[0078] S5.3: Determine whether the difference value calculated in S5.2 has converged. If yes, stop network training and use the graph convolutional network and parameters under the current convergence condition as the model selection for the building simplification algorithm; otherwise, return to step S5.1 and continue training.
[0079] Based on, but not limited to, the above methods, the specific implementation process of step S6 is as follows:
[0080] S6.1: Input the feature matrix of the building to be simplified into the simplification algorithm selection model trained in step S5 to obtain the classification result of the building;
[0081] S6.2: Based on the classification result, select the corresponding simplification algorithm to obtain the final simplified result of the building. The simplified result and the comparison results with other simplification algorithms are as follows: Figure 4 As shown, (a) is the original data, (b) is the simplified result obtained by this invention, (c) is the simplified result obtained by rectangular transformation, (d) is the simplified result obtained by the four nearest neighbor method, (e) is the simplified result obtained by template matching, and (f) is the simplified result obtained by recursive regression; Figure 4 It is known that the simplification results obtained by using the same simplification method are poor for buildings of different shapes. In some simplification results, the center position of the building changes greatly and the visual effect is not satisfactory. For some buildings with complex shapes, the rectangle transformation method cannot maintain the shape and structural features of the building itself. The template matching method relies heavily on the construction of the template library and cannot obtain good simplification results for buildings in some regions. However, the building simplification method based on shape selection proposed in this invention selects better simplification results for different types of buildings. The overall indicators are relatively balanced and the simplification effect is good.
[0082] The following describes a building simplification algorithm selection device based on graph convolutional networks provided by the present invention. The building simplification algorithm selection device described below can be referred to in correspondence with the building simplification algorithm selection method described above.
[0083] like Figure 5 As shown, a building simplification algorithm selection device based on graph convolutional networks includes the following modules:
[0084] The semantic segmentation module 001 is used to acquire high-resolution remote sensing images, perform cropping and data augmentation operations on the high-resolution remote sensing images, and segment the high-resolution remote sensing images using the HRnet method to obtain a dataset of semantic segmentation results for buildings.
[0085] Boundary vectorization module 002 is used to vectorize the building boundaries of the building semantic segmentation result dataset to obtain the building vector dataset;
[0086] Graph structure construction module 003 is used to construct graph structures and extract features from building vector datasets to obtain building graph structure datasets.
[0087] The label creation module 004 is used to construct a classification sample library based on the building vector dataset and the shape of the buildings, using manual identification, as a classification label dataset;
[0088] Model training module 005 is used to train a simplified algorithm selection model based on graph convolutional networks according to the building graph structure dataset and the classification label dataset. After training, the trained simplified algorithm selection model is obtained.
[0089] The algorithm selection module 006 is used to select a simplification algorithm for a single building using a trained simplification algorithm selection model, obtain the simplification algorithm applicable to the building, and use the simplification algorithm to simplify the building to obtain the final building simplification result.
[0090] like Figure 6As shown, a schematic diagram of the physical structure of an electronic device is illustrated. The electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call the logical instructions in the memory 630 to execute the steps of the above-mentioned building simplification algorithm selection method, specifically including: S1: acquiring high-resolution remote sensing images, performing cropping and data augmentation operations on the high-resolution remote sensing images, and segmenting the high-resolution remote sensing images using the HRnet method to obtain a building semantic segmentation result dataset; S2: vectorizing the building boundaries of the building semantic segmentation result dataset to obtain a building vector dataset; S3: constructing a graph structure and extracting features from the building vector dataset obtained in step S2 to obtain a building graph structure dataset; S4: based on the building vector dataset obtained in step S2, constructing a classification sample library according to the building shape using manual recognition as a classification label dataset; S5: training a simplification algorithm selection model based on a graph convolutional network based on the building graph structure dataset obtained in S3 and the classification label dataset obtained in S4, and obtaining a trained simplification algorithm selection model after training; S6: using the simplification algorithm selection model trained in step S5, selecting a simplification algorithm for a single building to obtain the applicable simplification algorithm for the building, and using the simplification algorithm to simplify the building to obtain the final building simplification result.
[0091] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0092] In another aspect, embodiments of the present invention also provide a storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps of the above-described building simplification algorithm selection method, specifically including: S1: acquiring high-resolution remote sensing images, performing cropping and data augmentation operations on the high-resolution remote sensing images, and using HR... The net method is used to segment high-resolution remote sensing images to obtain a building semantic segmentation result dataset; S2: The building semantic segmentation result dataset is vectorized to obtain a building vector dataset; S3: The building vector dataset obtained in step S2 is used to construct a graph structure and extract features to obtain a building graph structure dataset; S4: Based on the building vector dataset obtained in step S2, a classification sample library is constructed according to the building shape using manual identification, which serves as a classification label dataset; S5: A simplification algorithm selection model based on a graph convolutional network is trained based on the building graph structure dataset obtained in S3 and the classification label dataset obtained in S4. After training, the trained simplification algorithm selection model is obtained; S6: Using the simplification algorithm selection model trained in step S5, a simplification algorithm is selected for a single building to obtain the appropriate simplification algorithm for the building. The simplification algorithm is then used to simplify the building to obtain the final building simplification result.
[0093] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0094] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. The use of the terms first, second, and third, etc., does not indicate any order and can be interpreted as identifiers.
[0095] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. A building reduction algorithm selection method based on a graph convolution network, characterized in that, The method comprises the following steps: S1: obtaining high-resolution remote sensing images, performing cropping and data enhancement operations on the high-resolution remote sensing images, performing segmentation on the high-resolution remote sensing images by using an HR net method, and obtaining a building semantic segmentation result data set; S2: performing building boundary vectorization on the building semantic segmentation result data set to obtain a building vector data set; S3: performing graph structure construction and feature extraction on the building vector data set obtained in step S2 to obtain a building graph structure data set; the graph structure construction and feature extraction on the building vector data set need to perform graph structure construction on the shape feature of a single building in the building vector data set, and the constructed graph structure can effectively reflect the shape feature of the building; the step of performing graph structure construction on the shape feature of a single building in the building vector data set comprises: S3.1: generating a graph structure by using a Delaunay triangulation network for a single vectorized building boundary, so as to strengthen the connection between nodes; S3.2: performing feature extraction on the building by using the graph structure obtained in S3.1 to obtain an initial feature representation of the shape of the building, including global features and contour-based multi-scale sequence features, covering global and local, regional and contour cognitive perspectives, and constituting a feature matrix X of each building; S4: based on the building vector data set obtained in step S2, according to the shape of the building, constructing a classification sample library by using artificial recognition as a classification label data set; specifically comprising: annotating each building in the semantic segmentation result data set by using existing simplification knowledge to obtain the best simplification algorithm suitable for the building, and making a classification label data set Y; the existing simplification knowledge comprises: maintaining the shape structure feature of the building before and after simplification, maintaining the central position before and after simplification, maintaining the visual center of the building before and after simplification as much as possible, and maintaining the area of the building before and after simplification; S5: training a simplification algorithm selection model based on a graph convolution network according to the building graph structure data set obtained in S3 and the classification label data set obtained in S4, and obtaining a trained simplification algorithm selection model after training; specifically comprising: dividing the feature matrix X of each building obtained in step S3 and the classification label data set Y obtained in step S4 into a training set and a test set respectively, inputting the training set into a graph convolution neural network, and training the network; the training process is as follows: S5.1: input the feature matrix X of each building obtained in the step S3 into a graph convolutional neural network, perform convolution calculation and obtain ; S5.2: By calculating The cross-entropy loss with the classification label dataset Y is used to calculate the difference between the building label data set and the model prediction result, and the calculation method is as follows: In the formula, the true label probability of a sample is represented by the predicted probability is represented by the sum of the classification data; S5.3: determining whether the difference value calculated in S5.2 converges, if yes, stopping the network training, and taking the graph convolution network in the current convergence condition as a building simplification algorithm selection model; otherwise, returning to step S5.1 and continuing the training; S6: using the trained simplification algorithm selection model in step S5 to select a simplification algorithm for a single building, obtaining a simplification algorithm suitable for the building, and simplifying the building by using the simplification algorithm to obtain a final building simplification result.
2. The building reduction algorithm selection method of claim 1, wherein, In step S1, the step of performing cropping and data enhancement operations on the high-resolution remote sensing images comprises: S1.1: crop high-resolution remote sensing images to a size suitable for semantic segmentation method; S1.2: random flip, rotation and noise point addition operations are performed on the cropped images to increase data diversity.
3. The building reduction algorithm selection method of claim 1, wherein, Step S2 specifically includes: The building binary graph result in the building semantic segmentation result dataset is subjected to an edge detection vectorization method to obtain a building vector dataset.
4. The building reduction algorithm selection method of claim 1, wherein, Step S6 specifically includes: S6.1: input the feature matrix of the building to be simplified into the simplified algorithm selection model trained in step S5 to obtain the classification result of the building; S6.2: select the corresponding simplified algorithm through the classification result to obtain the final simplified result of the building.
5. A building reduction algorithm selection device based on a graph convolution network, for implementing the building reduction algorithm selection method according to any one of claims 1-4, characterized in that, The method comprises the following modules: A semantic segmentation module is configured to obtain high-resolution remote sensing images, crop and perform data enhancement operations on the high-resolution remote sensing images, perform segmentation on the high-resolution remote sensing images by using an HR net method, and obtain a building semantic segmentation result dataset; A boundary vectorization module is configured to perform building boundary vectorization on the building semantic segmentation result dataset to obtain a building vector dataset; A graph structure construction module is configured to perform graph structure construction and feature extraction on the building vector dataset to obtain a building graph structure dataset; A label making module is configured to construct a classification sample library as a classification label dataset by using artificial recognition based on the building vector dataset and according to building shapes; A model training module is configured to train a simplified algorithm selection model based on a graph convolution network according to the building graph structure dataset and the classification label dataset, and obtain the trained simplified algorithm selection model after the training is completed; An algorithm selection module is configured to select a simplified algorithm for a single building by using the trained simplified algorithm selection model, obtain a simplified algorithm suitable for the building, and simplify the building by using the simplified algorithm to obtain a final building simplified result.
6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the building simplified algorithm selection method according to any one of claims 1-4 when executing the program.