A multi-fractal-based urban green view rate dynamic evolution pattern recognition method

Abstract

The application discloses a kind of urban green visibility dynamic evolution pattern recognition method based on multiple fractals, comprising the following steps: S100, urban street green visibility data acquisition;S200, urban multiscale isochrone data acquisition, based on city road network, 5, 10, 15, 20 minutes of walkable area is obtained by OpenRouteService platform API interface;S300, green visibility dynamic evolution pattern recognition;S400, identification result demonstration, including the spatial distribution of green visibility dynamic evolution pattern, key threshold identification and spatial structure differentiation, for assisting urban greening structure optimization and scientific intervention strategy formulation.The method solves the problem of existing urban street view green visibility analysis, such as static, single scale and insufficient structure expression.

Description

Technical Field

[0001] This invention relates to the field of geographic information science, and in particular to a method for identifying the dynamic evolution pattern of urban green visibility based on multifractals. Background Technology

[0002] With the deepening of refined urban governance and green city construction, the distribution structure and perceived effect of green resources in urban spaces have received increasing attention. Traditional greening evaluation methods mostly rely on remote sensing, cadastral data, or statistical data from urban landscaping departments, which are difficult to reflect the level of visible greenery in residents' daily walking experience. To make up for this deficiency, street view images, as an important data source reflecting green visibility at the human visual scale, have been widely used in recent years to calculate the "Green View Index" (GVI) and to conduct research on the accessibility and equity of urban green spaces.

[0003] The application of street view images in urban green space visibility analysis is gradually being promoted, and the Green Visibility Index (GVI) has become an important means of assessing the green and livable environment of cities. However, related research and tools still have the following technical shortcomings: 1. Static evaluation results: Existing methods are mostly based on single-time point or regional average green visibility, ignoring the dynamic changes in green exposure along walking paths, making it difficult to reproduce the real green experience process; 2. Coarse spatial analysis: Most methods use single-scale or single-level analysis, failing to reveal the evolution pattern of green visibility in different walking access areas and its spatial heterogeneity in cities; 3. Failure to capture structural features: Existing models lack systematic classification and pattern recognition of the evolution structure of green visibility, failing to reflect the complex mechanisms and potential risks behind the trends.

[0004] Therefore, there is an urgent need for an analytical method that can comprehensively consider the multi-scale spatiotemporal dynamic characteristics, structural complexity and spatial heterogeneity of street view green visibility rate, so as to achieve more intelligent and scientific identification and evolution modeling of urban green space, and serve the early warning of green ecological risks and the optimized intervention of urban green space system. Summary of the Invention

[0005] This invention proposes a method for identifying the dynamic evolution pattern of urban green view rate based on multifractals. This method solves the problems of static approach, single scale, and insufficient structural representation in existing urban street view green view rate analysis.

[0006] To achieve this objective, the present invention adopts the following technical solution:

[0007] A method for identifying dynamic evolution patterns of urban green view rate based on multifractals includes the following steps:

[0008] S100, City Street Green View Rate Data Acquisition: Obtain the city-wide street view image data of the target city through the Baidu Map API interface, and calculate the proportion of vegetation in the street view image in the city-wide street view image data, i.e., the green view rate, based on the image semantic segmentation algorithm.

[0009] S200 and urban multi-scale walking time circle data acquisition: Based on the urban road network, the 5, 10, 15 and 20 minute walking reach areas are obtained through the OpenRouteService platform API interface;

[0010] S300, dynamic evolution pattern recognition of green visibility rate, uses an improved multifractal algorithm to identify the dynamic evolution pattern of green visibility rate within 0-20 minutes, including rising, falling, rising then falling, falling then rising, and fluctuating patterns;

[0011] S400, identification results demonstration, including the spatial distribution of the dynamic evolution pattern of green view rate, key threshold identification and spatial structure differentiation, to assist in the optimization of urban greening structure and the formulation of scientific intervention strategies.

[0012] Preferably, step S100 specifically includes the following steps:

[0013] S110 Road network data acquisition: The road network vector data of the target city is extracted through the OpenStreetMap open platform. The data is in shapefile format.

[0014] S120. Street view sampling point generation: In the GIS platform, road segments with walkable attributes are selected, and a linear interpolation method is used to generate street view image sampling points at 50-meter intervals to obtain their latitude and longitude coordinates.

[0015] S130. Street view image acquisition: Based on the coordinates of the sampling points, the panoramic image data of the street view corresponding to each point is downloaded and stored by calling the panoramic image API interface of Baidu Maps.

[0016] S140. Image semantic segmentation: Based on the acquired panoramic street view image of the whole city, the DeepLabv3+ semantic segmentation model is used to extract the greening elements in the image with high precision. The semantic segmentation model uses ResNet-269 as the backbone network and is pre-trained on the ADE20K semantic segmentation dataset to ensure that the model can effectively identify multiple types of greening semantic targets in the street view image.

[0017] Specifically, it includes seven categories of green vegetation elements: trees, grasslands, plants, flowers, mountains, hills, and palms; among them, trees are called trees, grasslands are called grass, plants are called plants, flowers are called flowers, mountains are called mountains, hills are called hills, and palms are called palms.

[0018] S150. Calculation of Green Vision Ratio (GVI) of Street View Images: Each street view image is classified pixel-by-pixel. The proportion of pixels corresponding to the seven semantic categories mentioned above is calculated to determine the GVI value of the image. The formula is:

[0019] ;

[0020] ;

[0021] Where k is the number of green semantic categories, It is the number of pixels corresponding to the green semantic objects of type 𝑖, such as trees, grasslands, plants, flowers, mountains, hills, and palms; This represents the total number of pixels in the image.

[0022] Finally, by combining the geographic coordinates of each image sampling point, the GVI value is bound to the spatial location and used as the basic data input for subsequent dynamic evolution pattern recognition.

[0023] Preferably, step S200 specifically includes the following steps:

[0024] S210, Road Network Data Acquisition: The road network vector data of the target city is extracted through the OpenStreetMap open platform, and the data is in shp format;

[0025] S220, isochronous sampling point generation: linear interpolation is performed on walkable roads in GIS software to generate isochronous sampling points at 100-meter intervals and obtain their latitude and longitude coordinates;

[0026] S230 and isochronous circle data were acquired using the OpenRouteService path planning service. Through the API interface, the walking range boundaries under time thresholds of 5, 10, 15, and 20 minutes were obtained, forming a multi-scale accessibility spatial database, which provides an analytical basis for the dynamic evolution of green view rate.

[0027] Furthermore, step S300 specifically includes the following steps:

[0028] S310. Average green visibility data of isochronous circles: Starting from each isochronous circle sampling point, the 0-20 minute walking time is divided into four continuous intervals: 0-5, 5-10, 10-15, and 15-20 minutes. Street view sampling images in each time interval are statistically analyzed, and their average green visibility values ​​are calculated to obtain a green visibility representative value sequence for each interval.

[0029] The calculation of the segmented change rate of green visibility rate (S320, 0-20 minutes) is based on the singular index principle in multifractal analysis. It calculates the local slope sequence (β1, β2, β3) of the green visibility rate sequence at each isochronous sampling point within a continuous time interval, representing the green visibility rate change rate in the three intervals: 0-5~5-10, 5-10~10-15, and 10-15~15-20, respectively. The formula for calculating the local slope is:

[0030] ;

[0031] Among them, t i Represents the i-th time period. This represents the average green visibility rate during that time period;

[0032] S330, 0-20 minute green visibility rate change trend clustering, based on the three slope vectors β1, β2, β3 of the city's isochronous sampling points, k-means clustering is performed, the number of clusters k is determined by the elbow method and the silhouette coefficient, in order to identify the typical dynamic evolution pattern of green visibility rate in the 0-20 minute reachable space, including rising, falling, rising then falling, falling then rising and fluctuating types, etc.

[0033] S340. Fractal dimension calculation of specific green view rate evolution trends: Each typical evolution pattern is mapped to the corresponding geographic spatial region, and its fractal dimension is calculated using the box-counting algorithm. The formula is as follows:

[0034] ;

[0035] in, In the box dimension algorithm, the box size is... When used to cover vectors with the same trend of change. The number of boxes required for isochronous sampling points; This is the fractal dimension corresponding to the evolutionary trend; this calculation is used to measure the structural complexity and spatial occupancy of the pattern in the city, thereby helping to understand the spatial expansion and balance of greening evolution.

[0036] Further, step S400 includes the following steps:

[0037] S410. Identification of the Geographical Distribution of Green Visibility Evolution Trends: The various typical trends of dynamic evolution of green visibility identified by clustering are mapped to urban geospatial space. The spatial distribution of the same type of trend is clustered and identified using the density clustering algorithm, namely DBSCAN, and the clustered patches of each type of trend in the city are extracted. Geometric measurements and statistics are performed on each clustered patch, including shape attribute indicators such as the number of patches, average area, and aspect ratio, in order to quantify the spatial scale and morphological characteristics of different green visibility evolution trends.

[0038] The morphological differences of clustered patches can reflect the spatial consistency and fragmentation in the process of urban greening evolution. For example, large-scale, regularly shaped clusters may reflect the structural clarity and stability of greening development, while small-scale, fragmented patches may indicate the fragmented development of greening structure, showing greater flexibility and diversity potential. The above indicators provide data support for subsequent optimization of urban green space structure and differentiated intervention.

[0039] Further, step S400 includes the following steps:

[0040] S420, Key Threshold Identification for Dynamic Evolution of Green Visibility: Statistically analyze the initial green visibility of all sampling points, i.e., the average value within a 0-5 minute walking range, and classify it into five level intervals as the initial value benchmark for dynamic evolution:

[0041] 0–0.1 represents an extremely low green visibility range;

[0042] 0.1–0.2 represents a range with relatively low green visibility.

[0043] 0.2–0.3 represents a moderate green visibility range;

[0044] 0.3–0.6 represents a relatively high green visibility range;

[0045] 0.6–1.0 represents a range with extremely high green visibility.

[0046] For each initial green view rate level, the fractal dimension of the distribution of various dynamic evolution trends in urban space is calculated again, constructing a fractal dimension spectrum. This spectrum is used to characterize the degree of occupation of different evolution trends in urban space under similar initial green view rates. By comparing the changing trends of the fractal dimension spectrum under five initial green view rate levels, the critical threshold at which the trend changes significantly is identified, which can be regarded as the key threshold. When the initial green view rate is lower or higher than this threshold, the overall evolution structure will change significantly. Therefore, this threshold can be regarded as the key point of dynamic evolution pattern transformation, providing a basis for greening risk early warning and graded intervention.

[0047] Further, step S400 includes the following steps:

[0048] S430. Spatial structural differentiation identification of dynamic evolution of green visibility rate: First, for each initial green visibility rate level, calculate the dispersion of various evolutionary trend fractal dimensions in its fractal dimension spectrum, and quantify it using the coefficient of variation (CV); the formula is:

[0049] ;

[0050] in, The standard deviation of the fractal dimension corresponding to various evolutionary trends. The CV value is its average value; it reflects the degree of difference in the dominance of different trend types in space. A CV value greater than 0.2 is generally considered highly dispersed, indicating that under similar initial green visibility, some trend types are absolutely dominant in space, while others are relatively marginalized. A CV value less than 0.2 is generally considered low dispersed, indicating that the difference in the intensity of different trend types in space is small, that is, the evolution pattern shows a relatively balanced coexistence.

[0051] Secondly, the urban area is divided into multi-scale grids ranging from 200m to 2000m. Based on five levels of initial green view rate, the Spearman correlation algorithm is used to calculate the spatial correlation between different dynamic evolution trend types under the multi-scale grid. This process can identify whether different evolution trends exhibit clustering and coexistence or obvious spatial differentiation in the multi-scale spatial structure under similar initial green view rate conditions. Combined with geographic visualization results, it can further identify which areas have more synergistic evolution and which areas have structural fragmentation, providing a more spatially targeted strategy basis for urban greening intervention.

[0052] One of the above technical solutions includes the following beneficial effects: 1. Through dynamic green visibility spatial differentiation identification, the spatial clustering and differentiation patterns of urban greening evolution can be intuitively reflected, and key greening and risk warning areas can be identified. Overall, the proposed method for identifying dynamic evolution patterns of urban green view rate not only provides a refined technical path for structural analysis and trend identification of urban streetscape green spaces, but also theoretically promotes the shift from "outcome-oriented green quantity evaluation" to "process-driven evolutionary modeling." Secondly, by combining streetscape big data with deep learning segmentation algorithms, the method achieves high-precision extraction of greening elements and automated calculation of green view rate, providing a foundation for large-scale urban greening assessment. By constructing isochronous accessibility spaces for walking, it can realistically depict the actual activity space of residents within walking reach, providing spatial data support for subsequently incorporating greening assessment into a spatiotemporally continuous walking network. Through an improved multifractal method, it enhances sensitivity to nonlinear, multi-scale green view rate sequences, identifying abnormal fluctuations and critical turning points in local trends. Thirdly, through trend clustering and fractal dimension analysis, it can identify typical patterns and characterize their structural complexity and occupancy in urban space, systematically revealing the structural characteristics and evolutionary patterns of green space distribution formed by different cities under specific geographical backgrounds and governance logics. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the overall process;

[0054] Figure 2This is a schematic diagram illustrating the process of acquiring urban street green view rate data;

[0055] Figure 3 This is a schematic diagram of the process for acquiring multi-scale walking isochronous data in the city;

[0056] Figure 4 This is a schematic diagram of the process for identifying the dynamic evolution pattern of urban green view rate;

[0057] Figure 5 This is a schematic diagram demonstrating the process of identifying the dynamic evolution of urban green view rate;

[0058] Figure 6 It involves acquiring urban road network data;

[0059] Figure 7 Street view sampling points and isochronous sampling points are generated along the city's road network.

[0060] Figure 8 It is the semantic segmentation of vegetation elements in a street view panoramic image;

[0061] Figure 9 This is a diagram illustrating the acquisition of isochron data for walking distances of 0-20 minutes.

[0062] Figure 10 This is a schematic diagram illustrating the trend of green visibility over a 0-20 minute period.

[0063] Figure 11 This is a cluster diagram illustrating the evolution trend of green visibility at all isochronous sampling points;

[0064] Figure 12 This is a geographical distribution diagram illustrating the evolution trend of green visibility.

[0065] Figure 13 It is the fractal dimension spectrum of different initial green visibility (I-GVI) dynamic evolution types.

[0066] Figure 14 This is a geographical distribution diagram showing the evolution trend of green view rate in city A when the initial green view rate is in the range of 0-0.1.

[0067] Figure 15 This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city A is in the range of 0-0.1.

[0068] Figure 16 This is a geographical distribution diagram showing the evolution trend of green view rate in city A when the initial green view rate is in the range of 0.1-0.2.

[0069] Figure 17This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city A is in the range of 0.1-0.2.

[0070] Figure 18 This is a geographical distribution diagram showing the evolution trend of green view rate in city A when the initial green view rate is in the range of 0.2-0.3.

[0071] Figure 19 This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city A is in the range of 0.2-0.3.

[0072] Figure 20 This is a geographical distribution diagram showing the evolution trend of green view rate in city A when the initial green view rate is in the range of 0.3-0.6.

[0073] Figure 21 This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city A is in the range of 0.3-0.6.

[0074] Figure 22 This is a geographical distribution diagram showing the evolution trend of green view rate in city A when the initial green view rate is in the range of 0.6-1.0.

[0075] Figure 23 This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city A is in the range of 0.6-1.0.

[0076] Figure 24 This is a geographical distribution diagram showing the evolution trend of green view rate in city B when the initial green view rate is in the range of 0-0.1.

[0077] Figure 25 This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city B is in the range of 0-0.1.

[0078] Figure 26 This is a geographical distribution diagram showing the evolution trend of green view rate in city B when the initial green view rate is in the range of 0.1-0.2.

[0079] Figure 27 This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city B is in the range of 0.1-0.2.

[0080] Figure 28 This is a geographical distribution diagram showing the evolution trend of green view rate in city B when the initial green view rate is in the range of 0.2-0.3.

[0081] Figure 29 This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city B is in the range of 0.2-0.3.

[0082] Figure 30 This is a geographical distribution diagram showing the evolution trend of green view rate in city B when the initial green view rate is in the range of 0.3-0.6.

[0083] Figure 31 This is a schematic diagram showing the geospatial correlation and differentiation of five dynamic evolution types in the 200-2000 meter range when the initial green view rate of city B is in the range of 0.3-0.6. Detailed Implementation

[0084] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0085] like Figure 1 As shown, a method for identifying the dynamic evolution pattern of urban green view rate based on multifractals includes the following steps:

[0086] S100, City Street Green View Rate Data Acquisition: Obtain the city-wide street view image data of the target city through the Baidu Map API interface, and calculate the proportion of vegetation in the street view image in the city-wide street view image data, i.e., the green view rate, based on the image semantic segmentation algorithm.

[0087] S200 and urban multi-scale walking time circle data acquisition: Based on the urban road network, the 5, 10, 15 and 20 minute walking reach areas are obtained through the OpenRouteService platform API interface;

[0088] S300, dynamic evolution pattern recognition of green visibility rate, uses an improved multifractal algorithm to identify the dynamic evolution pattern of green visibility rate within 0-20 minutes, including rising, falling, rising then falling, falling then rising, and fluctuating patterns;

[0089] S400, identification results demonstration, including the spatial distribution of the dynamic evolution pattern of green view rate, key threshold identification and spatial structure differentiation, to assist in the optimization of urban greening structure and the formulation of scientific intervention strategies.

[0090] Beneficial Effects: The solution combines street view big data with deep learning segmentation algorithms to achieve high-precision extraction of greening elements and automated calculation of green view rate, providing a foundation for large-scale urban greening assessment. By constructing pedestrian isochronous accessibility spaces, it can realistically depict residents' actual activity space within walking reach, providing spatial data support for incorporating greening assessment into spatiotemporally continuous walking networks. Through improved multifractal methods, it enhances sensitivity to nonlinear, multi-scale green view rate sequences, identifying abnormal fluctuations and critical turning points in local trends. Through trend clustering and fractal dimension analysis, it can identify typical patterns and characterize their structural complexity and occupancy in urban space, systematically revealing the structural characteristics and evolutionary patterns of green space distribution formed by different cities under specific geographical backgrounds and governance logics. By identifying key thresholds for dynamic green view rate, it establishes a threshold response model for the dynamic evolution of urban greening, constructing a forward-looking tool for risk identification and control. Through spatial differentiation identification of dynamic green view rate, it can intuitively reflect the spatial aggregation and differentiation patterns of urban greening evolution, assisting in identifying key greening areas and risk warning areas. Overall, the proposed method for identifying the dynamic evolution pattern of urban green view rate not only provides a refined technical path for the structural analysis and trend identification of urban streetscape green spaces, but also theoretically promotes the shift from "outcome-oriented green quantity assessment" to "process-driven evolutionary modeling." This framework possesses good scalability and generalization, and can be extended to various fields such as ecological security pattern monitoring, green infrastructure layout optimization, and human settlement environment quality assessment. It has broad prospects for promotion and profound practical significance in the construction of smart cities, green infrastructure, and urban adaptive governance systems.

[0091] like Figure 2 As shown, step S100 specifically includes the following steps:

[0092] S110 Road network data acquisition: The road network vector data of the target city is extracted through the OpenStreetMap open platform. The data is in shapefile format.

[0093] S120. Street view sampling point generation: In the GIS platform, road segments with walkable attributes are selected, and a linear interpolation method is used to generate street view image sampling points at 50-meter intervals to obtain their latitude and longitude coordinates.

[0094] S130. Street view image acquisition: Based on the coordinates of the sampling points, the panoramic image data of the street view corresponding to each point is downloaded and stored by calling the panoramic image API interface of Baidu Maps.

[0095] like Figure 8As shown, S140, image semantic segmentation, based on the acquired panoramic street view image of the whole city, uses the DeepLabv3+ semantic segmentation model to extract the greening elements in the image with high precision; the semantic segmentation model uses ResNet-269 as the backbone network and is pre-trained on the ADE20K semantic segmentation dataset to ensure that the model can effectively identify multiple types of greening semantic targets in the street view image.

[0096] Specifically, it includes seven categories of green vegetation elements: trees, grasslands, plants, flowers, mountains, hills, and palms; among them, trees are called trees, grasslands are called grass, plants are called plants, flowers are called flowers, mountains are called mountains, hills are called hills, and palms are called palms.

[0097] S150. Calculation of Green Vision Ratio (GVI) of Street View Images: Each street view image is classified pixel-by-pixel. The proportion of pixels corresponding to the seven semantic categories mentioned above is calculated to determine the GVI value of the image. The formula is:

[0098] ;

[0099] ;

[0100] Where k is the number of green semantic categories, It is the number of pixels corresponding to the green semantic objects of type 𝑖, such as trees, grasslands, plants, flowers, mountains, hills, and palms; This represents the total number of pixels in the image.

[0101] Finally, by combining the geographic coordinates of each image sampling point, the GVI value is bound to the spatial location and used as the basic data input for subsequent dynamic evolution pattern recognition.

[0102] Beneficial effects: This step generates GVI vector data with spatial attributes by binding the GVI value of each street view image with the latitude and longitude coordinates of the corresponding sampling point, thus improving the spatial expressiveness of the green view rate indicator. This spatialized data can be directly integrated into a GIS platform, supporting dynamic evolution pattern recognition, temporal change analysis, and spatial differentiation management, providing accurate data support for urban greening monitoring, regional greening intervention, and optimization of the living environment.

[0103] like Figure 6 The diagram shows the acquisition of urban road network data in steps S110 and S210:

[0104] The OpenStreetMap (OSM) open map platform was used to obtain complete road network data for the target city. Data was scraped by calling OSM's API interfaces (such as the Overpass API) to extract vectorized road network data containing multiple fields of attributes such as road geometry, road class (highway tag), and number of lanes (lanes tag).

[0105] To simplify the complexity of the network structure in subsequent analysis and avoid semantic confusion or duplication in the assessment of the actual pedestrian environment caused by multi-lane settings, this scheme uniformly converts multi-lane roads in the original road network data into single-lane representations. During the process, all road entities containing multiple lanes are normalized to a single centerline representation based on the "lanes" field, and redundant lateral lines are removed, thus ensuring that the spatial structure of the street network is consistent with pedestrian traffic logic. This processing strategy not only improves the computational efficiency and connectivity stability of the road network structure in isochronous circle calculations but also provides a clearer and more consistent geometric basis for subsequent isochronous sampling and spatial fractal analysis of street scene samples.

[0106] like Figure 3 As shown, step S200 specifically includes the following steps:

[0107] S210, Road Network Data Acquisition: The road network vector data of the target city is extracted through the OpenStreetMap open platform, and the data is in shp format;

[0108] S220, isochronous sampling point generation: linear interpolation is performed on walkable roads in GIS software to generate isochronous sampling points at 100-meter intervals and obtain their latitude and longitude coordinates;

[0109] S230 and isochronous circle data were acquired using the OpenRouteService path planning service. Through the API interface, the walking range boundaries under time thresholds of 5, 10, 15, and 20 minutes were obtained, forming a multi-scale accessibility spatial database, which provides an analytical basis for the dynamic evolution of green view rate.

[0110] Beneficial Effects: By introducing isochronous circle walkable accessibility modeling, this step realizes the transformation from road network topology to spatiotemporal behavioral space, constructing a walkable accessibility spatial framework with temporal attributes. This method can quantify the spatial activity boundaries of individuals under different time thresholds, thus providing realistic spatial semantic constraints for the study of the dynamic evolution of green view rate. Combined with the OpenRouteService path planning service to generate multi-scale (5, 10, 15, 20 minutes) walkable accessibility circle boundaries, a multi-scale accessibility spatial database is established, providing a spatiotemporal analysis foundation for the dynamic evolution analysis of green view rate.

[0111] like Figure 7 The diagram shows the generation of street view sampling points and isochronous sampling points along the road network in steps S120 and S220, respectively:

[0112] Based on the pre-processed road network data, street view sampling points and isochronous sampling points are generated for the road network according to different spatial resolutions.

[0113] The street view sampling points are linearly deployed along the road network at fixed intervals of 50 meters to ensure high-frequency and high-coverage acquisition of street view images within the urban road area. This resolution balances image download cost and scene variation density, making it suitable for continuous distribution modeling of street micro-environmental perception features (such as green view rate).

[0114] The isochronous circle sampling points are distributed throughout the entire road network at intervals of 100 meters. This sampling density ensures the accuracy of fractal structure characterization while controlling the complexity and redundancy of isochronous circle calculations.

[0115] like Figure 9 The image shows the acquisition of the 0-20 minute walking space range corresponding to the isochronous sampling points in step 230:

[0116] Each isochronous circle sampling point serves as the starting point for calculating the walking range. Combined with the OpenRouteService path service platform, isochronous circle regions with walking time thresholds of 5, 10, 15, and 20 minutes are generated to form the basic spatial unit for subsequent spatial multifractal analysis.

[0117] like Figure 4 As shown, step S300 specifically includes the following steps:

[0118] S310. Average green visibility data of isochronous circles: Starting from each isochronous circle sampling point, the 0-20 minute walking time is divided into four continuous intervals: 0-5, 5-10, 10-15, and 15-20 minutes. Street view sampling images in each time interval are statistically analyzed, and their average green visibility values ​​are calculated to obtain a green visibility representative value sequence for each interval.

[0119] like Figure 10 As shown, the calculation of the segmented change rate of green visibility rate (S320, 0-20 minutes) is based on the singular index principle in multifractal analysis. The local slope sequence (β1, β2, β3) of the green visibility rate sequence at each isochronous sampling point within a continuous time interval is calculated, representing the green visibility rate change rate in the three intervals: 0-5~5-10, 5-10~10-15, and 10-15~15-20, respectively. The formula for calculating the local slope is:

[0120] ;

[0121] Among them, ti Represents the i-th time period. This represents the average green visibility rate during that time period;

[0122] like Figure 11 As shown, the S330 and 0-20 minute green visibility rate change trend clustering is performed based on the three slope vectors (β1, β2, β3) of the city's isochronous sampling points. The number of clusters k is determined by the elbow method and the silhouette coefficient to identify the typical dynamic evolution pattern of green visibility rate in the 5-20 minute reachable space, including rising, falling, rising then falling, falling then rising, and fluctuating patterns.

[0123] Cluster analysis identifies different trend patterns, effectively revealing typical dynamic trends in green view rate changes across different urban areas within accessible space. This enables a structured summary of the evolutionary paths of multi-scale, heterogeneous green view environments. This step enhances the interpretability and guiding value of the results in practical applications such as spatial planning, green space allocation, and pedestrian environment intervention.

[0124] S340. Fractal dimension calculation of specific green view rate evolution trends: Each typical evolution pattern is mapped to the corresponding geographic spatial region, and its fractal dimension is calculated using the box-counting algorithm. The formula is as follows:

[0125] ;

[0126] in, In the box dimension algorithm, the box size is... When used to cover vectors with the same trend of change. The number of boxes required for isochronous sampling points; This is the fractal dimension corresponding to the evolutionary trend; this calculation is used to measure the structural complexity and occupancy of the pattern in urban space, thereby helping to understand the spatial expansion and balance of greening evolution.

[0127] Beneficial effects: By dividing a continuous 20-minute walking time into four equally spaced intervals and calculating the average green view rate of each interval, the green view rate data of massive street view images is transformed into an ordered "representative value sequence". This processing preserves the changing trend over time while avoiding the interference of random fluctuations of single sample points on the overall trend, providing a stable and comparable data foundation for subsequent analysis.

[0128] Local slope sequence calculated based on the multifractal singularity index principle It can accurately capture the rate of change of green visibility in adjacent time intervals, such as the slope from 0-5 minutes to 5-10 minutes. Compared to simple difference calculation, the singularity index can better reflect the local characteristics of data in nonlinear and non-uniform changes, such as the difference between a rapid increase or a slow decrease in green visibility, thus improving the distinguishability of the trend.

[0129] By grouping the three slope vectors β1, β2, and β3 using k-means clustering, and combining the elbow method with the silhouette coefficient to determine the optimal number of clusters, the goal of automatically extracting typical evolution patterns such as "ascending" and "descending" from massive data was achieved.

[0130] The fractal dimension of each evolutionary pattern is calculated by the box dimension algorithm, which transforms the distribution of scattered sampling points in geographic space into a value that reflects the structural complexity. For example, the higher the fractal dimension, the stronger the pattern’s ability to occupy space and the more spatially dominant it is. This indicator makes up for the shortcomings of the traditional area ratio, which can only reflect the size of the area but cannot reflect the complexity and organizational characteristics of the spatial structure.

[0131] like Figure 12 As shown, step S400 includes the following steps:

[0132] 410. Identification of the Geographical Distribution of Green Visibility Evolution Trends: The various typical trends of dynamic evolution of green visibility identified by clustering are mapped to urban geospatial space. The spatial distribution of the same type of trend is clustered and identified using the density clustering algorithm, namely DBSCAN, and the clustered patches of each type of trend in the city are extracted. Geometric measurements and statistics are performed on each clustered patch, including shape attribute indicators such as the number of patches, average area, and aspect ratio, to quantify the spatial scale and morphological characteristics of different green visibility evolution trends.

[0133] The morphological differences of clustered patches can reflect the spatial consistency and fragmentation in the process of urban greening evolution. For example, large-scale, regularly shaped clusters may reflect the structural clarity and stability of greening development, while small-scale, fragmented patches may indicate the fragmented development of greening structure, showing greater flexibility and diversity potential. The above indicators provide data support for subsequent optimization of urban green space structure and differentiated intervention.

[0134] Mapping the green view rate evolution trend identified by clustering onto urban space enables a dual-linked analysis of time-series changes and geographical location, allowing for a visual representation of the spatial distribution characteristics of various evolution patterns within the city. This process not only enhances the intuitive identification of trend spatial clusters but also provides a geographical reference for subsequent greening intervention and optimization decisions based on regional characteristics.

[0135] like Figure 13 As shown, step S400 includes the following steps:

[0136] S420, Key Threshold Identification for Dynamic Evolution of Green Visibility: Statistically analyze the initial green visibility of all sampling points, i.e., the average value within a 0-5 minute walking range, and classify it into five level intervals as the initial value benchmark for dynamic evolution:

[0137] 0–0.1 represents an extremely low green visibility range;

[0138] 0.1–0.2 represents a range with relatively low green visibility.

[0139] 0.2–0.3 represents a moderate green visibility range;

[0140] 0.3–0.6 represents a relatively high green visibility range;

[0141] 0.6–1.0 represents a range with extremely high green visibility.

[0142] For each initial green view rate level, the fractal dimension of the distribution of various dynamic evolution trends in urban space is calculated again to construct a fractal dimension spectrum. This spectrum is used to characterize the degree of occupation of different evolution trends in urban space under similar initial green view rates. By comparing the changing trends of the fractal dimension spectrum under five initial green view rate levels, the critical threshold at which the trend changes significantly is identified, which can be regarded as the key threshold. Figure 13 The data shows that in the two cities illustrated, when the initial green view ratio (I-GVI) is below 0.2 (i.e., the ranges of 0-0.1 and 0.1-0.2), its fractal dimension spectrum trend is opposite to that of the ranges above 0.2 (0.2-0.3, 0.3-0.6, and 0.6-1.0). This indicates that 0.2 can be considered a key turning point in the evolutionary pattern of urban green view ratio. When the initial green view ratio is below or above this threshold, the overall evolutionary structure will change significantly; therefore, this threshold can be regarded as a key point for the transformation of the dynamic evolutionary pattern, providing a basis for greening risk early warning and graded intervention.

[0143] Grading the initial green view rate helps to accurately identify the evolution path under different green foundation conditions, thus providing a scientific basis for formulating tiered and differentiated urban greening intervention strategies.

[0144] like Figure 5 As shown, step S400 includes the following steps:

[0145] S430. Spatial structural differentiation identification of dynamic evolution of green visibility rate: First, for each initial green visibility rate level, calculate the dispersion of various evolutionary trend fractal dimensions in its fractal dimension spectrum, and quantify it using the coefficient of variation (CV); the formula is:

[0146] ;

[0147] in, The standard deviation of the fractal dimension corresponding to various evolutionary trends. The CV value is its average value; it reflects the degree of difference in the dominance of different trend types in space. A CV value greater than 0.2 is generally considered highly dispersed, indicating that under similar initial green visibility, some trend types are absolutely dominant in space, while others are relatively marginalized. A CV value less than 0.2 is generally considered low dispersed, indicating that the difference in the intensity of different trend types in space is small, that is, the evolution pattern shows a relatively balanced coexistence.

[0148] Secondly, the urban area is divided into multi-scale grids ranging from 200m to 2000m. Based on five levels of initial green view rate, the Spearman correlation algorithm is used to calculate the spatial correlation between different dynamic evolution trend types under the multi-scale grid. This process can identify whether different evolution trends exhibit clustering and coexistence or obvious spatial differentiation in the multi-scale spatial structure under similar initial green view rate conditions. Combined with geographic visualization results, it can further identify which areas have more synergistic evolution and which areas have structural fragmentation, providing a more spatially targeted strategy basis for urban greening intervention.

[0149] The scheme achieves quantitative identification of spatial structural differences in the dynamic evolution of green view rate by calculating the coefficient of variation (CV) of the fractal dimension of various evolutionary trends. The CV value characterizes the dispersion of the degree of occupation of different trend types in urban space. When CV > 0.2, it indicates that a certain trend has a significant dominance in spatial distribution, and urban greening evolution exhibits a clustering characteristic driven by a single trend. When CV < 0.2, it indicates that multiple trends coexist in a balanced manner in space, and greening evolution is affected by multiple factors, showing a pattern of spatial diversity and local synergy. This analytical method breaks through the limitations of traditional methods that rely solely on the existence of trends, and achieves a quantitative characterization of the spatial dominance and diversity structure of greening evolution.

[0150] Through discrete quantification and multi-scale correlation analysis, the spatial dominance and scale dependence of green view rate evolution are transformed into operable quantitative indicators for the first time. This deepens the scientific understanding of the heterogeneity of urban green space, upgrading from knowing that green view rate varies in different places to clearly understanding where the differences are manifested, why they are different, at what scale they are most obvious, and what impact these differences will have. It also provides support for greening intervention from priority determination to scale adaptation and regional customization, ultimately promoting the upgrading of urban green space regulation towards precision and differentiation.

[0151]

[0152] Table 1

[0153]

[0154] Table 2

[0155] Tables 1 and 2 are the spatial structural differentiation identification of the dynamic evolution of green visibility rate in step S430.

[0156] The technical principles of the present invention have been described above with reference to specific embodiments. These descriptions are merely for explaining the principles of the invention and should not be construed as limiting the scope of protection of the invention in any way. Based on this explanation, those skilled in the art can readily conceive of other specific embodiments of the invention without inventive effort, and these embodiments will all fall within the scope of protection of the present invention.

Claims

1. A method for recognizing dynamic evolution patterns of urban green view rate based on multifractals, characterized in that, Includes the following steps: S100, City Street Green View Rate Data Acquisition: Obtain the city-wide street view image data of the target city through the Baidu Map API interface, and calculate the proportion of vegetation in the street view image in the city-wide street view image data, i.e., the green view rate, based on the image semantic segmentation algorithm. S200 and urban multi-scale walking time circle data acquisition: Based on the urban road network, the 5, 10, 15 and 20 minute walking reach areas are obtained through the OpenRouteService platform API interface; S300, dynamic evolution pattern recognition of green visibility rate, uses an improved multifractal algorithm to identify the dynamic evolution pattern of green visibility rate within 0-20 minutes, including rising, falling, rising then falling, falling then rising, and fluctuating patterns; S400, identification results demonstration, including the spatial distribution of the dynamic evolution pattern of green view rate, key threshold identification and spatial structure differentiation, to assist in the optimization of urban greening structure and the formulation of scientific intervention strategies; Step S300 specifically includes the following steps: S310. Average green visibility data of isochronous circles is obtained by dividing the 0-20 minute walking time of each isochronous circle into four continuous intervals: 0-5, 5-10, 10-15, and 15-20 minutes, taking each sampling point of the isochronous circle as the starting point. Street view sampling images in each time interval are statistically analyzed, and their average green visibility values ​​are calculated to obtain the representative value of green visibility for each interval. The calculation of the segmented change rate of green visibility rate (S320, 0-20 minutes) is based on the singular index principle in multifractal analysis. It calculates the local slope sequence (β1, β2, β3) of the green visibility rate at each isochronous sampling point within a continuous time interval, representing the rate of change of green visibility rate in the three intervals: 0-5~5-10, 5-10~10-15, and 10-15~15-20, respectively. The formula for calculating the local slope is: ； Among them, t i Represents the i-th time period. This represents the average green visibility rate during that time period; S330, 0-20 minute green visibility rate change trend clustering, based on the three slope vectors β1, β2, β3 of the city's isochronous sampling points, k-means clustering is performed, the number of clusters k is determined by the elbow method and the silhouette coefficient, in order to identify the typical dynamic evolution pattern of green visibility rate in the 0-20 minute reachable space, including rising type, falling type, rising then falling type, falling then rising type and fluctuating type; S340. Fractal dimension calculation of specific green view rate evolution trends: Each typical evolution pattern is mapped to the corresponding geographic spatial region, and its fractal dimension is calculated using the box-counting algorithm. The formula is as follows: ； in, In the box dimension algorithm, the box size is... When used to cover vectors with the same trend of change. The number of boxes required for isochronous sampling points; This is the fractal dimension corresponding to the evolutionary trend; this calculation is used to measure the structural complexity and spatial occupancy of the pattern in the city, thereby helping to understand the spatial expansion and balance of greening evolution.

2. The method for identifying the dynamic evolution pattern of urban green view rate based on multifractals according to claim 1, characterized in that, Step S100 specifically includes the following steps: S110 Road network data acquisition: The road network vector data of the target city is extracted through the OpenStreetMap open platform. The data is in shapefile format. S120. Street view sampling point generation: In the GIS platform, road segments with walkable attributes are selected, and a linear interpolation method is used to generate street view image sampling points at 50-meter intervals to obtain their latitude and longitude coordinates. S130. Street view image acquisition: Based on the coordinates of the sampling points, the panoramic image data of the street view corresponding to each point is downloaded and stored by calling the panoramic image API interface of Baidu Maps. S140. Image semantic segmentation: Based on the acquired panoramic street view image of the whole city, the DeepLabv3+ semantic segmentation model is used to extract the greening elements in the image with high precision. The semantic segmentation model uses ResNet-269 as the backbone network and is pre-trained on the ADE20K semantic segmentation dataset to ensure that the model can effectively identify multiple types of greening semantic targets in the street view image. Specifically, it includes seven categories of green vegetation elements: trees, grasslands, plants, flowers, mountains, hills, and palms; among them, trees are called trees, grasslands are called grass, plants are called plants, flowers are called flowers, mountains are called mountains, hills are called hills, and palms are called palms. S150. Calculation of Green Vision Ratio (GVI) of Street View Images: Each street view image is classified pixel-by-pixel. The proportion of pixels corresponding to the seven semantic categories mentioned above is calculated to determine the GVI value of the image. The formula is: ； ； Where k is the number of green semantic categories, It represents the number of pixels corresponding to the green semantic object of type 𝑖, which represents trees, grassland, plants, flowers, mountains, hills, and palm trees; This represents the total number of pixels in the image. Finally, by combining the geographic coordinates of each image sampling point, the GVI value is bound to the spatial location and used as the basic data input for subsequent dynamic evolution pattern recognition.

3. The method for identifying the dynamic evolution pattern of urban green view rate based on multifractals according to claim 2, characterized in that, Step S200 specifically includes the following steps: S210, Road Network Data Acquisition: The road network vector data of the target city is extracted through the OpenStreetMap open platform, and the data is in shp format; S220, isochronous sampling point generation: linear interpolation is performed on walkable roads in GIS software to generate isochronous sampling points at 100-meter intervals and obtain their latitude and longitude coordinates; S230 and isochronous circle data were acquired using the OpenRouteService path planning service. Through the API interface, the walking range boundaries under time thresholds of 5, 10, 15, and 20 minutes were obtained, forming a multi-scale accessibility spatial database, which provides an analytical basis for the dynamic evolution of green view rate.

4. The method for identifying the dynamic evolution pattern of urban green view rate based on multifractals according to claim 3, characterized in that, Step S400 includes the following steps: S410. Identification of the Geographical Distribution of Green Visibility Evolution Trends: The various typical trends of dynamic evolution of green visibility identified by clustering are mapped to urban geospatial space. The spatial distribution of the same type of trend is clustered and identified using the density clustering algorithm, namely DBSCAN, and the clustered patches of each type of trend in the city are extracted. Geometric measurements and statistics are performed on each clustered patch, including the number of patches, average area, and aspect ratio, to quantify the spatial scale and morphological characteristics of different green visibility evolution trends. The morphological differences in clustered patches can reflect the spatial consistency and fragmentation in the process of urban greening evolution.

5. The method for identifying the dynamic evolution pattern of urban green view rate based on multifractals according to claim 4, characterized in that, Step S400 includes the following steps: S420, Key Threshold Identification for Dynamic Evolution of Green Visibility: Statistically analyze the initial green visibility of all sampling points, i.e., the average value within a 0-5 minute walking range, and classify it into five level intervals as the initial value benchmark for dynamic evolution: 0–0.1 represents an extremely low green visibility range; 0.1–0.2 represents a range with relatively low green visibility. 0.2–0.3 represents a moderate green visibility range; 0.3–0.6 represents a relatively high green visibility range; 0.6–1.0 represents a range with extremely high green visibility. For each initial green view rate level, the fractal dimension of the distribution of various dynamic evolution trends in urban space is calculated again to construct a fractal dimension spectrum. This spectrum is used to characterize the degree of occupation of different evolution trends in urban space under similar initial green view rates. By comparing the changing trends of the fractal dimension spectrum under the five initial green view rate levels, the critical threshold at which the trend changes significantly is identified as the key threshold. When the initial green view rate is lower or higher than the key threshold, the overall evolution structure will change significantly.

6. The method for recognizing the dynamic evolution pattern of urban green view rate based on multifractals according to claim 5, characterized in that, Step S400 includes the following steps: S430. Spatial structural differentiation identification of dynamic evolution of green visibility rate: First, for each initial green visibility rate level, calculate the dispersion of various evolutionary trend fractal dimensions in its fractal dimension spectrum, and quantify it using the coefficient of variation (CV); the formula is: ； in, The standard deviation of the fractal dimension corresponding to various evolutionary trends. The CV value is its average value; the CV value reflects the degree of difference in the dominance of different trend types in the space; a CV value greater than 0.2 is considered highly dispersed; a CV value less than 0.2 is generally considered low dispersed. Secondly, the urban area is divided into multi-scale grids ranging from 200m to 2000m. Based on five levels of initial green view rate, the Spearman correlation algorithm is used to calculate the spatial correlation between different dynamic evolution trend types under the multi-scale grid. This process can identify whether different evolution trends exhibit clustering and coexistence or obvious spatial differentiation in the multi-scale spatial structure under similar initial green view rate conditions. Combined with geographic visualization results, it can further identify which areas have more synergistic evolution and which areas have structural fragmentation, providing a more spatially targeted strategy basis for urban greening intervention.

Images