A small and medium-sized city edge area extraction method based on multi-source data

Abstract

This application discloses a method for extracting the edge areas of small and medium-sized cities based on multi-source data, including: acquiring remote sensing images, POI (Point of Interest) data, light imagery, and population data of the city; analyzing the remote sensing images to obtain landscape disorder; evaluating the kernel density based on the POI data to obtain the POI kernel density; determining the nighttime light intensity based on the light imagery; determining the degree of single explanatory power of landscape disorder, POI kernel density, and nighttime light intensity on the population data; determining the weights of landscape disorder, POI kernel density, and nighttime light intensity; determining a comprehensive index for each region based on the weights; and classifying each region into multiple categories based on the comprehensive index, including the urban edge area. The model proposed in this application is significantly superior to the landscape disorder threshold method and the POI kernel density breakpoint analysis method in terms of accuracy, detail, and completeness, and has high versatility, meeting the research needs of edge areas in small and medium-sized cities.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method for extracting the edge areas of small and medium-sized cities based on multi-source data. Background Technology

[0002] As a connecting link between urban and rural areas, the urban fringe is the region where land use and spatial structure change most rapidly during urban expansion, characterized by diversity, dynamism, and transition. Rapidly and accurately identifying the urban fringe is of significant practical importance for optimizing urban spatial layout, controlling unlimited urban expansion, and protecting land resources.

[0003] Currently, methods for identifying urban fringe areas mainly include the urban-rural gradient view method, the threshold method, and the abrupt change / breakpoint analysis method. The urban-rural gradient view method identifies urban fringe areas based on spatial gradient changes in factors such as regional land use, socio-economic conditions, and population density. However, this method struggles to overcome subjectivity when determining boundary points in areas with dispersed landscape structures. The threshold method determines urban fringe areas based on threshold ranges for indicators such as distance from the built-up area, population density, building ratio, and information entropy. While simple and practical, the threshold determination typically requires repeated experiments, resulting in low efficiency, discontinuous results, and poor universality. The abrupt change / breakpoint analysis method determines urban fringe areas by calculating abrupt / breakpoint values ​​in different directions for single or combined indicators such as nighttime light intensity, impermeability index, and landscape disorder, and is currently the mainstream method.

[0004] However, research on urban edge identification has mainly focused on large cities, and there is no model applicable to the extraction of urban edges in small and medium-sized cities. Unlike large cities, the urban edges of small and medium-sized cities are often smaller, and the available data related to urban development (economy, population, light imagery) has a lower spatial resolution, which increases the difficulty of accurately identifying urban edges. Summary of the Invention

[0005] This application provides a method for extracting the edge areas of small and medium-sized cities based on multi-source data, in order to solve the problems of high difficulty, low accuracy, poor versatility, and discontinuous results in the extraction of edge areas of small and medium-sized cities by existing technical methods.

[0006] On the one hand, embodiments of this application provide a method for extracting the edge areas of small and medium-sized cities based on multi-source data, including:

[0007] Acquire remote sensing imagery, POI data, light imagery, and population data of the city;

[0008] Land use classification is performed on remote sensing images to obtain the urban landscape disorder level;

[0009] The core density of a city is assessed based on POI data to obtain the city's POI core density;

[0010] Determine the intensity of nighttime lights in a city based on light images;

[0011] Determine the individual explanatory power of landscape disorder, POI kernel density, and nighttime light intensity on population data;

[0012] The weights of landscape disorder, POI kernel density, and nighttime light intensity are determined based on the proportion of a single explanatory power to the total explanatory power.

[0013] The comprehensive index of each area in the city is determined based on its weight.

[0014] The city is divided into several categories based on a comprehensive index, including the urban fringe.

[0015] The method for extracting the edge areas of small and medium-sized cities based on multi-source data in this application has the following advantages:

[0016] Based on GF-2 imagery, POI data, NPP / VIIRS imagery, and Woldpop multi-source data, a model for extracting the edge areas of small and medium-sized cities was constructed, using landscape disorder, POI kernel density, and nighttime light intensity as urban characteristic factors. Tests in three small and medium-sized cities—Hantai District, Shangzhou District, and Hanbin District—showed that the proposed model significantly outperformed the landscape disorder threshold method and the POI kernel density breakpoint analysis method in terms of accuracy, detail, and completeness. Furthermore, it demonstrated high versatility and can meet the research needs of edge areas in small and medium-sized cities. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a method for extracting the edge areas of small and medium-sized cities based on multi-source data, provided in this application embodiment;

[0019] Figure 2 A schematic diagram of landscape disorder in Hantai District provided for an embodiment of this application;

[0020] Figure 3 A schematic diagram of kernel density under different bandwidths in the Hantai region provided for embodiments of this application;

[0021] Figure 4A schematic diagram of the urban edge area extraction results of Hantai District provided in this application embodiment;

[0022] Figure 5 A schematic diagram showing the results of extracting the urban edge areas of Shangzhou District and Hanbin District as provided in this embodiment of the application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Figure 1 A flowchart illustrating a method for extracting the edge areas of small and medium-sized cities based on multi-source data, provided in this application embodiment. This application embodiment provides a method for extracting the edge areas of small and medium-sized cities based on multi-source data, including:

[0025] S100 acquires remote sensing images, POI data, light images, and population data of the city.

[0026] For example, the data used in this application includes five categories: GF-2, POI, Woldpop, NPP / VIIRS, and administrative boundaries. Table 1 shows the specific information of the data. Among them, the remote sensing images acquired by the GF-2 satellite achieve sub-meter spatial resolution and multispectral integrated remote sensing data acquisition, and have the characteristics of high positioning accuracy, high spatial resolution, and high temporal resolution. This application uses four GF-2 remote sensing images covering the study area in July 2020.

[0027] Points of Interest (POIs) are geographic objects that can be abstracted into points, especially geographic entities closely related to people's lives. The POI data in this application comes from Amap (https: / / lbs.amap.com / ).

[0028] The light imagery uses NPP / VIIRS data, which is sourced from the National Geophysical Data Center (NGDC) and obtained by the Visible Infrared Imaging Radiometric Instrument on the Suomi NPP satellite. This application uses the monthly light data product from July 2020 provided by NGDC, with a spatial resolution of 500m.

[0029] Population data were selected from Woldpop data. This application uses Woldpop data from July 2020 at a resolution of 100m.

[0030] Table 1. Detailed Data Information

[0031]

[0032] S110 classifies land use in remote sensing images to obtain the urban landscape disorder level.

[0033] For example, S110 specifically includes: classifying land use in remote sensing images using an object-oriented SVM (Supported Vector Machine) classification method; and determining landscape disorder based on land use type.

[0034] This application employs an object-oriented SVM classification method. Object-oriented classification overcomes the limitations of traditional classification methods that use individual pixels as the basic classification and processing unit, classifying images at the object level and reducing the loss of semantic information inherent in traditional pixel-based classification methods. SVM is a classification algorithm based on the VC dimension theory and structural risk minimization principle of statistical learning theory. Compared to neural networks or traditional statistical classification methods, SVM controls model complexity by the number of vectors, eliminating the need for dimensionality reduction to decrease feature variables. Therefore, during classification, the SVM classifier does not lose feature information of ground objects, reducing the occurrence of overfitting.

[0035] The object-oriented SVM classification method combines the advantages of object-oriented multi-scale segmentation and SVM. It first performs multi-scale segmentation based on the properties of object regions in the image, considering not only spectral information but also features such as texture, geometry, and spatial topology. Then, it uses training samples for support vector machine classification. The object-oriented SVM classification method has significant advantages in accuracy, generalization, and high-dimensional data processing, and has been widely used in remote sensing image classification applications.

[0036] Furthermore, before using the object-oriented SVM classification method to classify land use in the remote sensing imagery, preprocessing is performed on the remote sensing imagery. This preprocessing includes atmospheric correction, fusion, mosaicking, and cropping operations on the remote sensing imagery.

[0037] After preprocessing the remote sensing images, land use classification can be performed to determine landscape disorder. Specifically, landscape disorder represents the degree of fragmentation and dispersion of the urban landscape, reflecting the heterogeneity and homogeneity of landscape space. The higher the heterogeneity of land use patches per unit area, the greater the landscape disorder. Urban and rural areas typically have a single land use type, mostly contiguous construction land or agricultural land, resulting in lower landscape disorder. The urban fringe is an active extension zone between the urban landscape and the agricultural hinterland, characterized by diverse land use types and higher landscape disorder. Therefore, the extent of the urban fringe can be determined by differences in landscape disorder. The formula for landscape disorder is as follows:

[0038]

[0039] In the formula, W is the landscape disorder value, and X... n This represents the percentage of land use type n within a unit area; N represents the total number of land use types per unit area.

[0040] S120 assesses the core density of a city based on POI data to obtain the city's POI core density.

[0041] For example, S120 specifically includes: evaluating the POI data using a kernel density assessment tool to obtain the POI kernel density.

[0042] Kernel density analysis is commonly used to assess the density values ​​of the neighborhood of point or line features, simulating the spatial distribution of features, and is widely applied in geospatial analysis research. Its main principle is that within a certain bandwidth, the estimated density of a feature decreases with increasing distance, with the highest kernel density at the feature's center and zero at the bandwidth edges. Kernel density analysis follows the spatial correlation law: the closer the features are, the greater their correlation. POI data also conforms to this law. The formula for POI kernel density is as follows:

[0043]

[0044] In the formula, λ (s) This calculates the kernel density of POIs in the s-th region, where r is the bandwidth set by the kernel density function, n' is the total number of all elements involved in the calculation, and d... ls It is the distance between point l and point s of POI. The weights are for distance.

[0045] Furthermore, the POI data is preprocessed before being evaluated using kernel density assessment tools. This preprocessing includes filtering and reprojection of the POI data.

[0046] S130 determines the intensity of nighttime city lights based on light images.

[0047] For example, S130 specifically includes: determining the DN (Digital Number) value of each image in the light image to obtain the nighttime light intensity.

[0048] Similarly, before determining the DN value of each image in the light imagery, the light imagery is preprocessed, which includes reprojection, denoising, and cropping of the NPP / VIIRS data.

[0049] S140, determine the degree of individual explanatory power of landscape disorder, POI kernel density, and nighttime light intensity on population data.

[0050] For example, S140 specifically includes: using the differentiation and factor detection of the geographic detector to determine the degree of individual explanatory power of landscape disorder, POI kernel density and nighttime light intensity on population data.

[0051] Geographic detectors are a set of statistical methods for detecting spatial heterogeneity and interpreting its underlying driving forces. They include heterogeneity and factor detection, interaction detection, risk zone detection, and ecological detection. The main principle of geographic detectors is that the study area is assumed to be divided into several sub-regions. If the sum of the variances of the sub-regions is less than the total variance of the region, spatial heterogeneity exists; if the spatial distributions of two variables tend to be consistent, a statistical correlation exists between them. Geographic detectors can assess spatial heterogeneity, detect explanatory factors, and analyze interactions between variables, and have been widely applied in fields such as natural sciences, environmental sciences, and human health. Differentiation and factor detection detects the spatial heterogeneity of attribute Y and the explanatory power of a factor X on attribute Y, measured by the q-value. Population density in population data is closely related to urban development. This application determines the weights of each factor based on the explanatory power of landscape disorder, POI kernel density, and nighttime light intensity on population data. The formula for q is:

[0052]

[0053]

[0054] In the formula: L represents the stratification, i.e., classification or zoning, of population data Y or landscape disorder, POI kernel density, and nighttime light intensity X; N h σ and N' are the number of units in the h-th layer and the entire region, respectively. h 2 and σ 2 S1 and S2 are the variances of the population data Y in the h-th layer and the entire region, respectively. SSW and SST are the sum of the variances within the layer and the sum of the variances of the entire region, respectively.

[0055] S150 determines the weights of landscape disorder, POI kernel density, and nighttime light intensity based on the proportion of a single explanatory power to the total explanatory power.

[0056] S160 determines the comprehensive index of each region in the city based on weights.

[0057] For example, a comprehensive index can be obtained by weighting landscape disorder, POI kernel density, nighttime light intensity, and corresponding weights.

[0058] S170 divides the city’s various areas into multiple categories based on a comprehensive index, including the city’s fringe.

[0059] For example, the natural breakpoint method can be used to divide the various areas of a city into multiple categories. This application includes three categories: the urban core area, the urban periphery area, and the rural hinterland.

[0060] The natural break method is a map grading algorithm that divides data into two groups, namely binary natural breaks. The natural break method assumes that any sequence of numbers contains natural inflection points and breaks, which can be used to group the data into groups with similar properties. The main principle of this algorithm is to cluster the numerical values, and the clustering ends when the between-group variance is maximized and the within-group variance is minimized. The natural break method is implemented in three steps:

[0061] (1) Calculate the sum of squared deviations of the array mean, SDAM:

[0062]

[0063] SDAM is the sum of squared deviations of the array mean. This is the average value of the array.

[0064] (2) For each region within a range, calculate the sum of squared deviations of the category mean, SDCM, and find the one with the smallest deviation.

[0065] (3) Calculate the variance goodness of fit (GVF):

[0066] GVF = (SDAM - SDCM) / SDAM

[0067] The GVF value ranges from 0 to 1, where 1 indicates an excellent fit and 0 indicates a very poor fit. The variance goodness of fit is used to verify whether the classification results achieve the goal of minimizing intra-class differences and maximizing inter-class differences.

[0068] After dividing the city’s various areas into multiple categories based on a comprehensive index, this application further performs raster-to-vector conversion and smoothing on the raster images obtained after category division to obtain the city’s edge area range.

[0069] Experimental instructions

[0070] Overview of the Study Area. Hantai District is located in the center of the Hanzhong Basin in southwestern Shaanxi Province, bordered by the Han River to the south and the Qinling Mountains to the north. The terrain of Hantai District slopes from north to south. The northern part is the southern slope of the Qinling Mountains, with an elevation of 700–2000 meters, accounting for 34% of the total area; the central part is a hilly area, with an elevation of 541–700 meters, accounting for 28% of the total area; and the southern part is the alluvial plain of the Han River, accounting for 38% of the total area. Hantai District is the largest commodity distribution center in southern Shaanxi and is the core area of ​​the Qinling and Bashan Mountains, possessing significant economic and ecological value. Since urban development in Hantai District is mainly concentrated in the south, this application uses the administrative boundaries of eight towns in the south to form the study area.

[0071] Results and Analysis. Landscape Disorder Thresholding Method Results: Based on preprocessed GF-2 remote sensing images of the study area, land use was classified into four categories using an object-oriented SVM classification method: vegetation (cultivated land, forest land, grassland), construction land, water bodies, and unused land. Figure 2 (a) As shown in the figure, construction land is mainly distributed in the central and southern parts of the study area, water bodies are distributed along the southern boundary, unused land is mainly distributed in the north, and vegetation is mainly distributed in the northwest, northeast, and southeast. The total vegetation area of ​​the study area is 80.34 km². 2 The construction land area is 60.83 km². 2 The water area is 6.95 km². 2 The unused land area is 3.31 km². 2 .

[0072] To highlight the layered structure of landscape disorder in the study area, a 100m×100m grid was constructed as the spatial calculation unit. ArcGIS 10.3 software was used to calculate the area ratio of vegetation, built-up land, water bodies, and unused land within each cell, and finally, the landscape disorder of the study area was calculated. Figure 2 (b) As shown in the figure, the urban core area exhibits prominent landscape structure characteristics, low landscape disorder, and concentrated low-value areas. After repeated experiments, a threshold of less than 0.46 was used as the marker for identifying the urban core area. However, when there is a large area of ​​green space in the urban core area, the landscape disorder is high, and the landscape disorder threshold method cannot identify the complete urban core area. There is no significant difference in landscape disorder between the urban fringe area and rural areas; both have high landscape disorder. Unlike large cities, small and medium-sized cities have smaller populations and villages. Furthermore, the study area is located in the Hanzhong Plain, where villages are relatively concentrated and farmland is relatively scattered, resulting in higher landscape disorder between the urban fringe area and rural areas. Although the landscape disorder threshold method has been widely used in identifying the urban fringe area of ​​large cities, in the process of identifying the urban fringe area in the study area, the shortcomings of the single-factor threshold method, such as discontinuous results, insufficient detail, and poor universality, were significantly amplified. Therefore, the landscape disorder threshold method is not suitable for identifying the urban fringe area of ​​small and medium-sized cities.

[0073] Results of POI kernel density breakpoint analysis: Based on the preprocessed POI data of the study area, the kernel density of POIs was calculated using the Kernel Density tool in ArcGIS 10.3. The bandwidth in kernel density analysis has a crucial impact on the results. Referring to existing research, kernel density analyses were performed at bandwidths of 500m, 1000m, and 1500m, respectively. The results are as follows: Figure 3 As shown in (a)-(c), the kernel density analysis results are fragmented and discontinuous when the bandwidth is 500m, and the overall distribution pattern of urban POIs is not obvious. When the bandwidth is 1500m, the local features of the overall distribution pattern of urban POIs are difficult to show, and the details are insufficient. When the bandwidth is 1000m, the kernel density analysis results have good stability, and the overall distribution pattern is obvious, which can meet the analysis needs of the urban edge area of ​​the study area.

[0074] Figure 3 (d) shows the results of classifying the POI kernel density of the study area with a bandwidth of 1000m into three categories using the natural breakpoint method. Combined with... Figure 3 As shown in (b) and (d), the study area exhibits a distinct concentric structure, with large, continuous, high-density areas in the urban core, lower density in the urban periphery, and near-zero density in most rural areas. Compared to the landscape disorder threshold method, POI kernel density analysis can more clearly and completely identify the urban core, but both methods are less effective in identifying the urban periphery. POI kernel density breakpoint analysis may identify some well-developed villages with a large number of POIs as urban peripheries. Furthermore, urban peripheries are still developing, with limited POI data and slow update rates, leading to significant errors between the results and the actual urban periphery. Unlike large cities, the POI data in the peripheries of small and medium-sized cities is less complete and updates more slowly, making it difficult for single-factor POI kernel density breakpoint analysis to accurately extract the urban periphery of the study area.

[0075] Results of the model for extracting the edge areas of small and medium-sized cities: Based on the model for extracting the edge areas of small and medium-sized cities, the landscape disorder, POI kernel density, and nighttime light intensity of the study area were first calculated. Then, the weights of each factor were determined by combining the geographic detector and Woldpop data (landscape disorder: 0.10, POI kernel density: 0.51, nighttime light intensity: 0.39), and a comprehensive index was constructed accordingly. Finally, the natural breakpoint method was used to identify the urban edge areas, and the results were post-processed. Figure 4 (a) shows the results of dividing the composite index into three categories using the natural breakpoint method; Figure 4Figure (b) shows the final result of the small and medium-sized city edge extraction model. As can be seen from the figure, the model can accurately and completely identify the urban edge boundaries of the study area. The urban edge areas of the study area are mainly concentrated in the north and east, covering an area of ​​approximately 39 km². 2 Compared to single-factor landscape disorder thresholding and POI kernel density discontinuity analysis, the performance of the urban fringe extraction model for small and medium-sized cities has been significantly improved, especially in the results for the outer boundary of the urban fringe. The overall pattern of the inner boundary (urban core area) of the urban fringe extracted by the model in this application and the POI kernel density discontinuity analysis method is relatively consistent, but the former provides more detail. The differences between the two results are mainly concentrated in the southwest of the study area. This is primarily because the southwest is the Hanjiang New Area, which is still in a rapid development phase, with faster changes in landscape patterns and slower updates to POI data. Therefore, the inner boundary of the urban fringe extracted by the POI kernel density discontinuity analysis method has an error compared to the actual boundary. The model in this application emphasizes the comprehensive performance of regional landscape disorder, POI kernel density, and nighttime light intensity, with less dependence on the performance of a single factor. Therefore, it can more accurately identify the extent of the urban fringe.

[0076] Model Accuracy Evaluation: Since the landscape disorder threshold method struggles to identify the extent of urban fringe areas in the study area, this application only evaluates the accuracy of the POI kernel density breakpoint analysis method and the small-to-medium-sized city fringe extraction model. Detailed visual analysis revealed that the small-to-medium-sized city fringe extraction model significantly outperformed the POI kernel density breakpoint analysis method in terms of detail and completeness. To further evaluate the extraction accuracy of different methods, this application employed two approaches: field verification and landscape pattern index assessment. The field verification involved uniformly selecting 100 sample points along roadsides around the urban fringe areas. Figure 4 Table 2 shows the accuracy of the extraction results analyzed through field verification. The table indicates that the overall accuracy of the small and medium-sized city edge area extraction model is significantly higher than that of the POI kernel density breakpoint analysis method, reaching 98%. The POI kernel density breakpoint analysis method has a higher number of false positives, with an overall accuracy of only 67%.

[0077] Table 2 Extraction accuracy of different methods

[0078]

[0079] Landscape pattern indices are often used to assess the accuracy of urban fringe area extraction. Patch density (PD) and Shannon diversity index (SHDI) are selected to evaluate the accuracy of two methods at both the rank and landscape level. PD represents the degree of landscape fragmentation, while SHDI represents the richness and complexity of landscape types. Generally, PD and SHDI are higher in urban fringe areas and lower in urban centers and rural areas. Table 3 shows the PD and SHDI values ​​of the two methods in different regions, calculated using Fragstats 4.2 software. The table shows that in urban fringe areas, the PD and SHDI values ​​of the small-to-medium-sized city fringe extraction model are significantly higher than those of the POI kernel density breakpoint analysis method. This indicates that the former extracts landscape fragmentation, complexity, and diversity levels within the urban fringe area that are more accurately reflected by the latter. In rural hinterlands, the PD and SHDI values ​​of the small-to-medium-sized city fringe extraction model are significantly lower than those of the POI kernel density breakpoint analysis method. However, in urban core areas, the PD and SHDI values ​​are approximately the same due to the smaller difference in the boundaries extracted by the two methods. In summary, the model proposed in this application has high accuracy and can achieve precise extraction of urban fringe areas in the study area.

[0080] Table 3 PD and SHDI values ​​in different regions

[0081]

[0082] Model Generality Analysis: To further verify the applicability of the proposed model to different regions and types of small and medium-sized cities, urban fringe area identification was conducted in Shangzhou District of Shangluo City and Hanbin District of Ankang City. Shangzhou District has a typical linear urban structure and is significantly constrained by resource conditions. From... Figure 5 As shown in (a), the POI kernel density breakpoint analysis method yields incomplete results for extracting the urban fringe area of ​​Shangzhou District, mainly concentrated in the southeast. The primary reason is that the southeast is an industrial park, with few and scattered POIs, insufficient to support the identification of the urban fringe area. Hanbin District is divided by the Han River, with the southeast mainly consisting of the old city and the northwest of the new city, exhibiting a multi-center urban structure. From... Figure 5 As shown in (b), the POI kernel density breakpoint analysis method yields poor results in extracting the urban fringe areas of Hanbin District, especially in the southeastern old city area. This is mainly because the single-factor method requires extremely high data quality when extracting multi-center urban fringe areas, while the old city area is relatively underdeveloped, with a concentrated population distribution and incomplete POI data. In contrast, the small and medium-sized city fringe area extraction model can accurately and completely extract the urban fringe areas of both regions. The model identifies urban fringe areas based on the comprehensive differences in regional landscape disorder, POI kernel density, and nighttime light intensity, with less dependence on single-factor performance and adaptability to different regions and types of small and medium-sized cities.

[0083] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0084] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for extracting small and medium-sized city fringe areas based on multi-source data, characterized in that, include: Acquire remote sensing imagery, POI data, light imagery, and population data of the city; Land use classification is performed on the remote sensing images to obtain the urban landscape disorder level; The kernel density of a city is assessed based on the POI data to obtain the city's POI kernel density; The intensity of nighttime lights in the city is determined based on the light images; Determine the individual explanatory power of the landscape disorder, POI kernel density, and nighttime light intensity on the population data; The weights of the landscape disorder, POI kernel density, and nighttime light intensity are determined based on the proportion of the individual explanatory power to the total explanatory power. The comprehensive index of each area in the city is determined by a weighted sum of the landscape disorder, the POI kernel density, and the nighttime light intensity, and the corresponding weights, wherein the weights of the landscape disorder, the POI kernel density, and the nighttime light intensity are 0.1, 0.51, and 0.39, respectively. The city's various areas are divided into multiple categories based on the comprehensive index, including the urban fringe.

2. The method of claim 1, wherein, The process of classifying land use in the remote sensing image to obtain the urban landscape disorder includes: The remote sensing image was classified for land use using an object-oriented SVM classification method. The landscape disorder level is determined based on the land use type.

3. The method for extracting the edge areas of small and medium-sized cities based on multi-source data according to claim 2, characterized in that, Before performing land use classification on the remote sensing image using the object-oriented SVM classification method, the following steps are also included: The remote sensing images are preprocessed.

4. The method for extracting the edge areas of small and medium-sized cities based on multi-source data according to claim 1, characterized in that, The step of evaluating the kernel density of a city based on the POI data to obtain the city's POI kernel density includes: The POI data are evaluated using a kernel density assessment tool to obtain the POI kernel density.

5. The method for extracting the edge areas of small and medium-sized cities based on multi-source data according to claim 4, characterized in that, Before evaluating the POI data using a kernel density assessment tool, the following steps are also included: The POI data is preprocessed.

6. The method of claim 1, wherein the method is characterized by, Determining the nighttime light intensity of the city based on the light image includes: The DN value of each image in the light image is determined to obtain the nighttime light intensity.

7. The method of claim 1, wherein the method is based on multi-source data. The determination of the individual explanatory power of the landscape disorder, POI kernel density, and nighttime light intensity for the population data includes: The degree of individual explanatory power of the landscape disorder, POI kernel density, and nighttime light intensity on the population data is determined by using the differentiation and factor detection of the geographic detector.

8. The method of claim 1, wherein the method is based on multi-source data. After dividing the city's various regions into multiple categories based on the comprehensive index, the process also includes: The raster images obtained after category classification are converted from raster to vector and smoothed to obtain the urban edge area range.

Images