A method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks
Through the multi-source data collaboration method based on graph convolution network, combined with remote sensing images and socio-economic data, extracting and performing graph convolution calculations, the problem of the gap in remote sensing image quality and data mode in urban functional area recognition is solved, and high Accurate urban functional area identification.
Patent Information
- Application Number
- CN202310981012.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-07
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2043-08-07
AI Technical Summary
In the identification of urban functional areas, the existing technology has problems such as high remote sensing image quality requirements, low resolution limit prediction accuracy, large mode gap between social and economic data and remote sensing data, and difficulty in effectively integrating, resulting in low accuracy of urban functional areas.
Using a multi-source data collaboration method based on graph convolution network, we use softmax to obtain and preprocess remote sensing images and socioeconomic data, extract images and data features, and perform graph convolution calculations. Functions and cross entropy loss functions establish graph structures of multi-source data features to achieve high-precision identification of urban functional areas.
Improve the recognition accuracy of urban functional areas, overcome the problem of low recognition accuracy caused by regional land complexity and variable spectral characteristics in the traditional method, and realize high-precision recognition of urban functional areas, which has important practical application significance. .
Smart Images

Figure CN117036939B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of multi-source and multi-modal data processing, analysis and application, and more specifically, to a method for identifying urban functional areas based on multi-source data collaboration based on a graph convolutional network. Background Art
[0002] Cities are the product of human society at a certain stage of development, including multiple elements such as space, economy, population, and function. In the process of its development, urban functional areas with different functional attributes such as administrative office land, education and scientific research land, and industrial land will gradually form in the urban space. Different functional area distribution forms have caused the complexity of cities. Under the background of rapid global urbanization, new elements are constantly integrated into the urban space structure, resulting in the differentiation of functional areas at different regional scales, followed by a series of urban problems, such as single urban functional structure, spatial differentiation, and improper resource allocation. Therefore, the identification of urban functional areas has become a necessary method to scientifically and reasonably grasp the changes in urban functional space, quickly grasp the development direction of urban spatial structure, and objectively understand the operating laws of urban spatial systems.
[0003] The development of remote sensing technology can help us obtain urban surface information quickly, accurately, at low cost and periodically. It is an important tool for urban information extraction. For example, by using pixel characteristics such as spectrum and texture, urban surface coverage and land use information can be extracted and classified. However, only relying on remote sensing images to complete tasks such as the extraction of "urban construction land" and "urban expansion" monitoring will result in the inability to update urban land use information in a timely manner. The reason is that there are several shortcomings in the use of remote sensing images: ① The quality requirements of remote sensing images themselves are high when identifying and classifying urban functional areas, and low-resolution images will limit the accuracy of prediction; ② The functional types of regional plots are related to human socio-economic activities. It is difficult to distinguish between objects with similar physical properties only by remote sensing images. ③ The lack of timeliness of satellite remote sensing images, and then considering the impact of seasons, clouds, etc. on images, will further reduce the accuracy of recognition. The development of Internet technology has led to the emergence of new social media such as social networks. These different Internet social media have generated a large amount of socio-economic data, attracting more and more researchers to pay attention to this information and explore the value of it.
[0004] Combining remote sensing images with socio-economic data is conducive to achieving information complementarity and completing the task of fine identification of urban functional areas. However, remote sensing data and socio-economic data differ in their sources and data organization. Generally speaking, remote sensing images cover the study area in the form of a surface, while socio-economic data are location-based and mostly appear as points, dashes or polygons. In addition, socio-economic data may be time-based rather than space-based. It is not an easy task to fuse these multi-source data, especially to alleviate the modal gap and heterogeneity between them. Although the existing deep learning-based methods can achieve feature association by virtue of the feature mapping space they construct. However, not all features are of equal importance. Therefore, the development of high-precision recognition methods for multi-source data collaboration for urban functional areas can assist management in urban structural layout, make scientific guidance, change unreasonable spatial layout and formulate future development plans in the detailed study of the pattern of urban functions in the city, which is of great significance for ensuring the quality of engineering construction and improving production efficiency. Summary of the invention
[0005] The purpose of this invention is to address the deficiencies in the prior art and propose a method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks.
[0006] In the first aspect, a method for identifying urban functional areas based on multi-source data collaboration based on a graph convolutional network is provided, comprising:
[0007] Step 1: Obtain remote sensing images and socioeconomic data, and perform data preprocessing;
[0008] Step 2: Segment the preprocessed remote sensing image to determine the unit for image feature extraction; and determine the type of socio-economic data features;
[0009] Step 3: Extract and construct features based on the multi-source data representation;
[0010] Step 4: Concatenate the features extracted by convolution with the features of the socioeconomic data as the basic feature data for graph computing, and perform graph convolution calculations;
[0011] Step 5: Based on the results of graph calculation, identify the division form of urban functional areas.
[0012] Preferably, in step 1, the socio-economic data includes road network data, POI data and building footprint data; step 1 includes:
[0013] Step 1.1, perform geometric correction and radiation correction on remote sensing images;
[0014] Step 1.2: Simplify, merge and geometrically correct the road network data; establish buffer zones according to different levels of road networks, divide the study area into a series of road network units, and provide a data basis for image segmentation;
[0015] Step 1.3, filter the POI data to exclude data without clear category description and location information; and reclassify the POI data according to the preset planning standards, and then transform the coordinate system of the POI data;
[0016] Step 1.4: Compare the building footprint data with the remote sensing image, and delete or add the building footprint data based on the comparison results.
[0017] Preferably, in step 2, the type of determining the socio-economic data feature includes: calculating the POI frequency feature within the road network unit; and calculating the area, perimeter and number of floors features of the buildings within the road network unit.
[0018] Preferably, in step 2, the formula for calculating the POI frequency feature is:
[0019]
[0020]
[0021] Where i represents the POI type; n i N represents the number of the i-th POI type in the block; i represents the total number of POIs of the i-th type; F i represents the frequency density of the POI of the i-th type in the total number of POIs of this type; C i Represents the ratio of the frequency density of the ith type of POI to the frequency density of all types of POIs in the block.
[0022] Preferably, in step 3, a convolutional network is used to extract the spectrum and texture features of the remote sensing image. The calculation formula for the convolutional network feature extraction is:
[0023]
[0024] Among them, x is the convolution kernel, (K, L) represents the width and height of the convolution kernel respectively; w k,l is the weight parameter of k rows and l columns; b is the bias term; f represents nonlinear calculation; O n,m is the value at the feature map (n, m); (i, j) is the number of rows and columns that the convolution kernel slides on the image.
[0025] Preferably, in step 4, the formula for graph convolution calculation is:
[0026]
[0027] in, I is the identity matrix of the same order as the adjacency matrix A, yes The degree matrix, H i is the feature matrix of layer i, W i is the weight matrix of the i-th layer, and σ is the nonlinear activation function.
[0028] Preferably, in step 5, the softmax function and the cross entropy loss function are used to establish the optimal graph structure composed of multi-source data features, realize the complementarity of multi-source data features, and complete the identification of urban functional areas.
[0029] In a second aspect, a system for identifying urban functional areas based on multi-source data collaboration of a graph convolutional network is provided, which is used to execute any of the methods for identifying urban functional areas based on multi-source data collaboration of a graph convolutional network described in the first aspect, including:
[0030] The preprocessing module is used to obtain remote sensing images and socio-economic data and perform data preprocessing;
[0031] A determination module is used to segment the pre-processed remote sensing image, determine the unit for image feature extraction, and determine the type of socio-economic data features;
[0032] The extraction module is used to extract and construct features based on the multi-source data representation;
[0033] The splicing module is used to splice the features extracted by convolution with the features of socioeconomic data as the basic feature data for graph computing and perform graph convolution computing;
[0034] The recognition module is used to identify the urban functional area division form based on the results of graph calculation.
[0035] In a third aspect, a computer storage medium is provided, in which a computer program is stored; when the computer program is run on a computer, the computer executes any of the multi-source data collaborative urban functional area identification methods based on graph convolutional networks as described in the first aspect.
[0036] The beneficial effects of the present invention are as follows: the multi-source data collaborative urban functional area identification method based on graph convolutional network proposed in the present invention is to identify urban functional areas by combining multi-source data features. The proposed identification framework is simple, effective and easy to implement. In addition, in the identification framework, not only the spatial information of the distribution status and properties of objects provided by the GF-2 multi-spectral remote sensing data is utilized, but also the characteristics of the functional area itself are taken into account, and the data of points of interest, building outline data, road network data and remote sensing data are used for collaborative application, and the complementarity between multi-source data is fully utilized to achieve high-precision urban functional area identification. In short, the method proposed in the present invention can improve the accuracy of urban functional areas, overcome the problem of low accuracy of urban functional area identification caused by complex regional object composition and changeable spectral characteristics when traditionally using a single multi-spectral image to identify urban functional areas, and achieve high-precision urban functional area identification. Therefore, the method proposed in the present invention has important practical application significance. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 This is a flowchart of a method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks;
[0038] Figure 2 A schematic diagram showing the recognition results. DETAILED DESCRIPTION
[0039] The present invention is further described below in conjunction with embodiments. The description of the following embodiments is only used to help understand the present invention. It should be noted that for ordinary persons in the art, without departing from the principle of the present invention, the present invention can also be modified in some ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.
[0040] Embodiment 1:
[0041] In view of the shortcomings of the existing technology that multi-source data cannot be fully combined for urban functional zone identification, the present invention proposes a multi-source data collaborative urban functional zone identification method based on graph convolutional network. The identification framework first performs a preliminary analysis of the data and determines the feature extraction method based on the data feature expression form; then, the extracted features are subjected to graph calculation to establish the correlation between multi-modal features; finally, a classifier is selected to classify the established graph structure, making full use of the complementary characteristics between multi-modal data to identify urban functional zones; so that the identification result is highly accurate, easy to implement, efficient and practical.
[0042] Specifically, the present invention provides a method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks, such as Figure 1 As shown, including:
[0043] Step 1: Obtain remote sensing images and socioeconomic data, and perform data preprocessing.
[0044] Specifically, the remote sensing image can be a GF-2 remote sensing image, and the socio-economic data includes road network data, POI data, and building footprint data. The GF-2 data is preprocessed to ensure that the image data features are fully utilized. The acquired socio-economic data needs to go through a data cleaning step to reduce data redundancy and improve data availability. For example, POI data, not all POI data are useful for identifying urban functional areas, such as newsstands, public toilets, transportation stations, etc., which will interfere with the experiment, so they need to be removed from the original data. In addition, the acquired POI data must be reclassified and the data coordinate system must be corrected according to certain standards to meet the purpose of urban functional zoning. The timeliness of building outline data The actual building distribution form in some areas does not match the data, and it is necessary to correct, edit, add or delete the corresponding building outlines.
[0045] Step 1 includes:
[0046] Step 1.1: Perform geometric correction and radiation correction on remote sensing images.
[0047] In step 1.1, accurate registration images can be obtained through geometric correction and radiation correction, while eliminating spectral distortion caused by factors such as sensors, atmospheric conditions, and sun altitude. Specific geometric correction and radiation correction technologies are existing technologies and will not be described in detail in this application.
[0048] Step 1.2: Use ArcMap 10.6 software to perform preprocessing operations such as simplification, merging, and geometric correction on the road network data; and establish buffer zones based on different levels of road networks, dividing the study area into a series of independent blocks, namely road network units, to provide a data basis for segmenting the image.
[0049] Among them, the setting of the cell network is conducive to the extraction and statistics of features. The present invention divides the data based on the road network unit and uses it as the representation and organization form of a certain type of functional area data.
[0050] Step 1.3: Filter the POI data to exclude data without clear category description and location information; and reclassify the POI data according to the "Urban Construction Land Classification and Planning Standards" issued by the Ministry of Housing and Urban-Rural Development of the People's Republic of China, and then convert the coordinate system of the POI data.
[0051] Specifically, based on Python programming, the coordinate conversion function is realized, and the above POI data is corrected from the Mars coordinate system referenced by Amap to the WGS84 coordinate system.
[0052] Step 1.4: Compare the building footprint data with the remote sensing image, and delete or add the building footprint data based on the comparison results.
[0053] Building footprint data can also be called building outline data. Specifically, the downloaded building outline data is compared with the corresponding remote sensing image in ArcMap10.6 software. If there is no building on the remote sensing image at the same location, but there is building outline data, you need to edit the feature to delete it. If there is a building on the remote sensing image, but there is no building outline data, you need to add the corresponding building outline surface feature.
[0054] Step 2: Segment the preprocessed remote sensing image to determine the unit for image feature extraction; and determine the type of socioeconomic data features.
[0055] Specifically, based on the road network processed in step 1.2, in ArcMap10.6 software, use the clipping tool to segment the image, and use the spatial analysis tool to divide the socioeconomic data.
[0056] Among them, determining the type of social and economic data characteristics includes: calculating the POI frequency characteristics within the road network unit; calculating the area, perimeter and number of floors of the buildings within the road network unit.
[0057] The formula for calculating POI frequency characteristics is:
[0058]
[0059]
[0060] Where i represents the POI type; n i N represents the number of the i-th POI type in the block; i represents the total number of POIs of the i-th type; F i represents the frequency density of the POI of the i-th type in the total number of POIs of this type; C i Represents the ratio of the frequency density of the ith type of POI to the frequency density of all types of POIs in the block.
[0061] Step 3: For multi-source data representation, feature extraction and feature construction are performed based on the divided blocks.
[0062] In step 3, a convolutional network is used to extract the spectral and texture features of the remote sensing image. The calculation formula for convolutional network feature extraction is:
[0063]
[0064] Among them, (K, L) represent the width and height of the convolution kernel respectively; w k,lis the weight parameter of k rows and l columns; b is the bias term; f represents nonlinear calculation; O n,m is the value at the feature map (n, m); (i, j) is the number of rows and columns that the convolution kernel slides on the image.
[0065] In addition, the statistical analysis tools in ArcMap10.6 software were used to construct the characteristics of socioeconomic data.
[0066] Step 4: Concatenate the features extracted by convolution with the features of socioeconomic data as the basic feature data for graph calculation, and perform graph convolution calculation.
[0067] Since the extracted features come from different data, the organization of the data makes it difficult to associate the extracted features with each other, such as the spectrum and texture features from the image, and the frequency features generated by the socioeconomic data. Due to the modal gap and heterogeneity, although the existing deep learning-based methods can establish feature associations by virtue of the feature mapping space they construct, this association is still weak and does not fully utilize the complementarity between data. Therefore, the present invention selects graph convolution for feature association and application, and establishes connections between multi-source data as the basis for functional area recognition and classification.
[0068] In step 4, the formula for graph convolution calculation is:
[0069]
[0070] in, I is the identity matrix of the same order as the adjacency matrix A, yes The degree matrix, H i is the feature matrix of layer i, W i is the weight matrix of the i-th layer, and σ is the nonlinear activation function.
[0071] The purpose of this step is to determine the association form of characteristic variables in urban functional area identification by considering the inherent characteristics of multi-source data. Due to the large heterogeneity and complex structure of multi-source data, it is difficult to establish effective associations between features. In this step, the characteristics of multi-source data are analyzed to determine the characteristic variables used to construct the classifier input in step 5.
[0072] Step 5: Based on the results of graph calculation, identify the division form of urban functional areas.
[0073] In step 5, the softmax function and the cross entropy loss function are used to establish the optimal graph structure composed of multi-source data features, realize the complementarity of multi-source data features, and complete the identification of urban functional areas.
[0074] The purpose of this step is to use graph convolution to strengthen feature association based on the feature variables determined in step 4, so that the feature variables can be used as much as possible to reflect the characteristics of the multi-semantic attributes of urban functional areas. Finally, based on the analysis and calculation of the constructed graph structure, accurate identification of urban functional areas is achieved.
[0075] Specifically, step 5 establishes an urban functional area identification classification model based on the data feature construction in step 3 and the graph calculated in step 4. In the urban functional area identification model, the socioeconomic characteristics come from the calculated frequency of points of interest and the frequency of building outlines. The frequency of building outlines in a unit uses a statistical method to construct features, and the maximum, minimum, average and mode of the building area, perimeter and number of floors in the area are used as features to represent the distribution of building features in the area.
[0076] Embodiment 2:
[0077] Based on Example 1, Example 2 of the present application provides a more specific method for identifying urban functional areas based on multi-source data collaboration of a graph convolutional network, including:
[0078] Step 1: Obtain remote sensing images and socioeconomic data, and perform data preprocessing;
[0079] Step 2: Segment the preprocessed remote sensing image to determine the unit for image feature extraction; and determine the type of socio-economic data features;
[0080] Step 3: Extract and construct features based on the multi-source data representation;
[0081] Step 4: Concatenate the features extracted by convolution with the features of the socioeconomic data as the basic feature data for graph computing, and perform graph convolution calculations;
[0082] Step 5: Based on the results of the graph calculation, identify the division form of urban functional areas;
[0083] Step 6: Figure 2 As shown, the accuracy of the recognition results is evaluated.
[0084] It can be seen that the accurate identification process of urban functional areas is achieved through feature extraction based on block units segmented by road network and feature calculation based on graph convolution, and multi-source complementary data is collaboratively applied in the identification framework. Specifically, the entire urban functional area identification framework is divided into three steps. The first step is to use pre-processed road network data to divide the functional unit area for feature calculation; the second step is to extract image features and construct socio-economic data features based on the divided unit areas; the third step is to use graph convolutional networks to establish connections between multi-source data features, realize data complementarity, and thus improve the recognition accuracy of urban functional areas.
[0085] In addition, Table 1 shows the algorithm implementation process, and Table 2 shows the accuracy evaluation of the recognition results in step 6.
[0086] Table 1 Algorithm implementation process
[0087]
[0088] Table 2 Experimental results accuracy evaluation
[0089]
[0090] Results analysis: From Table 1, Figure 2 It can be seen from the quantitative and qualitative evaluation indicators that the method proposed in this study has a better mapping effect than the traditional method multi-layer perceptron, with an overall mapping accuracy of 84%. The main reason is that the graph convolutional network considers the topological structure between nodes by performing convolution operations in graph calculations, and can effectively model graph data. In addition, when performing convolution operations, the graph convolutional network will consider the neighboring nodes of each node and fuse the information of the neighboring nodes into the representation of the current node. Such operations can effectively capture the local structural information between nodes. Areas with complex landforms such as industrial areas have a higher recognition accuracy of 87%, and the recognition accuracy of roads is 67%.
[0091] It should be noted that the parts in this embodiment that are the same or similar to those in Embodiment 1 can be referenced to each other and will not be described in detail in this application.
[0092] Embodiment 3:
[0093] Based on Example 1, Example 3 of the present invention provides an urban functional area recognition system based on multi-source data collaboration of a graph convolutional network, including:
[0094] The preprocessing module is used to obtain remote sensing images and socio-economic data and perform data preprocessing;
[0095] A determination module is used to segment the pre-processed remote sensing image, determine the unit for image feature extraction, and determine the type of socio-economic data features;
[0096] The extraction module is used to extract and construct features based on the multi-source data representation;
[0097] The splicing module is used to splice the features extracted by convolution with the features of socioeconomic data as the basic feature data for graph computing and perform graph convolution computing;
[0098] The recognition module is used to identify the urban functional area division form based on the results of graph calculation.
[0099] Specifically, the method provided in this embodiment is a system corresponding to the method provided in Embodiment 1. Therefore, the parts in this embodiment that are the same or similar to those in Embodiment 1 can be referenced to each other and will not be repeated in this application.
Claims
1. A method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks, characterized in that: include: Step 1: Obtain remote sensing images and socioeconomic data, and perform data preprocessing; In step 1, the socio-economic data includes road network data, POI data and building footprint data; step 1 includes: Step 1.1, perform geometric correction and radiation correction on remote sensing images; Step 1.2: Simplify, merge and geometrically correct the road network data; establish buffer zones according to different levels of road networks, divide the study area into a series of road network units, and provide a data basis for image segmentation; Step 1.3, filter the POI data to exclude data without clear category description and location information; and reclassify the POI data according to the preset planning standards, and then transform the coordinate system of the POI data; Step 1.4, compare the building footprint data with the remote sensing image, and delete or add the building footprint data according to the comparison results; Step 2: Segment the preprocessed remote sensing image to determine the unit for image feature extraction; and determine the type of socio-economic data features; Step 3: Extract and construct features based on the multi-source data representation; Step 4: Concatenate the features extracted by convolution with the features of the socioeconomic data as the basic feature data for graph computing, and perform graph convolution calculations; Step 5: Based on the results of graph calculation, identify the division form of urban functional areas.
2. The method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks according to claim 1 is characterized in that: In step 2, the type of social and economic data characteristics is determined, including: calculating the POI frequency characteristics within the road network unit; calculating the area, perimeter and number of floors characteristics of the buildings within the road network unit.
3. The method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks according to claim 2 is characterized in that: In step 2, the formula for calculating the POI frequency feature is: Where i represents the POI type; n i N represents the number of the i-th POI type in the block; i represents the total number of POIs of the i-th type; F i represents the frequency density of the POI of the i-th type in the total number of POIs of this type; C i Represents the ratio of the frequency density of the ith type of POI to the frequency density of all types of POIs in the block.
4. The method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks according to claim 3 is characterized in that: In step 3, a convolutional network is used to extract the spectral and texture features of the remote sensing image. The calculation formula for convolutional network feature extraction is: Among them, x is the convolution kernel, (K, L) represents the width and height of the convolution kernel respectively; w k,l is the weight parameter of k rows and l columns; b is the bias term; f represents nonlinear calculation; O n,m is the value at the feature map (n, m); (i, j) is the number of rows and columns that the convolution kernel slides on the image.
5. The method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks according to claim 4 is characterized in that: In step 4, the formula for graph convolution calculation is: in, I is the identity matrix of the same order as the adjacency matrix A, yes The degree matrix, H i is the feature matrix of layer i, W i is the weight matrix of the i-th layer, and σ is the nonlinear activation function.
6. The method for identifying urban functional areas based on multi-source data collaboration based on graph convolutional networks according to claim 5 is characterized in that: In step 5, the softmax function and the cross entropy loss function are used to establish the optimal graph structure composed of multi-source data features, realize the complementarity of multi-source data features, and complete the identification of urban functional areas.
7. A multi-source data collaborative urban functional area recognition system based on graph convolutional network, characterized in that: The method for identifying urban functional areas based on multi-source data collaboration of a graph convolutional network according to any one of claims 1 to 6 comprises: The preprocessing module is used to obtain remote sensing images and socio-economic data and perform data preprocessing; A determination module is used to segment the pre-processed remote sensing image, determine the unit for image feature extraction, and determine the type of socio-economic data features; The extraction module is used to extract and construct features based on the multi-source data representation; The splicing module is used to splice the features extracted by convolution with the features of socioeconomic data as the basic feature data for graph computing and perform graph convolution computing; The recognition module is used to identify the urban functional area division form based on the results of graph calculation.
8. A computer storage medium, characterized in that: The computer storage medium stores a computer program; when the computer program runs on a computer, the computer executes the urban functional area identification method based on multi-source data collaboration of a graph convolutional network as described in any one of claims 1 to 6.
