A method, device and medium for identifying typical information of a city-scale building

Abstract

The present application relates to a kind of typical information identification method, device and medium of urban scale building, its method includes: obtaining the building contour of target research city, POI data and AOI data, and import into the QGIS of pre-set to carry out spatial connection analysis and buffer analysis, obtain the number table of the POI type contained in each building contour;According to number table, each building contour is carried out main attribute identification and sub-type clustering analysis, and the identification success information or identification failure information of building contour belonging to building type is obtained;When obtaining identification failure information, based on random forest model and building geometry information, the building contour of identification failure is analyzed to obtain the building type belonging to it.The building volume of urban scale is often huge, the scheme proposed in the present application improves processing speed and recognition rate by combining unsupervised and supervised learning in identifying building type, not only saves time and effort in identification process, but also can cover as many building types as possible.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building information recognition, and in particular to a typical information recognition method and device for urban scale buildings and a medium. BACKGROUND

[0002] Urban scale building group energy consumption simulation is a bottom-up physical-based modeling method, which can be used to optimize new district energy planning schemes and evaluate old district energy saving retrofit schemes. In urban building group energy consumption simulation, the geometric information of buildings is obtained through geographic information system (GIS) data, while the types and performance parameters of various systems of the buildings need to be assumed according to typical buildings or relevant standards. The typical building database mainly classifies building energy consumption models according to different building types, construction years and climate regions, and building types and construction years are the main basis for reference of typical buildings.

[0003] At present, the most direct way to obtain these two types of information is to access the public data opening platform of government agencies. Some large cities in Europe and the United States, such as New York, Boston and Berlin, have publicly released data such as building contours, building heights, building types, construction years, etc. When there is no relevant information on the platform of the city, scholars at home and abroad use multi-source geographic big data to indirectly infer the typical information of buildings including building types and construction years.

[0004] For example, for building classification, Chinese patent "CN 110008542A A city mass building classification method" uses a logistic regression model to construct a training set based on building base plan and city electronic map points of interest (POI), determines the parameter weight, and then predicts the building type. It uses a supervised learning-based algorithm and needs a large amount of manual labeling and classification, which is more suitable for street scale in cities and does not have universal applicability. Chinese patent "CN107247938B A method for classifying functions of urban buildings in high-resolution remote sensing images" discloses using a convolutional neural network (CNN) to extract buildings in remote sensing images, and then performing kernel density estimation on classified POI data, and according to the average value of the kernel density of the buildings, classifying them into one of the three types of commercial service facility land, public management and public service land, and residential land. "CN113672788A A city building function classification method based on multi-source data and weight coefficient method" classifies building types into residential, school, office, commercial service and public service based on building vector data, POI and electronic map area of interest (AOI) data using a weight coefficient method. The classification schemes disclosed in CN107247938B and CN113672788A are not detailed enough in terms of building type classification, such as lacking hospitals, shopping malls and mixed buildings, because the input parameters in building energy consumption simulation need to be set based on more detailed building types.

[0005] For example, for the construction age, the commonly used CityGML model is mostly obtained from the public platform of European and American cities, and there is a lack of such data in China. The method based on street view images is greatly affected by surrounding obstructions such as trees or cars, thereby causing the recognition process to be easily disturbed and the recognition accuracy to be low. SUMMARY

[0006] (I) Technical problems to be solved

[0007] In view of the above-mentioned defects and deficiencies of the prior art, the present application provides a typical information recognition method and device for urban scale buildings and a medium, which solves the technical problems that the existing recognition scheme for the typical information of buildings does not have universality, the classification of buildings is not fine enough, the recognition process is easily disturbed, and the recognition accuracy is not high.

[0008] (II) Technical solutions

[0009] In order to achieve the above-mentioned purpose, the main technical solutions adopted by the present application include:

[0010] In a first aspect, the present application provides a typical information recognition method for urban scale buildings based on GIS data and high-resolution remote sensing images, which comprises:

[0011] Obtaining the building contour, POI data and AOI data of the target research city;

[0012] Importing the building contour, POI data and AOI data into a pre-set QGIS to perform spatial connection analysis and buffer analysis, and obtaining a quantity table of the POI types contained in each building contour;

[0013] According to the quantity table, performing main attribute recognition and sub-type clustering analysis on each building contour to obtain recognition success information or recognition failure information of the building type to which the building contour belongs;

[0014] When the recognition failure information is obtained, performing analysis on the building contour with recognition failure based on a pre-constructed random forest model and the obtained building geometric information to obtain the building type to which the building contour belongs.

[0015] Optionally, after obtaining the building contour, POI data and AOI data of the target city, the method further comprises:

[0016] The POI re-classified data is obtained by screening and re-classifying the POI data, and the POI re-classified data includes business office building data, government office data, residential area data, shopping mall data, hotel data, school data, hospital data, cultural art gallery data, company data, catering service site data, retail shopping site data, and entertainment and leisure site data.

[0017] Optionally, the building contour, the POI data, and the AOI data are imported into the pre-set QGIS to perform spatial connection analysis and buffer analysis, and a quantity table of POI types contained in each building contour is obtained, including:

[0018] After the building contour, the POI data, and the AOI data are imported into the QGIS, the following spatial connection analysis and buffer analysis are performed:

[0019] For the POI in the building contour, the type of the POI is directly assigned to the building contour;

[0020] For the POI not in the building contour, when the buffer distance of the POI and the adjacent building contour boundary meets the first requirement, the type of the POI is assigned to the building contour;

[0021] For the building contour in the AOI, the type of the AOI is directly assigned to the building contour;

[0022] For the building contour not completely in the AOI, when the buffer percentage of the AOI and the adjacent building contour meets the second requirement, the type of the AOI is assigned to the building contour;

[0023] After the spatial analysis and the buffer analysis, the quantity table of POI types contained in each building contour is obtained;

[0024] wherein,

[0025] The buffer distance is the shortest distance from the POI to the adjacent building contour boundary;

[0026] The buffer percentage is the percentage of the part of the building contour area in the AOI boundary to the total area of the building contour;

[0027] The first requirement is that the buffer distance is not more than 4 m;

[0028] The second requirement is that the buffer percentage is not less than 70%.

[0029] Optionally, according to the quantity table, main attribute recognition and sub-type clustering analysis are performed on each building contour to obtain recognition success information or recognition failure information of the building type of the building contour, including:

[0030] In the quantity table, when the building contour contains any one POI main attribute, the main category of the building is directly determined;

[0031] In the quantity table, when the building contour contains any one POI main attribute, the main category of the building is directly determined;

[0032] In the quantity table, when the building contour contains any one POI main attribute, the main category of the building is directly determined;

[0033] In the quantity table, when the building contour contains any one POI main attribute, the main category of the building is directly determined;

[0034] Based on the main attribute recognition and sub-type clustering analysis results, the recognition success information or recognition failure information of the building type to which each building contour belongs is output;

[0035] The POI main attribute includes a commercial office building attribute, a government office attribute, a residential attribute, a shopping mall attribute, a hotel attribute, a school attribute, a hospital attribute, and a cultural and art gallery attribute.

[0036] Optionally, when the recognition failure information is obtained, the building contour that fails to be recognized is analyzed based on a pre-constructed random forest model and obtained building geometric information, and the building type to which the building contour belongs is obtained, including:

[0037] The building geometric information of the building contour that fails to be recognized is obtained by analyzing the building contour and floor data using QGIS;

[0038] The sample building data on the public database obtained is used as a training sample, and the constructed random forest model is trained according to the training sample;

[0039] The building contour that fails to be recognized is classified based on the trained random forest model and the building geometric information, and the building type to which the building contour belongs is obtained;

[0040] The building geometric information includes the number of floors, the contour area, the contour perimeter, the minimum circumscribed rectangle short side width, the minimum circumscribed rectangle aspect ratio, and the approximate rectangle coefficient.

[0041] Optionally, after the building type to which the building contour that fails to be recognized belongs is obtained based on the pre-constructed random forest model and the obtained building geometric information when the recognition failure information is obtained, the method further includes:

[0042] For buildings of which the building type is residential, AOI data, building contours, and residential information retrieved from a public database are imported into a pre-configured QGIS for spatial join analysis to obtain residential building age information;

[0043] For buildings of which the building type is non-residential, different-year vector building contours are generated from different-year remote sensing images obtained by a pre-configured convolutional neural network model, and intersection analysis is performed on the different-year vector building contours in QGIS to obtain the age information of the non-residential buildings.

[0044] Optionally, for buildings of which the building type is non-residential, different-year vector building contours are generated from different-year remote sensing images obtained by a pre-configured convolutional neural network model, and intersection analysis is performed on the different-year vector building contours in QGIS to obtain the age information of the non-residential buildings, including:

[0045] Obtaining different-year satellite images of the non-residential buildings from a public database;

[0046] Selecting a region of buildings in a satellite image of a certain year, and selecting no less than 1000 corresponding building contours of various shapes as label data, aligning the label data with the buildings in the selected region in QGIS, and correcting the satellite image by adding missing contours and / or deleting redundant contours;

[0047] Dividing the aligned and corrected building contours and the satellite image into a training set and a validation set according to a split image size of 256x256 and a ratio of 9:1;

[0048] After training a pre-configured Mask R-CNN model according to the training set, inputting different-year satellite images into the trained model in sequence to generate corresponding different-year building contours;

[0049] Importing original building contours obtained from a public database and generated different-year building contours into QGIS, and performing pairwise intersection analysis between the original building contours and each different-year building contour in sequence;

[0050] If the intersection overlap between the original building contours and the building contours of a certain year exceeds a pre-configured value, it is determined that the buildings are the same, and it is further determined whether the building contours of a certain year have changed compared to the original building contours, and if so, it is determined that the construction age is between the original building contours and the building contours of a certain year;

[0051] Performing intersection analysis between the original building contours and the building contours of the remaining years multiple times to ultimately obtain different construction age information of the buildings in the selected region.

[0052] In a second aspect, an embodiment of the present application provides a device for identifying typical information of buildings at a city scale based on GIS data and high-resolution remote sensing images, which comprises:

[0053] a data acquisition module configured to acquire vector building outlines, POI data and AOI data of a target city;

[0054] a quantity table output module configured to import the building outlines, POI data and AOI data into a pre-set QGIS to perform spatial connection analysis and obtain a quantity table of POI types contained in each building outline;

[0055] a first analysis module configured to perform main attribute identification and sub-type clustering analysis on each building outline according to the quantity table to obtain building type identification success information or identification failure information of the building outline;

[0056] a second analysis module configured to perform analysis on the building outline for which identification fails based on a pre-constructed random forest model and acquired building geometric information to obtain a building type to which the building outline belongs.

[0057] Optionally, the device further comprises:

[0058] a residential building age information identification module configured to, for a building whose building type is residential, import the AOI data, building outlines and residential information retrieved from a public database into the pre-set QGIS to perform spatial connection analysis and obtain residential building age information;

[0059] a non-residential building age information identification module configured to, for a building whose building type is non-residential, generate vector building outlines of different ages by a pre-set convolutional neural network model according to acquired remote sensing images of different ages, and perform intersection analysis on the vector building outlines of different ages in the QGIS to obtain age information of the non-residential building.

[0060] In a third aspect, an embodiment of the present application provides a computer readable medium having computer executable instructions stored thereon, wherein the executable instructions are executed by a processor to implement a method for identifying typical information of buildings at a city scale based on GIS data and high-resolution remote sensing images.

[0061] (Three) beneficial effects

[0062] The present application has the beneficial effects that: since the volume of buildings at a city scale is often huge, the scheme of identifying typical information of buildings based on limited and public data proposed by the present application combines unsupervised learning and supervised learning to improve processing speed and identification rate in identifying building types, so that the building classification efficiency is higher and more detailed, the identification process is time-saving and labor-saving, and various building types can be covered as much as possible. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 A flowchart of a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images according to an embodiment of the present application is shown in FIG. 1.

[0064] Figure 2 A spatial connection and buffer diagram of POI and building contour of a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images according to an embodiment of the present application is shown in FIG. 2.

[0065] Figure 3 A spatial connection and buffer diagram of AOI and building contour of a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images according to an embodiment of the present application is shown in FIG. 3.

[0066] Figure 4 A specific flowchart of step S3 of a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images according to an embodiment of the present application is shown in FIG. 4.

[0067] Figure 5 A specific flowchart of step S4 of a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images according to an embodiment of the present application is shown in FIG. 5.

[0068] Figure 6 A specific flowchart of step S6 of a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images according to an embodiment of the present application is shown in FIG. 6.

[0069] Figure 7 A building type identification flowchart of a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images according to an embodiment of the present application is shown in FIG. 7.

[0070] Figure 8 A building age identification flowchart of a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images according to an embodiment of the present application is shown in FIG. 8. DETAILED DESCRIPTION

[0071] In order to better explain the present application and facilitate understanding, the present application is described in detail below through specific embodiments with reference to the accompanying drawings.

[0072] As Figure 1As shown, the embodiment of the present application proposes a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images, which comprises: obtaining building contours, POI data and AOI data of a target research city; importing the building contours, POI data and AOI data into a pre-set QGIS to perform spatial connection analysis and buffer analysis, and obtaining a quantity table of POI types contained in each building contour; performing main attribute identification and sub-type clustering analysis on each building contour according to the quantity table to obtain identification success information or identification failure information of the building type to which the building contour belongs; when the identification failure information is obtained, performing analysis on the building contour with identification failure based on a pre-constructed random forest model and obtained building geometric information to obtain the building type.

[0073] Since the building volume of urban scale is often huge, the scheme of the present application for identifying typical information of buildings based on limited and public data improves the processing speed and identification rate by combining unsupervised and supervised learning in identifying the building type, so that the building classification efficiency is higher and more refined, not only saving time and labor in the identification process, but also covering as many building types as possible.

[0074] In order to better understand the above technical solutions, the exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer, more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0075] Specifically, the present application provides a typical information identification method of urban scale buildings based on GIS data and high-resolution remote sensing images, comprising:

[0076] S1, obtaining building contours, POI data and AOI data of a target research city.

[0077] The building contour is a vector building contour, which can be obtained through BIGEMAP map downloader or online public resources and public databases. The point of interest (POI) mainly refers to some points of geographic entities closely related to people's life, containing information such as entity name, category, address and latitude and longitude and spatial attributes; the area of interest (AOI) also contains the above four basic information (entity name, category, address and latitude and longitude), which refers to regional data representing geographic entities.

[0078] In a geographic information system, a POI can be a building, a shop or a company, etc. An AOI can be a residential area, a university campus or a hospital campus, etc. The present application obtains different types of POI data and AOI boundary points through the disclosed Gaode map Web service API, and then converts the AOI boundary points into a surface through QGIS software.

[0079] Further, after step S1, the method further comprises: obtaining POI reclassified data by screening and reclassifying the POI data; wherein the POI reclassified data comprises business office building data, government office data, residential area data, shopping mall data, hotel data, school data, hospital data, cultural art gallery data, company data, catering service site data, retail shopping site data, and entertainment and leisure site data.

[0080] Since the POI data is classified into many categories, each major category contains multiple subcategories. Some types of classification are not comprehensive enough, for example, insurance and securities companies are located in financial insurance services, and law firms are located in life services, but they should also belong to companies, so it is necessary to select and reclassify the original data. For example, select business buildings to obtain business office buildings and residential areas, select government agencies and social groups to obtain government agencies above the township level, industrial and commercial tax agencies, and public security agencies and law enforcement agencies to divide into government offices, and select shopping services to obtain department stores, shopping centers and large supermarkets to divide into shopping malls. Finally, recombine 12 types of POI reclassified data, including business office buildings, government offices, residential areas, shopping malls, hotels, schools, hospitals, cultural art galleries, companies, catering services, retail shopping, and entertainment and leisure sites.

[0081] S2, import the building contour, POI data, and AOI data into the pre-set QGIS to perform spatial connection analysis and buffer analysis, and obtain a quantity table of POI types contained in each building contour.

[0082] Further, step S2 comprises:

[0083] After importing the building contour, POI data, and AOI data into QGIS, the following spatial connection analysis and buffer analysis are performed to obtain a quantity table of POI types contained in each building contour:

[0084] Reference Figure 2 and Figure 3After importing the building contour, POI data and AOI data into QGIS, the following spatial join analysis and buffer analysis are performed: for the POI within the building contour, the type of the POI is directly assigned to the building contour; for the POI not within the building contour, the type of the POI is assigned to the building contour when the buffer distance of the POI and the adjacent building contour boundary meets the first requirement; for the building contour within the AOI, the type of the AOI is directly assigned to the building contour; for the building contour not completely within the AOI, the type of the AOI is assigned to the building contour when the buffer percentage of the AOI and the adjacent building contour meets the second requirement; after the spatial analysis and buffer analysis, the quantity table of the POI type contained in each building contour is obtained. The buffer distance is the shortest distance from the POI to the adjacent building contour boundary; the buffer percentage is the percentage of the building contour area within the AOI boundary to the total building contour area; the first requirement is that the buffer distance is not more than 4m; and the second requirement is that the buffer percentage is not less than 70%.

[0085] In a specific embodiment, the building contour, POI data and AOI data are imported into QGIS, and the POI and AOI in the Gaode Mars coordinate system are converted into data in the WGS84 coordinate system. Each POI is a point in the WGS84 coordinate system, so for the POI within the building contour, the POI attribute is directly assigned to the building contour after the point-face intersection through the "add attribute by location" function in QGIS. For the POI outside the building contour, the error caused by the spatial position offset needs to be considered. The shortest distance from the POI to the adjacent building contour boundary is taken as the buffer distance. In order to determine the appropriate buffer distance, a block is selected to check whether the external POI belongs to the nearby building contour, and finally the buffer distance is set to 4m. For the building contour within the AOI, the type of the AOI (such as residence, hospital, school, etc.) is directly assigned to the building contour after the face-face intersection operation of QGIS. Due to the offset, the buffer analysis also needs to be performed on the building contour located on the boundary. A block is selected and the percentage of the building contour area within the boundary to the total contour area is calculated to determine the percentage of the building contour actually belonging to the AOI, and finally the buffer percentage is set to 70%.

[0086] It is worth mentioning that QGIS is a user-friendly, cross-platform open source desktop geographic information system developed based on Qt using C++, which can run on Linux, Unix, Mac OSX and Windows platforms.

[0087] S3, according to the quantity table, performing main attribute recognition and sub-type clustering analysis on each building contour to obtain recognition success information or recognition failure information of the building type to which the building contour belongs.

[0088] Further, as shown in Figure 4 Step S3 includes:

[0089] S31, when the building contour contains any one POI main attribute in the quantity table, the main category of the building is directly determined; wherein the POI main attribute includes commercial office building attribute, government office attribute, residential attribute, shopping mall attribute, hotel attribute, school attribute, hospital attribute and cultural art gallery attribute.

[0090] S32, when the building contour contains any multiple POI main attributes in the quantity table, the main category of the building contour is determined as mixed type.

[0091] S33, when the building contour does not contain POI main attribute in the quantity table, the main category of the building contour is determined according to the corresponding AOI type.

[0092] S34, after determining the main attribute, the mixed type building and the building without POI main attribute and AOI are clustered, and the average number of POI in each cluster and the proportion of each category are analyzed to determine the subcategory of the mixed type building and the building without POI main attribute and AOI.

[0093] S35, based on the main attribute recognition and subcategory clustering analysis results, the recognition success information or recognition failure information of the building type to which each building contour belongs is output.

[0094] After spatial analysis by QGIS, the quantity table of POI types contained in each building contour is finally obtained, and Table 1 takes one contour as an example. When the building contour contains any one POI attribute of commercial office building, government office, residential area, shopping mall, hotel, school, hospital and cultural art gallery, the main purpose of the building can be directly determined, so the eight attributes are defined as main attributes. When there are multiple main attributes, the building is determined as mixed type. When there is only one AOI attribute and no POI main attribute, the building type is determined according to the corresponding AOI type.

[0095] Table 1: Example of POI type quantity table contained in building contour

[0096] Category POI Number Category POI Number Hotel 0 School 0 Shopping Mall 0 Hospital 0 Office Building 1 Restaurant 12 Cultural Art Gallery 0 Company 20 Government Agency 0 Retail Shopping 28 Residence 0 Entertainment 0

[0097] In addition to the main attribute, commercial buildings usually have many other commercial POIs. For example, the building shown in Table 1 can be an office building with retail stores and restaurants. Subtype clustering analysis is used for further building use classification. Building subtype refers to buildings with other mixed uses. Clustering is a process of grouping similar data in a data set into the same class or cluster without the need for prior given criteria, and automatically classifying according to sample data. Some buildings only contain one main attribute and one or two other POIs, such as cultural art galleries, schools, hospitals, and government office buildings, which do not need further classification. Subtype clustering is used for residential, commercial office, hotel, and buildings without main attribute and AOI information, as they also contain a large number of other four types of commercial POIs. After clustering, the average number of POIs in each cluster and the proportion of each category are analyzed to identify mixed buildings.

[0098] Preferably, K-means clustering is the most widely used clustering algorithm, which is simple and efficient. It is a partition-based clustering algorithm that uses distance as a measure of similarity between data objects. The smaller the distance, the more likely they are in the same cluster. The distance calculation formula is as follows:

[0099]

[0100] where x =( x 1, x 2,..., x n ), y =( y 1, y 2,..., y n )represent two columns n dimensional vectors; usually the distance is selected as the Euclidean distance, and p =2.

[0101] S4, based on the pre-constructed random forest model and the obtained building geometry information, analyzing the building outline that fails to be recognized to obtain the building type to which it belongs.

[0102] Further, as shown in Figure 5 , step S4 includes:

[0103] S41, obtaining the building geometry information of the building outline that fails to be recognized by analyzing the building outline and floor data using QGIS, wherein the building geometry information includes: the number of floors above ground, the outline area, the outline perimeter, the minimum circumscribed rectangle short side width, the minimum circumscribed rectangle aspect ratio, and the approximate rectangle coefficient.

[0104] S42, the sample building data on the public database obtained is taken as a training sample, and the random forest model constructed is trained according to the training sample.

[0105] S43, the building contour whose recognition fails is classified based on the trained random forest model and the building geometric information, to obtain the building type.

[0106] Through the POI main attribute analysis and the sub-type clustering analysis, most of the building types can be identified, and the remaining building contours without POI and AOI attributes are classified based on the building geometric information by using the random forest method.

[0107] The geometric information of each building is obtained by analyzing the building contour and the floor data, including the number of floors above ground, the contour area, the contour perimeter, the minimum circumscribed rectangle short side width, the minimum circumscribed rectangle aspect ratio, and the approximate rectangle coefficient.

[0108] The number of floors above ground is obtained from the previous building vector contour data, the area and the perimeter of each building contour are obtained by using QGIS calculation. The approximate rectangle is introduced to represent the shape characteristics of the building contour, for example, most of the residential buildings are long and narrow. After the contour is rotated, a minimum circumscribed rectangle is used to enclose the contour boundary points. The approximate rectangle coefficient is defined as the ratio of the contour area to the minimum rectangle area, and the closer the coefficient is to 1, the closer the rectangle is. The length, width and area of the minimum circumscribed rectangle are calculated by QGIS.

[0109] According to the actual type of the sample building in Baidu Street View, the types of the building contours are divided into low-rise residential buildings, apartment-style residential buildings and other types.

[0110] Then, the random forest algorithm is used to train the model and make prediction. The model input is six geometric information, and the model output is the building type. The random forest is a classifier containing multiple decision trees, the decision tree is based on if-then-else rule, according to the corresponding attribute value in the classification item, the branch node is determined, until the leaf node is reached, and the classification result is obtained. The random forest is generated in a random way, and multiple independent decision trees are generated, each of which learns and predicts independently, and finally the voting results of multiple decision trees are counted to determine the final result, which is better than the classification result of any single classifier. The number of decision trees is 100, and the maximum depth is 10. Then, the 10-fold cross-validation method is used to evaluate the performance of the trained model, so as to avoid overfitting of the model. Finally, the trained model is used for classification of other buildings.

[0111] In addition, the embodiment of the present application also includes:

[0112] S5, for buildings of the residential type, AOI data, building contours, and residential information retrieved from public databases are imported into a pre-set QGIS for spatial connection analysis to obtain residential building age information.

[0113] For residential buildings, public data is collected based on a real estate website (such as Anjuke), and Python language programming is used to crawl the names, addresses, and building ages of residential communities in the city, and then the address is converted to latitude and longitude using the Gaode map geocoding API. Furthermore, the csv file containing the residential community information is imported into QGIS, presented as point data, and the latitude and longitude in the Gaode Mars coordinate system are converted to the latitude and longitude in the WGS84 coordinate system. After importing the building contours and community AOI, through the "add attributes by location" function, the building age attribute is directly assigned to the community AOI after point-face intersection, and then the community AOI age attribute is assigned to the building contours within it after face-face intersection.

[0114] S6, for buildings of the non-residential type, different year vector building contours are generated by a pre-set convolutional neural network model based on different year remote sensing images, and intersection analysis of different year vector building contours is performed in QGIS to obtain the age information of non-residential buildings.

[0115] Further, as shown in Figure 6 , step S6 includes:

[0116] S61, different year satellite images of non-residential buildings are obtained from public databases.

[0117] S62, select a building in a region in a satellite image of a certain year, and select no less than 1000 corresponding building contours of various shapes as label data, align the label data with the buildings in the selected region in QGIS, and correct the satellite image by adding missing contours or deleting redundant contours.

[0118] S63, the aligned and corrected building contours and satellite images are randomly divided into training set and validation set according to the segmentation image size 256x256 and the ratio of 9:1.

[0119] S64, after training the pre-built Mask R-CNN model according to the training set, different year satellite images are sequentially input into the trained model to generate corresponding different year building contours.

[0120] S65, the original building contours obtained from public databases and the generated different year building contours are imported into QGIS, and the original building contours are sequentially and pairwise intersected with each different year building contour.

[0121] S66, if the intersection of the original building contour and the building contour of a certain year exceeds the pre-set value, it is determined to be the same building, and then it is determined whether the building contour of a certain year has changed compared to the original building contour, if it has changed, it is determined that the construction year is between the original building contour and the building contour of a certain year.

[0122] S67, the intersection analysis of the original building contour and the building contour of the remaining years is performed multiple times, and finally the different construction year information of the buildings in the selected area is obtained.

[0123] For non-residential buildings, the contour is extracted based on high-resolution remote sensing images and the age is inferred by intersection analysis, specifically: first, download historical satellite images of different years from Google Map sources, with a resolution of 0.53m and an image format of tiff.

[0124] Second, select a region in a certain year's image, use the existing building contour vector data as labeled data, no less than 1000, and contain a variety of different shapes. Because there is a certain tilt angle when the satellite takes the image, resulting in a deviation between the building contour and the image, move the vector contour in QGIS to align the building in the image, and add missing or delete redundant contours.

[0125] Thirdly, the aligned and corrected building contour and satellite image are imported into ArcGIS Pro software. Set the segmentation image size to 256x256, and randomly divide the training set and validation set in a ratio of 9:1. Convolutional neural network (CNN) is a kind of deep neural network with convolutional layer, pooling layer and full connection layer structure, which is widely used in image recognition field. Typical representatives are semantic segmentation U-Net algorithm and instance segmentation Mask R-CNN algorithm. In the image, different building contours are identified and distinguished, which belongs to the task of instance segmentation, therefore Mask R-CNN algorithm is selected. In the deep learning module of ArcGIS Pro, set the backbone network to ResNet50, select 20 for the number of training rounds, and automatically extract the best learning rate in the learning curve. Finally, based on the trained model, input different year images in turn to predict and generate corresponding vector building contours.

[0126] Finally, the original building contour (e.g. 2018) and the building contour identified in different years (e.g. 2014, 2012, etc.) are imported into QGIS in turn. First, the data of 2018 and 2014 are subjected to face-face intersection, and if the overlapping part of the two buildings is more than 50%, it is judged as the same building, and then the changed building is detected to determine its construction year as 2015-2018. Similarly, the unchanged building in the original contour is subjected to face-face intersection with the data of 2012, and the distribution of buildings of different construction years is obtained by comparison in turn.

[0127] In addition, the embodiment of the present application also provides a device for identifying typical information of urban scale buildings based on GIS data and high-resolution remote sensing images, comprising:

[0128] A data acquisition module is configured to acquire vector building contours, POI data and AOI data of a target city.

[0129] A quantity table output module is configured to import the building contour, POI and AOI data into a pre-set QGIS to perform spatial connection analysis, so as to obtain a quantity table of POI types contained in each building contour.

[0130] A first analysis module is configured to perform main attribute identification and sub-type clustering analysis on each building contour according to the quantity table, so as to obtain building type identification success information or identification failure information of the building contour.

[0131] A second analysis module is configured to, when the identification failure information is acquired, analyze the building contour with identification failure based on a pre-constructed random forest model and acquired building geometric information, so as to obtain the building type.

[0132] Further, the device further comprises:

[0133] A residential building year information identification module is configured to, for the building with the building type of residence, import the AOI data, the building contour and residential information retrieved from a public database into a pre-set QGIS to perform spatial connection analysis, so as to obtain residential building year information.

[0134] A non-residential building year information identification module is configured to, for the building with the building type of non-residence, generate vector building contours of different years by a pre-set convolutional neural network model according to acquired remote sensing images of different years, and perform intersection analysis on the vector building contours of different years in QGIS to obtain the year information of the non-residential building.

[0135] In addition, the embodiment of the present application also provides a computer readable medium, which stores computer executable instructions, and the executable instructions are executed by a processor to realize the method for identifying typical information of buildings at urban scale based on GIS data and high-resolution remote sensing images.

[0136] In a specific implementation, taking the urban area of Changsha as an example, the following building typical information identification including building type and construction year is performed.

[0137] Firstly, the building contour vector data in 2018 (68966) from public databases such as Gaode, POI data in 2019 (281767) and AOI data (3367) are obtained. The POI data type information is shown in Table 2, which is divided into 11 categories, 104 secondary categories and 450 tertiary categories according to the Gaode reference document. The AOI data includes residential areas, schools, hospitals, scenic spots, industrial parks and the like.

[0138] Table 2 POI original data type table

[0139]

[0140] Secondly, the POI original data is recombined to obtain 12 types of POI reclassification data including commercial office buildings, government offices, residential areas, shopping malls, hotels, schools, hospitals, cultural art galleries, companies, catering services, retail shopping and entertainment leisure places.

[0141] Then, the building contour, POI and AOI data are imported into QGIS, and after point-face and face-face intersection and considering the buffer distance and buffer percentage, finally 200054 of the 281767 POIs are assigned to the building contour, and there are a total of 43831 building contours in the AOI.

[0142] Further, a table of the number of 12 types of POIs and AOI labels contained in each building contour is obtained by statistics, and the contour containing the main attribute is screened out by using Python language programming to determine the main purpose of the building, and then the sub-type clustering is performed. Finally, commercial office buildings, government office buildings, residential buildings, shopping malls, hotels, schools, hospitals, cultural art galleries, commercial and residential buildings, commercial mixed buildings (hotel-office, hotel-shopping mall, office-shopping mall and other mixed buildings) are identified, and the identification rate is 69%. Selecting 7895 building contours in the urban area, the actual building purpose is determined by using the Baidu map street view image to verify the accuracy of the algorithm, and the result shows that the total accuracy is 86%.

[0143] Then, for the remaining 31% unidentified building contours (21538), 6 geometric information of each contour was obtained by QGIS. According to the actual building type of 3036 building contours marked by Baidu Street View, it was divided into low-rise residential, apartment-style residential and other types. The random forest algorithm was used to train the sample, and then predict the type of 18502 building contours. The results showed that the overall accuracy of the model was 81.7%, and 10428 low-rise residential buildings, 5686 apartment-style residential buildings and 2388 other types (regarded as unidentified) were predicted. Through the combination of the two methods, 95.6% of the building types were identified in the 68966 building contours.

[0144] Then, the Python language programming was used to crawl the names, addresses and building age information of residential communities in Changsha urban area from Anjuke website. In QGIS, the community age attribute was assigned to the community AOI by point-face intersection, and the community AOI age attribute was assigned to the building contours in it by face-face intersection.

[0145] Finally, the satellite images of Changsha urban area in 2005, 2008, 2012 and 2014 from Google Maps were downloaded, with a resolution of 0.53m. The central area of the city in 2014 image was selected, containing 8000 building contours as labeled samples. After training the CNN model, the results showed that the average accuracy was 81%, which could better identify the buildings with spacing. According to the existing building vector data in 2018, the intersection calculation was carried out with the building contours extracted in each year, and the distribution of different construction years was obtained.

[0146] In summary, the present application provides a typical information identification method and system of urban scale buildings based on GIS data and high-resolution remote sensing images.

[0147] Reference Figure 7 It can be seen that the building type identification step in the typical information includes:

[0148] (1) Determine the research city, and obtain the vector building contours, POI and AOI data from Gaode Map.

[0149] (2) Reclassify the POI data.

[0150] (3) Perform spatial connection analysis and buffer analysis on GIS data. After spatial analysis and buffer analysis by QGIS, the number table of POI types contained in each building contour is finally obtained.

[0151] (4) Identify the main purpose of the building according to the main attribute of the POI.

[0152] (5) Perform subtype clustering analysis again.

[0153] (6) The majority of building types can be identified through POI main attribute and sub-type clustering analysis, and the remaining building outlines without POI and AOI attributes are classified based on building geometric information using the random forest method.

[0154] (7) The geometric information of each building, including the number of floors, outline area, outline perimeter, minimum circumscribed rectangle short width, minimum circumscribed rectangle aspect ratio, and approximate rectangle coefficient, is obtained by analyzing building outlines and floor data.

[0155] (8) Training samples are selected.

[0156] (9) The random forest algorithm is used to train the model and make predictions.

[0157] Referring to Figure 8 , it can be seen that the building age identification step in the typical information includes:

[0158] (10) For residential buildings, public data collected from real estate websites such as Anjuke is imported into QGIS along with building outlines and community AOIs for spatial join analysis.

[0159] (11) Obtain historical satellite images of different years.

[0160] (12) Label samples.

[0161] (13) Use the Mask R-CNN algorithm to train the model and make predictions.

[0162] (14) Perform intersection analysis on building outlines of different years.

[0163] Based on the above description, in the identification of building types, the unsupervised learning method can identify most of the buildings, which is time-saving and labor-saving and classifies the buildings more finely. Then for the remaining small part of buildings without POI and AOI information, the supervised learning method is used, which only needs to label a small amount of samples for training and prediction.

[0164] Generally, residential buildings in cities account for more than 50%, and in the identification of construction years, the invention first uses public community data from websites to determine the age of residential buildings, quickly and efficiently completing at least half of the workload. For non-residential buildings (commercial buildings), the historical satellite image year selection refers to the time node (2005 and 2015) issued by the Public Building Energy Saving Design Standard, which can be spaced 3 or 4 years, without the need to process year-by-year images, saving a part of the algorithm running time.

[0165] Compared with the prior art, the method based on public data can be popularized to different cities. In the identification of building types, the combination of unsupervised and supervised learning algorithms can improve the processing speed and identification rate, and cover various building types as much as possible. In the identification of construction years, satellite images have more comprehensive coverage of the entire city than street view images. Through automatic comparison of historical satellite images, the changed buildings can be quickly detected and their years can be identified.

[0166] Since the system / device for implementing the method of the above-mentioned embodiments of the present application is described in the above-mentioned embodiments of the present application, the specific structure and modification of the system / device can be understood by those skilled in the art based on the method described in the above-mentioned embodiments of the present application, and thus will not be described here. Any system / device used in the method of the above-mentioned embodiments of the present application belongs to the scope of the present application.

[0167] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0168] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions.

[0169] It should be noted that the word "a" or "an" located before a component does not exclude the presence of multiple such components. The present application can be implemented by means of hardware including several different components and by means of a suitably programmed computer. The use of the words first, second, third, etc. is only for the purpose of description and does not indicate any order or sequence. These words can be understood as part of the component name.

[0170] Moreover, it is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. It is also to be understood that the terminology could be interchangeable under the pertinent statutory provisions, that the mere use of certain terms does not exclude mecha nism from other terms and / or combinations of the terms, and / or that features described as meth ods and / or units could be implemented using other features as methods and / or units.

[0171] Although the preferred embodiments of the application have been disclosed, a worker of ordinary skill in this art would realize that other modifications to and variations of the preferred em bodiments could be made and would fall within the spirit and scope of the application.

[0172] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the ap plication, the application can be practiced otherwise than as specifically described.

Claims

1. A method for identifying typical information of urban scale buildings based on GIS data and high-resolution remote sensing images, characterized in that, The method comprises the following steps: obtaining building outlines, POI data and AOI data of a target city; importing the building outlines, the POI data and the AOI data into a pre-set QGIS to perform spatial connection analysis and buffer analysis, and obtaining a quantity table of POI types contained in each building outline, including: after importing the building outlines, the POI data and the AOI data into the QGIS, performing the following spatial connection analysis and buffer analysis: for a POI located in a building outline, directly assigning a type of the POI to the building outline; for a POI not located in a building outline, when a buffer distance of the POI to a boundary of an adjacent building outline meets a first requirement, assigning a type of the POI to the building outline; for a building outline located in an AOI, directly assigning a type of the AOI to the building outline; for a building outline not completely located in an AOI, when a buffer percentage of the AOI to an adjacent building outline meets a second requirement, assigning a type of the AOI to the building outline; after the spatial analysis and the buffer analysis, obtaining the quantity table of the POI types contained in each building outline; wherein the buffer distance is a shortest distance from the POI to the boundary of the adjacent building outline; the buffer percentage is a percentage of a partial building outline area in the AOI boundary to a total area of the building outline; the first requirement is that the buffer distance is not more than 4 m; and the second requirement is that the buffer percentage is not less than 70%; performing main attribute recognition and sub-type clustering analysis on each building outline according to the quantity table, and obtaining recognition success information or recognition failure information of a building type of each building outline, including: when the quantity table contains any one POI main attribute of a building outline, directly determining a main category of the building; when the quantity table contains any multiple POI main attributes of a building outline, determining that a main category of the building outline is a mixed type; when the quantity table does not contain a POI main attribute of a building outline, determining a main category of the building outline according to a corresponding AOI type; after determining the main attribute, clustering the mixed type building and the building without the POI main attribute and the AOI, and determining sub-categories of the mixed type building and the building without the POI main attribute and the AOI by analyzing an average number of POIs in each cluster and a proportion of each category after clustering; based on the main attribute recognition and the sub-type clustering analysis result, outputting the recognition success information or the recognition failure information of the building type of each building outline; wherein the POI main attribute includes a commercial office building attribute, a government office attribute, a residential attribute, a shopping mall attribute, a hotel attribute, a school attribute, a hospital attribute and a cultural and artistic museum attribute; when the recognition failure information is obtained, analyzing the building outline with recognition failure based on a pre-constructed random forest model and obtained building geometric information, and obtaining a building type of the building outline. 2.The method of claim 1, wherein the method is characterized by, After obtaining the building outlines, the POI data and the AOI data of the target city, the method further comprises the following steps: The POI re-classification data is obtained by screening and re-classifying the POI data; wherein, the POI re-classification data comprises business office building data, government office data, residential area data, shopping mall data, hotel data, school data, hospital data, cultural art gallery data, company data, catering service site data, retail shopping site data, and entertainment and leisure site data. 3.The method of claim 1, wherein the method further comprises: determining a building type of each of the buildings based on the building information and the building type information; and determining a building type of each of the buildings based on the building information and the building type information. When the identification failure information is obtained, the identification failure building contour is analyzed based on the pre-constructed random forest model and the obtained building geometry information, and the building type to which the building contour belongs is obtained, including: The building geometry information of the identification failure building contour is obtained by analyzing the building contour and the floor data by using QGIS; The sample building data on the public database is obtained as a training sample, and the pre-constructed random forest model is trained according to the training sample; The identification failure building contour is classified based on the trained random forest model and the building geometry information, and the building type to which the building contour belongs is obtained. The building geometry information includes the number of floors, the contour area, the contour perimeter, the minimum circumscribed rectangle short side width, the minimum circumscribed rectangle aspect ratio, and the approximate rectangle coefficient.

4. The method according to any one of claims 1-3, wherein the method is characterized by, After obtaining the building type to which the building contour belongs based on the pre-constructed random forest model and the obtained building geometry information when the identification failure information is obtained, the method further comprises: For the building whose building type is residence, the AOI data, the building contour, and the residence information retrieved from the public database are imported into the pre-set QGIS for spatial connection analysis to obtain the residence building age information; For the building whose building type is non-residence, different year vector building contours are generated from different year remote sensing images by using the pre-set convolutional neural network model, and the intersection analysis of the different year vector building contours is performed in QGIS to obtain the age information of the non-residence building. 5.The method of claim 4, wherein the method further comprises: determining a building type of the building based on the building information and the building type information; and determining a building type of the building based on the building information and the building type information. For the building whose building type is non-residence, different year satellite images of the non-residence building are obtained from the public database; A region of the building in the satellite image of a certain year is selected, and a corresponding building contour of no less than 1000 and containing multiple different shapes is selected as the label data, the label data is aligned with the selected region in QGIS, and the satellite image is corrected by adding missing contours and / or deleting redundant contours; The aligned and corrected building contour and the satellite image are randomly divided into a training set and a validation set according to a segmentation image size of 256x256 and a ratio of 9:1; After the pre-constructed Mask R-CNN model is trained according to the training set, the different year satellite images are sequentially input into the trained model to generate the corresponding different year building contours. ​ The original building contour and the generated building contour of different years obtained from the public database are imported into QGIS, and the original building contour is intersected with each building contour of different years in turn; If the intersection overlap of the original building contour and the building contour of a certain year exceeds the pre-set value, it is judged as the same building, and then it is further judged whether the building contour of a certain year has changed compared with the original building contour, if it has changed, it is judged that the construction year is between the original building contour and the building contour of a certain year; The intersection analysis of the original building contour and the building contour of the remaining years is performed multiple times, and finally the different construction year information of the buildings in the selected area is obtained.

6. A device for identifying typical information of buildings at urban scale based on GIS data and high-resolution remote sensing images, applied to the method of any one of claims 1-5, characterized in that, It comprises: A data acquisition module for acquiring vector building contours, POI data and AOI data of a target city; A quantity table output module for importing the building contour, POI and AOI data into the pre-set QGIS to perform spatial connection analysis and obtain the quantity table of the POI types contained in each building contour; A first analysis module for performing main attribute recognition and sub-type clustering analysis on each building contour according to the quantity table to obtain building type discrimination success information or discrimination failure information of the building contour; A second analysis module for analyzing the building contour of which the recognition fails based on the pre-constructed random forest model and the obtained building geometric information to obtain the building type. 7.The device for identifying typical information of buildings at urban scale based on GIS data and high-resolution remote sensing images according to claim 6, wherein, It further comprises: A residential building year information recognition module for buildings of which the building type is residential, importing the AOI data, building contour and residential information retrieved from the public database into the pre-set QGIS to perform spatial connection analysis and obtain the residential building year information; A non-residential building year information recognition module for buildings of which the building type is non-residential, generating vector building contours of different years by the pre-set convolutional neural network model according to the obtained remote sensing images of different years, and performing intersection analysis on the vector building contours of different years in QGIS to obtain the year information of the non-residential buildings.

8. A computer readable medium having stored thereon computer- executable instructions, characterized in that, The executable instructions are executed by the processor to implement the method for identifying typical information of buildings at city scale based on GIS data and high-resolution remote sensing images according to any one of claims 1-5.

Images