A method for identifying urban functional areas based on multi-source feature fusion

Abstract

The present invention relates to the field of identification of urban functional areas, and specifically discloses an urban functional area identification method based on multi-source feature fusion. The method comprehensively utilizes multi-source and multi-modal data such as remote sensing images and POI data, and utilizes advanced algorithms such as deep learning to fully mine multi-source data information. Based on high-resolution stereo satellite stereo image pairs, high-resolution night light satellite images, initial POI data, initial OSM data, and initial building function classification data, multi-dimensional features such as ground feature distribution features representing urban functional areas, night light index of human activity features, social function features, and building distribution features are extracted. The accuracy of urban functional area identification is improved through multi-source feature fusion and integrated learning. The method can efficiently and automatically extract and identify urban functional areas, provide data support and technical guarantee for urban planning management, have the advantages of low production cost, high efficiency, and the ability to achieve periodic updates, and have broad application prospects.

Description

Technical Field

[0001] The present invention relates to the field of urban functional area recognition, and in particular to an urban functional area recognition method based on multi-source feature fusion. Background Art

[0002] With the acceleration of China's urbanization process, urban construction is changing with each passing day, more and more people are migrating to cities, and the urban structure has undergone major changes. At the same time, this has also caused problems such as traffic congestion, environmental pollution, land shortage and duplicate construction in some parts of the city. Different urban functional areas are formed according to the different social functions they undertake, such as commercial land, residential land, industrial land, public service land, etc. Accurately identifying urban functional areas can better understand the internal structure and dynamic characteristics of the city, so as to allocate and utilize resources in a targeted manner, provide a strong basis for the division of urban functional areas, and help promote the sustainable development of urbanization.

[0003] Traditional urban functional zone identification mostly uses methods such as statistical surveys and expert judgment. Through on-site visits and questionnaires, functional zones are identified using expert experience or simple statistical indicators. These methods are time-consuming and labor-intensive, inefficient, inaccurate, highly subjective, and difficult to dynamically update, and cannot adapt to the needs of monitoring dynamic changes in urban structure. In recent years, with the continuous development of remote sensing, sensor and other technologies, a large amount of valuable multi-source and multi-modal data has been provided for urban analysis. At the same time, the rapid development of computer, machine learning and other technologies has also provided technical means for data processing, information extraction and analysis of urban analysis. Therefore, automatic identification of urban functional zones based on rich multi-source data and computer technology has been widely studied. The existing methods mainly include the following three methods:

[0004] 1) Methods based on remote sensing images;

[0005] Remote sensing data has the characteristics of wide coverage, short imaging cycle, and rich data sources. It is one of the most important and efficient means to obtain land cover / land use. Since surface cover and land use can reflect the function of a region to a certain extent, it is feasible and effective to identify urban functional areas based on remote sensing images. The urban functional area identification method based on remote sensing images mainly distinguishes different categories of regions by analyzing simple features such as spectral characteristics and texture characteristics of each region or deep image features extracted by deep learning. However, the method based on remote sensing images only uses the information of the surface cover of the region and cannot explore the real functional attributes of the region.

[0006] 2) Methods based on point of interest data;

[0007] Points of Interest (POI) data are point entities abstracted from geographic elements in space. In addition to the spatial information such as latitude and longitude coordinates and addresses of geographic entities, they are also connected to various attribute information, such as POI name, category, administrative division, etc. POI data contains rich urban function information and is widely used in research on spatiotemporal behavior, urban planning, geographic information, etc. The urban functional area identification method based on POI data is to spatially associate and statistically analyze urban functional areas with POI data, and classify urban functional areas by counting the categories and numbers of POI points in different urban functional areas and using methods such as kernel density analysis. However, the POI-based method only considers the information of POI, and POI has various problems such as uneven distribution and data duplication. At the same time, different POI categories have different degrees of influence on urban functional areas. It is impossible to accurately identify urban functional areas relying solely on POI information.

[0008] 3) Methods based on spatiotemporal big data;

[0009] With the continuous accumulation of human activity data and social network data, the identification of urban functional areas has gradually shifted from using single data to using multi-source data. Human activity data includes mobile phone signaling data, vehicle floating data, etc. By analyzing human activity data, we can explore the movement patterns of humans between regions. Social network data reflects people's attitudes and tendencies towards urban areas through text and image information. This information greatly enriches the regional feature attributes available for functional area classification. Based on the method of spatiotemporal big data, we extract features of different dimensions by mining from spatiotemporal big data, and apply these features to machine learning classification to identify urban functional areas. However, spatiotemporal big data is difficult to obtain. At the same time, data such as social networks are affected by factors such as the age and gender of the user population, and their reliability is poor.

[0010] In summary, the existing urban functional area recognition methods do not fully exploit and utilize data information. Most methods only use one or two of the data such as remote sensing images, POIs or mobile phone signals, and do not fully integrate the features of multi-source data. At the same time, most of the existing urban functional area recognition methods still rely on single classifiers such as random forests and Softmax, and rarely use ensemble learners and deep learning represented by convolutional neural networks. The ability to extract and analyze multi-source features needs to be improved. How to fully mine the information of multi-source and multi-modal data, take advantage of the advantages of multi-source feature fusion, and improve the accuracy of urban functional areas is still a problem that needs to be solved in current research. Summary of the invention

[0011] In order to solve the above technical problems, the present invention provides a method for identifying urban functional areas based on multi-source feature fusion. It comprehensively utilizes multi-source and multi-modal data such as remote sensing images and POI data, and uses advanced algorithms such as deep learning to fully mine multi-source data information, extract multi-dimensional features such as the distribution characteristics of land objects representing urban functional areas, the night light index of human activity characteristics, social function characteristics and building distribution characteristics. Through multi-source feature fusion and integrated learning, the accuracy of urban functional area identification is improved, and urban functional areas can be extracted and identified efficiently and automatically, providing data support and technical guarantee for urban planning management. It has the advantages of low production cost, high efficiency and the ability to achieve periodic updates, and has broad application prospects.

[0012] The present invention provides a method for identifying urban functional areas based on multi-source feature fusion, which adopts the following technical solution:

[0013] A method for identifying urban functional areas based on multi-source feature fusion, comprising the following steps:

[0014] S1, obtain multi-source data and preprocess them;

[0015] S1-1, obtain high-resolution stereo satellite stereo image pairs and preprocess them to obtain high-resolution stereo image pairs and orthophoto multispectral images;

[0016] S1-2, obtaining high-resolution nighttime light satellite images and preprocessing them to obtain nighttime light images;

[0017] S1-3, obtaining initial POI data and preprocessing it to obtain preprocessed POI data;

[0018] S1-4, obtaining initial OSM data and preprocessing it to obtain preprocessed OSM data;

[0019] S1-5, obtaining initial building function classification data, preprocessing the attributes of the initial building function classification data, and obtaining preprocessed building function classification data, wherein the attributes of the initial building function classification data include building function category and building area;

[0020] S2, superimposing the line layer of the pre-processed OSM data obtained in step S1-4 with the surface layer of the area to be identified, performing line cutting surface operation on the area to be identified according to the pre-processed OSM data, and obtaining the urban functional area plot unit data of the area to be identified;

[0021] S3, based on the high-resolution stereo image pair obtained in S1-1, inverts the height of the surface objects.

[0022] Get the resampled nDSM data;

[0023] S4, superimposing the orthophoto multispectral image obtained in step S1-1 and the resampled nDSM data obtained in step S3 to obtain a multidimensional feature image including a multispectral channel and an altitude channel;

[0024] S5, performing image segmentation based on a deep learning semantic segmentation model on the multi-dimensional feature image obtained in step S4 to obtain a vectorized feature segmentation result;

[0025] S6, performing spatial association and statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the vectorized feature segmentation result obtained in step S5, and calculating the feature distribution characteristics F of the urban functional area cover , F cover Indicates the area proportion of each type of land feature in each urban functional area plot unit;

[0026] S7, performing spatial correlation and surface statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the preprocessed night light image obtained in step S1-2, and calculating the night light index F representing the characteristics of human activities according to the brightness values ​​indicated by the pixels located in the urban functional area unit in the night light image. light ;

[0027] S8, reclassifying the pre-processed POI data obtained in step S1-3 according to the mapping relationship between the categories of the POI data and the categories of the urban functional areas, reclassifying the categories of the pre-processed POI data into categories for marking urban functional areas, and setting weights for each type of POI point to obtain reclassified POI data;

[0028] S9, performing spatial association and statistical analysis on the urban functional area plot unit data of the to-be-identified area obtained in step S2 and the reclassified POI data obtained in step S8, and calculating the social function characteristics F of the urban functional area use , F use Indicates the proportion of the number and weight product of each type of POI points inside and around each urban functional area plot unit;

[0029] S10, performing spatial association and statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the pre-processed building function classification data obtained in steps S1-5, and calculating the building distribution characteristics F of the urban functional area building , F building Indicates the number, area and average building height of each building function category in each urban functional area unit;

[0030] S11, for the land feature distribution feature F of the urban functional area obtained in step S6 cover, the night light index F representing the characteristics of human activities in the urban functional area obtained in step S7 light , the social function characteristics F of the urban function area obtained in step S9 use and the building distribution characteristics F of the urban functional area obtained in step S10 building , identification of urban functional areas based on XGBoost ensemble classifier;

[0031] S11-1, the land feature distribution characteristics F of the urban functional area cover , Night Light Index F light 、Social Functional Characteristics F use and building distribution characteristics F building Connect them in series to obtain the comprehensive characteristics F of the urban functional areas;

[0032] S11-2, constructing a training sample set, taking the comprehensive feature F of the urban functional area obtained in step S11-1 as input and the real category of the urban functional area as output to train the XGBoost integrated classifier;

[0033] S11-3, using the XGBoost integrated classifier trained in step S11-2, taking the comprehensive features F of the urban functional areas to be identified as input, identify the urban functional areas and obtain the urban functional areas identification results.

[0034] Preferably, S3 comprises the following steps:

[0035] S3-1, based on the high-resolution stereo image pair obtained in S1-1, the digital surface model data is extracted through the binocular stereo matching algorithm, referred to as DSM data;

[0036] S3-2, filtering the DSM data obtained in step S3-1 to obtain digital elevation model data representing ground height, referred to as DEM data;

[0037] S3-3, performing a subtraction operation on the DSM data obtained in step S3-1 and the DEM data obtained in step S3-2 to obtain normalized digital surface model data representing the relative height of surface objects, referred to as nDSM data;

[0038] S3-4, resampling the nDSM data obtained in step S3-3 with reference to the orthophoto multispectral image obtained in step S1-1 to obtain resampled nDSM data.

[0039] Preferably, S5 comprises the following steps:

[0040] S5-1, slicing the multi-dimensional feature image according to the same preset size to obtain an image slice data set;

[0041] S5-2, inputting the image slices in the image slice data set into the trained semantic segmentation model in sequence to perform image segmentation, and obtaining a set object segmentation result corresponding to the image slice set;

[0042] S5-3, merging the collective feature segmentation results to obtain a regional feature segmentation result;

[0043] S5-4, vectorizing the regional feature segmentation result to obtain a vectorized feature segmentation result, wherein the attributes of the vectorized feature segmentation result include the type of feature and the area of ​​the feature.

[0044] Preferably, the land feature distribution feature F of the urban functional area of ​​S6 cover The calculation formula is:

[0045]

[0046] Among them, F cover is the distribution characteristics of the land features in the urban functional area, S i is the area of ​​the i-th type of land feature in each urban functional area plot unit; i∈1,2,3…m is the number of the land feature category; m is the total number of land feature categories.

[0047] Preferably, the night light index F in step S7 is light The calculation formula is:

[0048]

[0049] Among them, F light It is the night light index representing the human activity characteristics of each building, S is the area of ​​the building's outline, n is the number of pixels of the night light image located within the building's outline, and DNi is the pixel value of each pixel of the night light image located within the building's outline.

[0050] Preferably, the formula for the mapping relationship between the categories of POI data and the categories of urban functional areas in S8 is:

[0051] f:type poi →type region ;

[0052] Among them, type poi Is the category of POI data, type region It is the category of urban functional area. For any POI category, there is a unique urban functional area category corresponding to it.

[0053] Preferably, the S9 comprises the following steps:

[0054] S9-1, presetting a buffer distance for each urban function area plot unit in the urban function area plot unit data of the area to be identified obtained in step S2 and buffering the data to obtain buffered urban function area plot unit data;

[0055] S9-2, spatially connecting the buffered urban functional area plot unit data obtained in step S9-1 with the reclassified POI data obtained in step S8, calculating the number and type of reclassified POI data associated with each urban functional area plot unit data and presetting the weight;

[0056] S9-3, according to the number, type and weight of the reclassified POI data associated with the plot unit data of each urban functional area obtained in step S9-2, calculate the social function characteristics F of the urban functional area use .

[0057] Preferably, the social function characteristics F of the urban function area in S9 use The calculation formula is:

[0058]

[0059] Among them, F use It represents the social function characteristics of each urban functional area, which indicates the proportion of the product of the number of each POI category and its weight inside and around the plot unit of each urban functional area to the sum of the total number of associated POIs and the product of their weights. i is the number of POIs of the i-th category associated with the plot unit in the functional area of ​​the city, w i is the weight value of the i-th category POI, and m is the total number of POI categories.

[0060] Preferably, the building distribution characteristics F of the urban functional area in S10 building The calculation formula is:

[0061]

[0062] Among them, F building It represents the building distribution characteristics of each urban functional area; F num is the proportion of buildings of each building function type in each urban functional area plot unit; F area is the area ratio of each building function type in each urban functional area plot unit; F heig h t is the average height of all buildings in each urban functional area plot unit; n i is the number of buildings of each building function category in each urban functional area plot unit; S iis the sum of the areas of buildings of each building function category in each urban functional area plot unit; i∈1,2,3…m is the number of each building function category; m is the number of all building function categories; h j is the height of each building in each urban functional area plot unit; j∈1,2,3…k is the serial number of each building in each urban functional area plot unit; k is the number of buildings in each urban functional area plot unit.

[0063] Preferably, the calculation formula for the comprehensive characteristic F of the urban functional area in S11-1 is:

[0064] F=CONTACT(F cover , F light , F use , F building );

[0065] Among them, F represents the comprehensive characteristics of urban functional areas, F cover represents the distribution characteristics of land features in urban functional areas, F cover The night light index that represents the characteristics of human activities in urban functional areas, F use is the social function characteristic of the urban functional area, F building It is the architectural distribution characteristic of urban functional areas.

[0066] In summary, the present invention includes the following beneficial technical effects:

[0067] 1. The present invention makes comprehensive use of multi-source and multi-modal data such as remote sensing images and POI data, and uses advanced algorithms such as deep learning to fully mine multi-source data information, extract multi-dimensional features such as the distribution characteristics of land objects representing urban functional areas, the night light index of human activity characteristics, social function characteristics, and building distribution characteristics. Through multi-source feature fusion and integrated learning, the accuracy of urban functional area identification is improved, and urban functional areas can be extracted and identified efficiently and automatically, providing data support and technical guarantee for urban planning management. It has the advantages of low production cost, high efficiency, and the ability to achieve periodic updates, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flow chart of the method for identifying urban functional areas in an embodiment of the present invention.

[0069] Figure 2 It is a flow chart of step S3 in the embodiment of the present invention.

[0070] Figure 3 It is a schematic diagram of the flow of steps S4 and S5 in an embodiment of the present invention.

[0071] Figure 4Schematic diagram of the urban functional area recognition result obtained in the embodiment of the present invention. DETAILED DESCRIPTION

[0072] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0073] The embodiment of the invention discloses a method for identifying urban functional areas based on multi-source feature fusion.

[0074] This embodiment is described by taking the identification of urban functional areas within the second ring road of Shijiazhuang City as an example.

[0075] Reference Figure 1 , the urban functional area identification method based on multi-source feature fusion includes the following steps:

[0076] S1, obtain multi-source data and preprocess them;

[0077] S1-1, obtain high-resolution stereo satellite stereo image pairs and preprocess them to obtain high-resolution stereo image pairs and orthophoto multispectral images;

[0078] Among them, the preprocessing process of high-resolution stereo satellite stereo image pairs includes but is not limited to radiation correction, atmospheric correction, geometric correction, cropping, etc.;

[0079] Gaofen-7 is my country's first civilian sub-meter optical transmission stereo mapping satellite, which has the same orbit forward and backward stereo imaging capabilities and sub-meter spatial resolution advantages, and can meet the data needs of basic mapping, global geographic information security, urban and rural construction monitoring and evaluation, agricultural survey and statistics, etc. In this embodiment, the data of Gaofen-7 is used as the high-resolution stereo satellite stereo image pair data.

[0080] One forward image data DLC and one rearward image data BWD of the Gaofen-7 satellite covering the second ring road of Shijiazhuang City were obtained in the same period. The two images constitute a stereo pair, of which DLC contains a 0.8m panchromatic image and a 3.2m multispectral image, and BWD contains a 0.65m panchromatic image and a 2.6m multispectral image. Both multispectral images contain four bands: red, green, blue and near infrared. The panchromatic image of DLC and the panchromatic image of BWD are used as high-resolution stereo pairs. First, the panchromatic image and multispectral image of BWD are radiometrically calibrated and atmospherically corrected respectively. Then, the panchromatic image and multispectral image after atmospheric correction are band-fused to obtain a multispectral image with a resolution of 0.65m. Then, the multispectral image after band fusion is geometrically corrected to obtain an orthophoto multispectral image. Finally, the high-resolution stereo pair and the orthophoto multispectral image are cropped according to the range of the area to be identified to obtain the high-resolution stereo pair and the orthophoto multispectral image of the area to be identified.

[0081] S1-2, obtaining high-resolution nighttime light satellite images and preprocessing them to obtain nighttime light images;

[0082] Among them, the preprocessing of night light images includes but is not limited to radiation correction, atmospheric correction, cropping, etc.;

[0083] Jilin-1 07 is the first high-resolution night light satellite with on-board calibration. It can obtain fine night light images with three spectral bands of red, green and blue and a spatial resolution of 0.92m. Its images can show detailed light information inside the city. This embodiment uses Jilin-1 night light satellite images as high-resolution night light satellite images.

[0084] Four Jilin-1-07 satellite images covering the second ring road of Shijiazhuang at the same time as Gaofen-7 were obtained. First, the four images were radiometrically calibrated and atmospherically corrected, and then the four images after atmospheric correction were spliced. Finally, the spliced ​​images were cropped according to the range of the area to be identified, and the night light images were obtained after preprocessing.

[0085] S1-3, obtaining initial point of interest data and preprocessing, the initial point of interest data is referred to as initial POI data, and the preprocessed POI data is obtained after preprocessing;

[0086] Preprocessing of initial POI data includes but is not limited to data correction;

[0087] Since the initial POI data collected has various problems such as classification errors and missing coordinate information, preprocessing such as data correction is required. Data correction specifically means correcting the misclassified points; deleting the points with missing information; and deleting POIs whose functional characteristics are not obvious and cannot clearly represent the functional characteristics of the city (such as public toilets, bus stops, etc.).

[0088] The initial POI data acquisition channels include commercial map service providers such as Baidu Maps and AutoNavi Maps, as well as relevant government departments, etc. In this embodiment, the POI data of AutoNavi Maps is used as the source of the initial POI data.

[0089] Obtain the POI data of the Amap of Shijiazhuang area at the same time as the above remote sensing images. First, vectorize the original POI data in CSV table format to form POI data point data, and then crop it according to the range of the area to be identified to obtain the POI point data within the range of the area to be identified. Then, according to the above preprocessing process, classify and correct the misclassified POI points; delete the POI points with missing information; delete the POIs whose functional characteristics are not obvious and cannot clearly represent the functional characteristics of the city (such as public toilets, bus stops, etc.), and finally obtain the preprocessed POI data within the range of the area to be identified.

[0090] S1-4, acquiring OSM data and preprocessing it, and obtaining preprocessed OSM data after the preprocessing;

[0091] Preprocessing of OSM data includes but is not limited to registration, deletion, and closure;

[0092] Among them, OSM data is road network data, which refers to the road network data on the OpenStreetMap (OSM) platform, and its attribute information includes road name, type, width, length, etc. OSM road network data can be used to divide urban functional area units.

[0093] There are many messy roads, dead-end roads and repeated roads in the OSM road network data, which need to be preprocessed. The specific operations are to delete some messy roads, dead-end roads and internal roads that are less important and have smaller analysis result images; delete the repeated roads; and connect and close the more important dead-end roads in combination with remote sensing satellite images.

[0094] In this embodiment, the initial OSM data within the Second Ring Road of Shijiazhuang City is obtained from the OSM website, and processed according to the above-mentioned preprocessing process, and some messy roads, dead-end roads, internal roads and repeated roads with low importance and small analysis result images are deleted; the dead-end roads with high importance are connected and closed with the orthophoto multispectral image obtained in step S1-1, and finally the preprocessed OSM data within the range of the area to be identified is obtained.

[0095] S1-5, obtaining initial building function classification data, preprocessing the attributes of the initial building function classification data, and obtaining preprocessed building function classification data after preprocessing.

[0096] Among them, the attributes of the initial building function classification data include building function category, building area, etc.;

[0097] The initial building function classification data is obtained from commercial map service providers such as Baidu Maps and Amap, geographic data service providers such as BIGEMAP, relevant government departments, and self-production.

[0098] This embodiment uses the self-produced building function classification data within the second ring of Shijiazhuang City as the data source of the initial building function classification data, and its data format is a vector surface format, and the attribute information includes building area, building function category, and building roof category, etc. The initial building function classification data is preprocessed, including checking and correcting the attribute information such as the building function category, and the preprocessed building function classification data is obtained after preprocessing.

[0099] S-2, based on the preprocessed OSM data obtained in step S1-4, the line layer is superimposed with the surface layer of the area to be identified, and the area to be identified is subjected to line cutting surface operation according to the preprocessed OSM data to obtain the urban functional area plot unit data of the area to be identified, totaling 1284.

[0100] S3, based on the high-resolution stereo image pair obtained in S1-1, invert the height of the surface objects to obtain the resampled nDSM data; refer to Figure 2 , the specific process is as follows:

[0101] S3-1, based on the high-resolution stereo image pair obtained in S1-1, image a and image b constitute a stereo image pair of the same area; image a and image b are extracted with a binocular stereo matching algorithm to obtain digital surface model data, referred to as DSM data;

[0102] In this embodiment, based on the high-resolution stereo image pair obtained in step S1-1, binocular stereo matching and height recalculation are performed through the Generate Point Clouds and SM by Dense Image Matching tool in the ENVI software to obtain DSM data of the area to be identified. The pixel size is 0.8m, and its value includes the height of the terrain and the height of the surface objects.

[0103] S3-2, filtering the DSM data obtained in step S3-1 to obtain digital elevation model data representing ground height, referred to as DEM data;

[0104] In this embodiment, based on the DSM data obtained in step S3-1, the DSM filtering tool of SAGA GIS is used to filter the height of the surface objects to obtain DEM data containing only the terrain height;

[0105] S3-3, performing a subtraction operation on the DSM data obtained in step S3-1 and the DEM data obtained in step S3-2 to obtain normalized digital surface model data representing the relative height of surface objects, referred to as nDSM data;

[0106] In this embodiment, based on the programmed band operation tool, the DSM data obtained in step S3-1 and the DEM data obtained in step S3-2 are subtracted to obtain nDSM data, whose value is the height of the surface object relative to the ground.

[0107] S3-4, resampling the nDSM data obtained in step S3-3 with reference to the orthophoto multispectral image obtained in step S1-1, and obtaining resampled nDSM data after resampling;

[0108] The nDSM data with a pixel size of 0.8 m obtained in step S3-3 is resampled with reference to the 0.65 m resolution of the orthophoto multispectral image obtained in step S1-1, and the resampled nDSM data with a pixel size of 0.65 m is obtained after resampling.

[0109] S4, based on the orthophoto multispectral image obtained in step S1-1 and the resampled nDSM data obtained in step S3, a multidimensional feature image including a multispectral channel and an altitude channel is obtained by superimposing the image. Figure 3 ;

[0110] Based on the programmed band operation tool, the orthophoto multispectral image obtained in step S1-1 and the resampled nDSM data obtained in step S3 are superimposed to obtain a multidimensional feature image with a total of 5 bands including multispectral channels (red, green, blue, near infrared) and altitude channels, with a resolution of 0.65m.

[0111] S5: For the multi-dimensional feature image obtained in step S4, image segmentation is performed based on the semantic segmentation model of deep learning to obtain the vectorized ground feature segmentation result. The process is as follows: Figure 3 ;

[0112] S5-1, slicing the multi-dimensional feature image according to the same preset fixed size to obtain an image slice data set, the data format is TIFF, and the data type is floating point (Float);

[0113] S5-2, input the image slices in the image slice data set into the trained UNetFormer semantic segmentation model in sequence to perform image segmentation, and obtain the set object segmentation result corresponding to the image slice set, the data format is PNG format, the data type is integer (Int), and the value is the number of the object type;

[0114] The present invention selects the UnetFormer model as the model for remote sensing image ground object segmentation. The model adopts a Transformer-based decoder, combined with a global-local attention mechanism, and can comprehensively use the local feature extraction capability of convolutional neural networks (CNN) and the global spatial context modeling capability of Transformer to improve the feature extraction capability of images and the accuracy of image segmentation. At the same time, the model has high efficiency and is particularly suitable for semantic segmentation of remote sensing urban scene images. On the other hand, in terms of model input, the present invention introduces nDSM data representing height features in addition to the multispectral bands containing radiation features. By adding elevation information, ground objects with similar appearance features but different heights can be segmented, such as building roofs and hardened ground, thereby improving the accuracy of ground object segmentation.

[0115] S5-3, merging the set object segmentation results corresponding to the image slice set to obtain the regional object segmentation result, the data format is TIFF, and the data type is Float;

[0116] S5-4, vectorizing the regional feature segmentation result to obtain a vectorized feature segmentation result, wherein the attribute information of the vectorized feature segmentation result includes but is not limited to the type of feature, the area of ​​the feature, etc.;

[0117] S6, performing spatial association and statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the vectorized feature segmentation result obtained in step S5, and calculating the feature distribution characteristics F of the urban functional area cover , F cover The meaning of is the area proportion of each type of land feature in each urban functional area plot unit;

[0118] Among them, the distribution characteristics of land features in urban functional areas are cover The calculation formula is:

[0119]

[0120] Among them, F cover is the distribution characteristics of the land features in the urban functional area, S i is the area of ​​the i-th type of feature in each urban functional area plot unit; i∈1,2,3…m is the number of the feature category; m is the total number of feature categories;

[0121] S7, performing spatial correlation and surface statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the pre-processed night light image obtained in step S1-2, and calculating the night light index F representing the characteristics of human activities according to the brightness values ​​indicated by the pixels located in the urban functional area plot unit in the night light image. light ;

[0122] Among them, the night light index F light The calculation formula is:

[0123]

[0124] Among them, F light is the night light index representing the human activity characteristics of each building, S is the area of ​​the building's outline, n is the number of pixels of the night light image located within the building's outline, and DNi is the pixel value of each pixel of the night light image located within the building's outline;

[0125] S8, reclassifying the preprocessed POI data obtained in step S1-3 according to the mapping relationship between the POI data category and the urban functional area category, reclassifying the category of the preprocessed POI data into a category that can be used to mark the urban functional area, and setting different weights for each type of POI point to obtain reclassified POI data;

[0126] Among them, the mapping relationship between POI data categories and urban functional area categories is:

[0127] f:type poi →type region ;

[0128] Among them, type poi Is the category of POI data, type region It is the category of urban functional area. For any POI category, there is a unique urban functional area category corresponding to it;

[0129] S9, performing spatial association and statistical analysis on the urban functional area plot unit data of the to-be-identified area obtained in step S2 and the reclassified POI data obtained in step S8, and calculating the social function characteristics F of the urban functional area use , F use The meaning of is the proportion of the number and weight product of each type of POI points inside and around each urban functional area plot unit;

[0130] Among them, the social function characteristics of urban functional areas are use The calculation formula is:

[0131]

[0132] Among them, F use represents the social function characteristics of each urban functional area, F use The meaning is the proportion of the product of the number of each POI category and its weight inside and around each urban functional area plot unit to the sum of the total number of associated POIs and the product of their weights, n i is the number of POIs of the i-th category associated with the plot unit in the functional area of ​​the city, w i is the weight value of the i-th category POI, and m is the total number of POI categories;

[0133] S9-1, setting a certain buffer distance for each urban functional area plot unit in the urban functional area plot unit data of the area to be identified obtained in step S2, and obtaining cached urban functional area plot unit data after buffering;

[0134] S9-2, spatially connecting the buffered urban functional area plot unit data obtained in step S9-1 with the reclassified POI data obtained in step S8, and calculating the number and type of reclassified POI data associated with each urban functional area plot unit;

[0135] S9-3, according to the number, type and weight of the reclassified POI data associated with each urban functional area plot unit obtained in step S9-2, calculate the social function characteristics F of the urban functional area use .

[0136] S10, performing spatial association and statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the pre-processed building function classification data obtained in steps S1-5, and calculating the building distribution characteristics F of the urban functional area building , F building The meaning is the number ratio, area ratio and average building height of each building function category in each urban functional area unit;

[0137] Among them, the building distribution characteristics of urban functional areas are building The calculation formula is:

[0138]

[0139] Among them, F building It represents the building distribution characteristics of each urban functional area; F num is the proportion of buildings of each building function type in each urban functional area plot unit; F area is the area ratio of each building function type in each urban functional area plot unit; F heig h t is the average height of all buildings in each urban functional area plot unit; n i is the number of buildings of each building function category in each urban functional area plot unit; S i is the sum of the areas of buildings of each building function category in each urban functional area plot unit; i∈1,2,3…m is the number of each building function category; m is the number of all building function categories; h j is the height of each building in each urban functional area plot unit; j∈1,2,3…k is the serial number of each building in each urban functional area plot unit; k is the number of buildings in each urban functional area plot unit.

[0140] S11, for the land feature distribution feature F of the urban functional area obtained in step S6 cover , the night light index F representing the characteristics of human activities in the urban functional area obtained in step S7 light , the social function characteristics F of the urban function area obtained in step S9use and the building distribution characteristics F of the urban functional area obtained in step S10 building , identification of urban functional areas based on XGBoost ensemble classifier;

[0141] S11-1, the land feature distribution characteristics F of the urban functional area cover , Night Light Index F light 、Social Functional Characteristics F use and building distribution characteristics F building Connect them in series to obtain the comprehensive characteristics F of the urban functional areas;

[0142] Among them, the calculation formula of the comprehensive characteristics F of urban functional areas is:

[0143] F=CONTACT(F cover , F light , F use , F building )

[0144] Among them, F represents the comprehensive characteristics of urban functional areas, F cover represents the distribution characteristics of land features in urban functional areas, F cover The night light index that represents the characteristics of human activities in urban functional areas, F use is the social function characteristic of the urban functional area, F building It is the architectural distribution characteristic of urban functional areas.

[0145] S11-2, constructing a training sample set, taking the comprehensive feature F of the urban functional area obtained in step S11-1 as input and the real category of the urban functional area as output to train the XGBoost integrated classifier;

[0146] S11-3, using the XGBoost integrated classifier trained in step S11-2, taking the comprehensive feature F of the urban functional area to be identified as input, the urban functional area identification is performed to obtain the urban functional area identification results within the second ring road of Shijiazhuang City, such as Figure 4 shown.

[0147] Among them, XGBoost is an integrated learning model, a distributed gradient boosting machine learning algorithm that performs optimization under the gradient boosting framework, provides parallel tree boosting functions, and is widely used in regression, classification, and ranking problems.

[0148] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying urban functional areas based on multi-source feature fusion, characterized in that: The following steps are involved: S1, obtain multi-source data and preprocess them; S1-1, obtain high-resolution stereo satellite stereo image pairs and preprocess them to obtain high-resolution stereo image pairs and orthophoto multispectral images; S1-2, obtaining high-resolution nighttime light satellite images and preprocessing them to obtain nighttime light images; S1-3, obtaining initial POI data and preprocessing it to obtain preprocessed POI data; S1-4, obtaining initial OSM data and preprocessing it to obtain preprocessed OSM data; S1-5, obtaining initial building function classification data, preprocessing the attributes of the initial building function classification data, and obtaining preprocessed building function classification data, wherein the attributes of the initial building function classification data include building function category and building area; S2, superimposing the line layer of the pre-processed OSM data obtained in step S1-4 with the surface layer of the area to be identified, performing line cutting surface operation on the area to be identified according to the pre-processed OSM data, and obtaining the urban functional area plot unit data of the area to be identified; S3, based on the high-resolution stereo image pair obtained in S1-1, the height of the surface objects is inverted to obtain the resampled nDSM data; S4, superimposing the orthophoto multispectral image obtained in step S1-1 and the resampled nDSM data obtained in step S3 to obtain a multidimensional feature image including a multispectral channel and an altitude channel; S5, performing image segmentation based on a deep learning semantic segmentation model on the multi-dimensional feature image obtained in step S4 to obtain a vectorized feature segmentation result; S6, performing spatial association and statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the vectorized feature segmentation result obtained in step S5, and calculating the feature distribution characteristics F of the urban functional area cover , F cover Indicates the area proportion of each type of land feature in each urban functional area plot unit; S7, performing spatial correlation and surface statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the preprocessed night light image obtained in step S1-2, and calculating the night light index F representing the characteristics of human activities according to the brightness values ​​indicated by the pixels located in the urban functional area unit in the night light image. light ; S8, reclassifying the preprocessed POI data obtained in step S1-3 according to the mapping relationship between POI categories and urban functional area categories, reclassifying the categories of the preprocessed POI data into categories for marking urban functional areas, and setting weights for each type of POI point to obtain reclassified POI data; S9, performing spatial association and statistical analysis on the urban functional area plot unit data of the to-be-identified area obtained in step S2 and the reclassified POI data obtained in step S8, and calculating the social function characteristics F of the urban functional area use , F use Indicates the proportion of the number and weight product of each type of POI points inside and around each urban functional area plot unit; S10, performing spatial association and statistical analysis on the urban functional area plot unit data of the area to be identified obtained in step S2 and the pre-processed building function classification data obtained in steps S1-5, and calculating the building distribution characteristics F of the urban functional area building , F building Indicates the number, area and average building height of each building function category in each urban functional area unit; S11, for the land feature distribution feature F of the urban functional area obtained in step S6 cover , the night light index F representing the characteristics of human activities in the urban functional area obtained in step S7 light , the social function characteristics F of the urban function area obtained in step S9 use and the building distribution characteristics F of the urban functional area obtained in step S10 building , identification of urban functional areas based on XGBoost ensemble classifier; S11-1, the land feature distribution characteristics F of the urban functional area cover , Night Light Index F light 、Social Functional Characteristics F use and building distribution characteristics F building Connect them in series to obtain the comprehensive characteristics F of the urban functional areas; S11-2, constructing a training sample set, taking the comprehensive feature F of the urban functional area obtained in step S11-1 as input and the real category of the urban functional area as output to train the XGBoost integrated classifier; S11-3, using the XGBoost integrated classifier trained in step S11-2, taking the comprehensive features F of the urban functional areas to be identified as input, identify the urban functional areas and obtain the urban functional areas identification results.

2. The method for identifying urban functional areas based on multi-source feature fusion according to claim 1 is characterized in that: The S3 comprises the following steps: S3-1, based on the high-resolution stereo image pair obtained in S1-1, the digital surface model data is extracted through the binocular stereo matching algorithm, referred to as DSM data; S3-2, filtering the DSM data obtained in step S3-1 to obtain digital elevation model data representing ground height, referred to as DEM data; S3-3, performing a subtraction operation on the DSM data obtained in step S3-1 and the DEM data obtained in step S3-2 to obtain normalized digital surface model data representing the relative height of surface objects, referred to as nDSM data; S3-4, resampling the nDSM data obtained in step S3-3 with reference to the orthophoto multispectral image obtained in step S1-1 to obtain resampled nDSM data.

3. The method for identifying urban functional areas based on multi-source feature fusion according to claim 1 is characterized in that: The S5 comprises the following steps: S5-1, slicing the multi-dimensional feature image according to the same preset size to obtain an image slice data set; S5-2, inputting the image slices in the image slice data set into the trained semantic segmentation model in sequence to perform image segmentation, and obtaining a set object segmentation result corresponding to the image slice set; S5-3, merging the collective feature segmentation results to obtain a regional feature segmentation result; S5-4, vectorizing the regional feature segmentation result to obtain a vectorized feature segmentation result, wherein the attributes of the vectorized feature segmentation result include the type of feature and the area of ​​the feature.

4. The method for identifying urban functional areas based on multi-source feature fusion according to claim 1 is characterized in that: The distribution characteristics of the land features in the urban functional area of ​​S6 cover The calculation formula is: Among them, F cover is the distribution characteristics of the land features in the urban functional area, S i is the area of ​​the i-th type of land feature in each urban functional area plot unit; i∈1,2,3…m is the number of the land feature category; m is the total number of land feature categories.

5. The method for identifying urban functional areas based on multi-source feature fusion according to claim 1 is characterized in that: The night light index F described in step S7 light The calculation formula is: Among them, F light It is the night light index representing the human activity characteristics of each building, S is the area of ​​the building's outline, n is the number of pixels of the night light image located within the building's outline, and DNi is the pixel value of each pixel of the night light image located within the building's outline.

6. The method for identifying urban functional areas based on multi-source feature fusion according to claim 1 is characterized in that: The formula for the mapping relationship between the categories of POI data and the categories of urban functional areas in S8 is: f:type poi →type region ; Among them, type poi Is the category of POI data, type region It is the category of urban functional area. For any POI category, there is a unique urban functional area category corresponding to it.

7. The method for identifying urban functional areas based on multi-source feature fusion according to claim 1 is characterized in that: The S9 comprises the following steps: S9-1, presetting a buffer distance for each urban function area plot unit in the urban function area plot unit data of the area to be identified obtained in step S2 and buffering the data to obtain buffered urban function area plot unit data; S9-2, spatially connecting the buffered urban functional area plot unit data obtained in step S9-1 with the reclassified POI data obtained in step S8, calculating the number and type of reclassified POI data associated with each urban functional area plot unit data and presetting the weight; S9-3, according to the number, type and weight of the reclassified POI data associated with the plot unit data of each urban functional area obtained in step S9-2, calculate the social function characteristics F of the urban functional area use .

8. The method for identifying urban functional areas based on multi-source feature fusion according to claim 7 is characterized in that: The social function characteristics of the urban functional area in S9 use The calculation formula is: Among them, F use It represents the social function characteristics of each urban functional area, which indicates the proportion of the product of the number of each POI category and its weight inside and around the plot unit of each urban functional area to the sum of the total number of associated POIs and the product of their weights. i is the number of POIs of the i-th category associated with the plot unit in the functional area of ​​the city, w i is the weight value of the i-th category POI, and m is the total number of POI categories.

9. The method for identifying urban functional areas based on multi-source feature fusion according to claim 1 is characterized in that: The building distribution characteristics F of the urban functional area in S10 building The calculation formula is: F building =CONTACT(F num ,F area ,F height ); Among them, F building It represents the building distribution characteristics of each urban functional area; F num is the proportion of buildings of each building function type in each urban functional area plot unit; F area is the area ratio of each building function type in each urban functional area plot unit; F heig h t is the average height of all buildings in each urban functional area plot unit; n i is the number of buildings of each building function category in each urban functional area plot unit; S i is the sum of the areas of buildings of each building function category in each urban functional area plot unit; i∈1,2,3…m is the number of each building function category; m is the number of all building function categories; h j is the height of each building in each urban functional area plot unit; j∈1,2,3…k is the serial number of each building in each urban functional area plot unit; k is the number of buildings in each urban functional area plot unit.

10. The method for identifying urban functional areas based on multi-source feature fusion according to claim 1, characterized in that: The calculation formula for the comprehensive characteristics F of the urban functional area in S11-1 is: F=CONTACT(F cover ,F light ,F use ,F building ); Among them, F represents the comprehensive characteristics of urban functional areas, F cover represents the distribution characteristics of land features in urban functional areas, F cover The night light index that represents the characteristics of human activities in urban functional areas, F use is the social function characteristic of the urban functional area, F building It is the architectural distribution characteristic of urban functional areas.

Images