A method for rapid construction of urban digital base based on multi-source remote sensing data

Abstract

The present invention discloses a method for quickly constructing a city digital base based on multi-source remote sensing data, and belongs to the field of remote sensing technology. The present invention comprises the following steps: obtaining high-resolution orthophoto remote sensing images, high-resolution stereo image pairs, OSM road network data and POI data, and performing preprocessing; superimposing OSM road network data to generate plot unit data; classifying plots in combination with POI data; extracting building outlines from orthophotos using a deep learning instance segmentation model; classifying building functions in combination with POI data; extracting and processing digital surface models and digital elevation models through high-resolution stereo image pairs to obtain building height information; finally generating city three-dimensional model data, and integrating various data to form a city digital base data set. The method can quickly and accurately construct a city digital base, and provide effective support for urban planning and management.

Description

Technical Field

[0001] The invention relates to the field of remote sensing technology, and in particular to a method for quickly constructing a city digital base based on multi-source remote sensing data. Background Art

[0002] Against the backdrop of the rapid development of the digital economy, the urbanization process has brought unprecedented challenges and opportunities. Traditional urban planning and management methods are unable to cope with these changes. The emergence of digital twin technology provides a new approach to solving these problems. Digital twin technology can capture the spatial and structural characteristics of cities more accurately and efficiently, and provides strong data support for urban planning, management, and decision-making.

[0003] The application basis of urban digital twin technology is the construction of urban digital base. At present, the construction of urban three-dimensional digital base mainly includes a variety of methods such as BIM model integration, 3D software modeling, drone remote sensing surveying and mapping (including three-dimensional laser scanning and oblique photogrammetry) and satellite remote sensing. Each method has its advantages and disadvantages. For example, the method based on BIM model integration has mature technology and high model accuracy, but the model data volume is large, the rendering speed is slow, there is a lack of data accumulation, and the production cycle is long, which is difficult to meet the needs of urban digital twin applications in large areas. Several other methods, including 3D software modeling and drone surveying and mapping (three-dimensional laser scanning, oblique photogrammetry), also have the disadvantages of high production cost, long production cycle, large workload, and difficulty in achieving periodic monitoring. The satellite remote sensing-based method can take advantage of the characteristics of satellite remote sensing images, such as wide coverage, short update cycle, and rich data resources, to quickly build a digital data base for cities in large areas and provide support for urban digital twin applications. In addition, compared with other methods, the urban digital base construction solution based on satellite remote sensing has the advantages of low production cost, high efficiency, and the ability to achieve periodic updates. Therefore, the urban digital base construction technology based on multi-source remote sensing data has broad application prospects. Summary of the invention

[0004] The purpose of the present invention is to provide a method for quickly constructing a city digital base based on multi-source remote sensing data to solve the problems raised in the above-mentioned background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] A method for rapidly constructing a city digital base based on multi-source remote sensing data, characterized in that the method comprises the following steps:

[0007] Step S1. Acquire multi-source remote sensing data and perform preprocessing, wherein the multi-source remote sensing data includes high-resolution orthophoto remote sensing images, high-resolution stereo image pairs, OSM road network data, and POI data of the study area;

[0008] Step S2. Overlay the OSM road network data with the study area, perform line cutting and surface operations on the study area according to the OSM road network data, generate plot units, and obtain the current land use plot unit data of the study area;

[0009] Step S3. Overlay and analyze the POI data of the study area in step S1 and the current land use unit data of the study area in step S2, classify the current land use unit data by setting classification rules, and obtain the land use classification data of the study area;

[0010] Step S4. Using the deep learning-based instance segmentation model to perform building extraction on the high-resolution orthophoto remote sensing image preprocessed in step S1, obtain a building extraction result set corresponding to the image slice set; analyze the building extraction result set to obtain building outline vector data of the study area;

[0011] Step S5. Overlay analysis is performed on the POI data of the study area in step S1 and the building outline vector data of the study area in step S4, and building function classification is performed on the buildings in the study area by setting classification rules to obtain building function classification data of the study area;

[0012] Step S6. Based on the high-resolution stereo image pair in step S1, extract the digital surface model, filter and generate the digital elevation model, obtain the normalized digital surface model by difference, and analyze it in combination with the building outline vector data in step S4, so as to obtain the building height information in the study area;

[0013] Step S7. Based on the building outline vector data and building height information of the study area, a three-dimensional model is reconstructed to obtain the building three-dimensional model data of the study area; combined with high-resolution orthophoto remote sensing images, OSM road network data, POI data, land use classification data of the study area, building function classification data of the study area and building three-dimensional model data of the study area, registration and integration are performed to obtain the urban digital base data set of the study area.

[0014] Combining high-resolution remote sensing images with multi-source spatiotemporal data of cities, and comprehensively utilizing remote sensing, big data analysis, artificial intelligence and other technologies, we can quickly build a digital base for cities and improve the accuracy and practicality of data. At the same time, we can use the latest deep learning and machine learning algorithms to effectively improve the efficiency and accuracy of data processing. In addition, exploring the widespread use of urban digital bases in urban planning, management applications and other fields will provide new perspectives and tools for research and practice in related fields, and further explore and give play to the value of remote sensing data. This method has the advantages of low cost, good performance, and can realize the rapid construction of digital bases for cities in large areas, and is easy to promote and apply on a large scale.

[0015] Further, step S1 includes:

[0016] S1-1. Acquire high-resolution remote sensing images and perform preprocessing, the preprocessing process including but not limited to radiometric calibration, atmospheric correction, geometric correction and cropping, so as to obtain high-resolution orthophoto remote sensing images after preprocessing;

[0017] S1-2. Obtain a high-resolution stereo image pair from a high-resolution stereo satellite and perform preprocessing to obtain a high-resolution stereo image pair after preprocessing of the study area; wherein the preprocessing of the stereo image pair includes but is not limited to radiation correction, atmospheric correction, geometric correction, cropping, etc.;

[0018] S1-3. Obtain OSM road network data and perform preprocessing, including but not limited to data correction and registration, to obtain OSM road network data of the study area after preprocessing; wherein OSM road network data refers to road network data on the OpenStreetMap (OSM) platform; its attribute information includes road name, type, width, length, etc. OSM road network data can be used for the division of urban plot units;

[0019] S1-4, obtain POI data and perform preprocessing, including but not limited to data correction and registration, to obtain the preprocessed POI data of the research area; POI data refers to point of interest data, which abstracts urban functions in life into spatial point-shaped geographic information data. In addition to the spatial information such as the latitude and longitude coordinates and address of the geographic entity, it also connects various attribute information. Such as POI name, category, administrative division, etc. POI data can enable research on spatiotemporal behavior, urban planning, geographic information, etc.

[0020] Furthermore, in step S3, the existing land parcel unit data is classified by setting classification rules to obtain the land classification data of the study area. The specific process is as follows:

[0021] According to the urban construction land classification system, the POI data categories are reclassified into categories that can be used to mark the functions of plot units, and different weights are set for each type of POI data point based on expert scores. The current land plot unit data of the study area are superimposed and analyzed with the reclassified POI data. According to the number, type and weight of different types of POI data points in the plot units, the functional classification of the current land plot units in the study area is carried out based on the word frequency-inverse text frequency index method.

[0022] Further, step S4 includes:

[0023] S4-1. Slice the preprocessed high-resolution orthophoto remote sensing image according to a fixed size to obtain an image slice data set;

[0024] S4-2. Input the image slice data set into the instance segmentation model based on deep learning to extract the building, and obtain the building extraction result set corresponding to the image slice set;

[0025] S4-3. Vectorize the building extraction result set to obtain the building top outline vector data of the study area;

[0026] S4-4. Regularize the graph of the building top outline vector data in the study area to obtain regularized building top outline vector data in the study area;

[0027] S4-5, correcting the regularized building top outline vector data of the study area, correcting the building top outline to the building bottom footprint, and obtaining the building outline vector data of the study area.

[0028] Furthermore, the deep learning-based instance segmentation model in S4-2 is based on the Mask R-CNN model and is improved based on the characteristics of remote sensing images to obtain the improved instance segmentation model RS-Mask R-CNN; the specific improvement process includes simultaneously inputting the RGB three-channel data and elevation channel data of the image slice in the model input stage, and adding a dual attention module after the feature extraction backbone network, wherein the dual attention module includes a position attention branch and a channel attention branch.

[0029] Furthermore, in step S5, the buildings in the study area are classified by building functions by setting classification rules to obtain building function classification data in the study area; the specific process is as follows:

[0030] Referring to the POI classification rules, the urban building function classification system with 3 major categories and 12 minor categories was established through induction and merging. According to the urban building function classification system, the POI data categories were reclassified into categories that can be used to mark building functions, and different weights were set for each type of POI points based on expert scores. The building outline data of the study area were superimposed and analyzed with the reclassified POI data. According to the number, type and weight of different types of POI points in the building outline, the word frequency-inverse text frequency index method was used to classify the building functions of the building outline data of the study area.

[0031] The term frequency-inverse text frequency index method refers to the TF-IDF index method, and the formula of the TF-IDF index method is:

[0032]

[0033]

[0034] TF-IDF i,j =TF i,j ×IDFi

[0035] Where i represents a word, j represents a document; n i,j represents the frequency of occurrence of i in document j; ∑ k n k,j represents the total number of occurrences of all words in document j, where k represents the index of all terms in document j; TF i,j represents the frequency of term i in document j; IDF i The inverse document frequency D of term i represents the total number of documents; {j:t i ∈d j} represents the total number of documents containing word i, that is, n i,j ≠0 number of documents; TF-IDF i,j Represents the TF-IDF value of term i in document j.

[0036] Furthermore, the classification rules set in step S3 and step S5 are to perform functional classification using the word frequency-inverse text frequency index method, and the specific classification steps are as follows:

[0037] Collect POI data in the study area, and each POI data includes the type and location information of the POI; divide the study area into several plot units or building outlines and record them as spatial units; according to the division of spatial units, record the number of each type of POI in each spatial unit, and for each POI type in each spatial unit, calculate the word frequency TF according to the formula i,j And the corresponding inverse document frequency IDF i , thus obtaining TF-IDF i,j ; According to the expert scores, different weights are set for each type of POI, and the corresponding TF-IDF values ​​are weighted; for each spatial unit, the weighted TF-IDF values ​​of all POI types are aggregated to obtain a comprehensive score; according to the comprehensive score, spatial units with equal comprehensive scores are classified into one category, and traversed to all spatial units to achieve functional classification.

[0038] The above-mentioned spatial unit represents a plot unit or a building outline. The classification rules set in step S3 and step S5 are the same, but the applied data are different. In step S3, the classification rules are used to classify the current land plot unit data to obtain the land classification data of the study area; in step S5, the classification rules are used to classify the buildings in the study area according to their building functions to obtain the building function classification data of the study area.

[0039] Further, step S6 includes:

[0040] S6-1. Based on the high-resolution stereo image pair obtained in S1-2, the digital surface model DSM data of the study area is extracted by using the image dense matching algorithm;

[0041] S6-2. Filter the digital surface model DSM data of the study area obtained in step S6-1 to obtain the digital elevation model DEM data of the study area;

[0042] S6-3. Differentiate the extracted digital surface model DSM data with the digital elevation model DEM data to obtain normalized digital surface model nDSM data representing the height of various surface objects in the study area;

[0043] S6-4. Fit the building outline vector data of the study area obtained in step S4 with the normalized digital surface model nDSM data representing the height of surface objects in the above-mentioned study area, and perform surface analysis. According to the height information indicated by the pixels located in the building outline vector data in the normalized digital surface model nDSM data, the building height information in the study area is obtained.

[0044] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0045] By comprehensively utilizing multi-source data such as high-resolution remote sensing images, stereo image pairs, OSM road network data, and POI data, the accuracy and completeness of urban digital base data can be effectively improved. The cross-validation and fusion of various data sources can make up for the shortcomings of a single data source, thereby obtaining more accurate urban spatial information. The use of deep learning models for building extraction and classification, combined with the improved Mask R-CNN algorithm, has greatly improved the efficiency of building urban digital bases. Compared with traditional manual modeling methods, this method can generate three-dimensional digital models of cities more quickly, providing timely support for urban planning and management. By setting classification rules and combining POI data and TF-IDF index method for functional classification, urban digital base data is made more comprehensive and rich, which can not only reflect the morphological characteristics of buildings, but also provide the functional attributes of buildings, providing more diversified data support for urban management and development. The deep learning model is combined with remote sensing images and elevation data, and improvements such as dual attention modules are made to make the model better adapt to the characteristics of remote sensing images, thereby The adaptability and accuracy of building extraction have been improved. This adaptive optimization can be better applied to remote sensing data of different types and qualities. The building height is extracted by differential calculation of digital surface model and building outline vector data combined with elevation data, and finally a true and accurate three-dimensional model is constructed. This method can more realistically reflect the terrain and architectural features of the city, and improve the authenticity and credibility of the data. Through the registration and integration of multiple data sources, a complete urban digital base data set is obtained, ensuring the consistency and accuracy of the data. This comprehensive integration method can provide a comprehensive and reliable set of digital base data for urban management and planning, and promote the development of smart cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0047] Figure 1 It is a schematic diagram of a method for quickly constructing a city digital base based on multi-source remote sensing data of the present invention;

[0048] Figure 2 It is an improved RS-Mask R-CNN model structure of a method for rapidly constructing a city digital base based on multi-source remote sensing data of the present invention;

[0049] Figure 3 The invention discloses a dual attention module DAM structure of a method for rapidly constructing a city digital base based on multi-source remote sensing data. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] See also Figure 1 , the present invention provides a technical solution:

[0052] A method for rapidly constructing an urban digital base based on multi-source remote sensing data, characterized in that the method comprises the following steps:

[0053] Step S1. Acquire multi-source remote sensing data and perform preprocessing, wherein the multi-source remote sensing data includes high-resolution orthophoto remote sensing images, high-resolution stereo image pairs, OSM road network data, and POI data of the study area;

[0054] Step S2. Overlay the OSM road network data with the study area, perform line cutting and surface operations on the study area according to the OSM road network data, generate plot units, and obtain the current land use plot unit data of the study area;

[0055] Step S3. Overlay and analyze the POI data of the study area in step S1 and the current land use unit data of the study area in step S2, classify the current land use unit data by setting classification rules, and obtain the land use classification data of the study area;

[0056] Step S4. Using the deep learning-based instance segmentation model to perform building extraction on the high-resolution orthophoto remote sensing image preprocessed in step S1, obtain a building extraction result set corresponding to the image slice set; analyze the building extraction result set to obtain building outline vector data of the study area;

[0057] Step S5. Overlay analysis is performed on the POI data of the study area in step S1 and the building outline vector data of the study area in step S4, and building function classification is performed on the buildings in the study area by setting classification rules to obtain building function classification data of the study area;

[0058] Step S6. Based on the high-resolution stereo image pair in step S1, extract the digital surface model, filter and generate the digital elevation model, obtain the normalized digital surface model by difference, and analyze it in combination with the building outline vector data in step S4, so as to obtain the building height information in the study area;

[0059] Step S7. Based on the building outline vector data and building height information of the study area, a three-dimensional model is reconstructed to obtain the building three-dimensional model data of the study area; combined with high-resolution orthophoto remote sensing images, OSM road network data, POI data, land use classification data of the study area, building function classification data of the study area and building three-dimensional model data of the study area, registration and integration are performed to obtain the urban digital base data set of the study area.

[0060] Combining high-resolution remote sensing images with multi-source spatiotemporal data of cities, and comprehensively utilizing remote sensing, big data analysis, artificial intelligence and other technologies, we can quickly build a digital base for cities and improve the accuracy and practicality of data. At the same time, we can use the latest deep learning and machine learning algorithms to effectively improve the efficiency and accuracy of data processing. In addition, exploring the widespread use of urban digital bases in urban planning, management applications and other fields will provide new perspectives and tools for research and practice in related fields, and further explore and give play to the value of remote sensing data. This method has the advantages of low cost, good performance, and can realize the rapid construction of digital bases for cities in large areas, and is easy to promote and apply on a large scale.

[0061] Step S1 includes:

[0062] S1-1. Acquire high-resolution remote sensing images and perform preprocessing, the preprocessing process including but not limited to radiometric calibration, atmospheric correction, geometric correction and cropping, so as to obtain high-resolution orthophoto remote sensing images after preprocessing;

[0063] S1-2. Obtain a high-resolution stereo image pair from a high-resolution stereo satellite and perform preprocessing to obtain a high-resolution stereo image pair after preprocessing of the study area; wherein the preprocessing of the stereo image pair includes but is not limited to radiation correction, atmospheric correction, geometric correction, cropping, etc.;

[0064] S1-3. Obtain OSM road network data and perform preprocessing, including but not limited to data correction and registration, to obtain OSM road network data of the study area after preprocessing; wherein OSM road network data refers to road network data on the OpenStreetMap (OSM) platform; its attribute information includes road name, type, width, length, etc. OSM road network data can be used for the division of urban plot units;

[0065] S1-4, obtain POI data and perform preprocessing, including but not limited to data correction and registration, to obtain the preprocessed POI data of the research area; POI data refers to point of interest data, which abstracts urban functions in life into spatial point-shaped geographic information data. In addition to the spatial information such as the latitude and longitude coordinates and address of the geographic entity, it also connects various attribute information. Such as POI name, category, administrative division, etc. POI data can enable research on spatiotemporal behavior, urban planning, geographic information, etc.

[0066] In step S3, the existing land parcel unit data is classified by setting classification rules to obtain the land classification data of the study area. The specific process is as follows:

[0067] According to the urban construction land classification system, the POI data categories are reclassified into categories that can be used to mark the functions of plot units, and different weights are set for each type of POI data point based on expert scores. The current land plot unit data of the study area are superimposed and analyzed with the reclassified POI data. According to the number, type and weight of different types of POI data points in the plot units, the functional classification of the current land plot units in the study area is carried out based on the word frequency-inverse text frequency index method.

[0068] Step S4 includes:

[0069] S4-1. Slice the preprocessed high-resolution orthophoto remote sensing image according to a fixed size to obtain an image slice data set;

[0070] S4-2. Input the image slice data set into the instance segmentation model based on deep learning to extract the building, and obtain the building extraction result set corresponding to the image slice set;

[0071] S4-3. Vectorize the building extraction result set to obtain the building top outline vector data of the study area;

[0072] S4-4. Regularize the graph of the building top outline vector data in the study area to obtain regularized building top outline vector data in the study area;

[0073] S4-5, correcting the regularized building top outline vector data of the study area, correcting the building top outline to the building bottom footprint, and obtaining the building outline vector data of the study area.

[0074] The deep learning-based instance segmentation model in S4-2 is based on the Mask R-CNN model and is improved based on the characteristics of remote sensing images to obtain the improved instance segmentation model RS-Mask R-CNN; the specific improvement process includes simultaneously inputting the RGB three-channel data and elevation channel data of the image slice in the model input stage, and adding a dual attention module after the feature extraction backbone network, where the dual attention module includes a position attention branch and a channel attention branch.

[0075] In this embodiment, for the instance segmentation model based on deep learning, there are the following contents:

[0076] First, in the model input stage, the RGB three-channel data and elevation channel data of the image slice are input simultaneously. Adding elevation information can distinguish the hardened ground and road surface that are easily confused with buildings from the buildings, and at the same time improve the extraction accuracy of buildings of different heights and reduce confusion.

[0077] Secondly, remote sensing images are different from natural scene photos. Their scene contents are more complex and the scales of objects vary greatly. For example, the sizes of buildings in remote sensing image scenes vary greatly. There are large factories and commercial buildings, as well as small single-family buildings and self-built houses. The original Mask R-CNN model extracts buildings from remote sensing images, but its recognition ability for small targets is weak, which easily leads to the omission or misidentification of small buildings. The dual attention module (DAM) can model the semantic correlation of the spatial dimension and channel dimension of the model's feature map, improve the model's scene understanding ability and small target detection ability. By adding the DAM module after the backbone network of the model to extract features, the model's building extraction accuracy can be improved.

[0078] Among them, the improved RS-Mask R-CNN model structure is as follows Figure 2 As shown;

[0079] The model input is the original image with three RGB channels and the elevation map with a single channel of elevation information;

[0080] The model first inputs the original image and elevation map into the backbone network for feature extraction; then the extracted features are input into the dual attention module (DAM), which contains the position attention branch (PAM) and the channel attention branch (CAM); then the output of the DAM is input into the region generation network (RPN) module, which is responsible for generating candidate regions; then the candidate regions are aligned with the region of interest (ROI Align) operation, and the aligned features are input into two convolutional layers (Conv) for further feature extraction; finally, the output of the convolutional layer is input into the classification module (Classbox) for building category recognition, and the building outline vector and the original image overlay are output as the final building extraction result.

[0081] The dual attention module DAM structure is as follows Figure 3 As shown in the figure, PAM is the position attention branch, CAM is the channel attention branch, and the specific operation processes of the two branches include convolution, reshaping, Softmax, transposition and other operations; PAM and CAM use spatial information and channel information respectively to improve feature representation capabilities.

[0082] In step S5, the buildings in the study area are classified by building functions by setting classification rules to obtain building function classification data in the study area. The specific process is as follows:

[0083] Referring to the POI classification rules, the system of urban building function classification with 3 major categories and 12 minor categories was established through induction and merging. According to the urban building function classification system, the POI data categories were reclassified into categories that can be used to mark building functions, and different weights were set for each type of POI points based on expert scores. The building outline data of the study area were superimposed and analyzed with the reclassified POI data. According to the number, type and weight of different types of POI points in the building outline, the building function classification of the building outline data of the study area was carried out based on the word frequency-inverse text frequency index method.

[0084] The term frequency-inverse text frequency index method refers to the TF-IDF index method, and the formula of the TF-IDF index method is:

[0085]

[0086]

[0087] TF-IDF i,j =TF i,j ×IDF i

[0088] Where i represents a word, j represents a document; n i,j represents the frequency of occurrence of i in document j; ∑ k n k,j represents the total number of occurrences of all words in document j, where k represents the index of all terms in document j; TF i,j represents the frequency of term i in document j; IDF i The inverse document frequency D of term i represents the total number of documents; {j:t i ∈d j} represents the total number of documents containing word i, that is, n i,j ≠0 number of documents; TF-IDF i,j Represents the TF-IDF value of term i in document j.

[0089] The classification rules set in step S3 and step S5 are to perform functional classification using the word frequency-inverse text frequency index method. The specific classification steps are as follows:

[0090] Collect POI data in the study area, and each POI data includes the type and location information of the POI; divide the study area into several plot units or building outlines and record them as spatial units; according to the division of spatial units, record the number of each type of POI in each spatial unit, and for each POI type in each spatial unit, calculate the word frequency TF according to the formula i,j And the corresponding inverse document frequency IDF i , thus obtaining TF-IDF i,j ; According to the expert scores, different weights are set for each type of POI, and the corresponding TF-IDF values ​​are weighted; for each spatial unit, the weighted TF-IDF values ​​of all POI types are aggregated to obtain a comprehensive score; according to the comprehensive score, spatial units with equal comprehensive scores are classified into one category, and traversed to all spatial units to achieve functional classification.

[0091] The above-mentioned spatial unit represents a plot unit or a building outline. The classification rules set in step S3 and step S5 are the same, but the applied data are different. In step S3, the classification rules are used to classify the current land plot unit data to obtain the land classification data of the study area; in step S5, the classification rules are used to classify the buildings in the study area according to their building functions to obtain the building function classification data of the study area.

[0092] Step S6 includes:

[0093] S6-1. Based on the high-resolution stereo image pair obtained in S1-2, the digital surface model DSM data of the study area is extracted by using the image dense matching algorithm;

[0094] S6-2. Filter the digital surface model DSM data of the study area obtained in step S6-1 to obtain the digital elevation model DEM data of the study area;

[0095] S6-3. Differentiate the extracted digital surface model DSM data with the digital elevation model DEM data to obtain normalized digital surface model nDSM data representing the height of various surface objects in the study area;

[0096] S6-4. Fit the building outline vector data of the study area obtained in step S4 with the normalized digital surface model nDSM data representing the height of surface objects in the above-mentioned study area, and perform surface analysis. According to the height information indicated by the pixels located in the building outline vector data in the normalized digital surface model nDSM data, the building height information in the study area is obtained.

[0097] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0098] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for rapidly constructing an urban digital base based on multi-source remote sensing data, characterized in that: The method comprises the following steps: Step S1. Acquire multi-source remote sensing data and perform preprocessing, wherein the multi-source remote sensing data includes high-resolution orthophoto remote sensing images, high-resolution stereo image pairs, OSM road network data, and POI data of the study area; Step S2. Overlay the OSM road network data with the study area, perform line cutting and surface operations on the study area according to the OSM road network data, generate plot units, and obtain the current land use plot unit data of the study area; Step S3. Overlay and analyze the POI data of the study area in step S1 and the current land use unit data of the study area in step S2, classify the current land use unit data by setting classification rules, and obtain the land use classification data of the study area; Step S4. Using the deep learning-based instance segmentation model to perform building extraction on the high-resolution orthophoto remote sensing image preprocessed in step S1, obtain a building extraction result set corresponding to the image slice set; analyze the building extraction result set to obtain building outline vector data of the study area; Step S5. Overlay analysis is performed on the POI data of the study area in step S1 and the building outline vector data of the study area in step S4, and building function classification is performed on the buildings in the study area by setting classification rules to obtain building function classification data of the study area; Step S6. Based on the high-resolution stereo image pair in step S1, extract the digital surface model, filter and generate the digital elevation model, obtain the normalized digital surface model by difference, and analyze it in combination with the building outline vector data in step S4, so as to obtain the building height information in the study area; Step S7. Based on the building outline vector data and building height information of the study area, a three-dimensional model is reconstructed to obtain the building three-dimensional model data of the study area; combined with high-resolution orthophoto remote sensing images, OSM road network data, POI data, land use classification data of the study area, building function classification data of the study area and building three-dimensional model data of the study area, registration and integration are performed to obtain the urban digital base data set of the study area.

2. The method for rapidly constructing a city digital base based on multi-source remote sensing data according to claim 1, characterized in that: The step S1 comprises: S1-1. Acquire high-resolution remote sensing images and perform preprocessing to obtain high-resolution orthophoto remote sensing images after preprocessing; S1-2. Obtain a high-resolution stereo image pair from a high-resolution stereo satellite and perform preprocessing to obtain a high-resolution stereo image pair after preprocessing of the study area; S1-3. Obtain OSM road network data and preprocess it to obtain OSM road network data of the study area after preprocessing; wherein OSM road network data refers to road network data on the Open Street View map platform; S1-4, obtain POI data and preprocess it to obtain the preprocessed POI data of the study area; wherein POI data refers to point of interest data, which abstracts urban functions in life into geographic information data in the form of spatial points.

3. The method for rapidly constructing a city digital base based on multi-source remote sensing data according to claim 1, characterized in that: In step S3, the existing land parcel unit data is classified by setting classification rules to obtain the land classification data of the study area. The specific process is as follows: According to the urban construction land classification system, the POI data categories are reclassified into categories that can be used to mark the functions of plot units, and different weights are set for each type of POI data point based on expert scores. The current land plot unit data of the study area are superimposed and analyzed with the reclassified POI data. According to the number, type and weight of different types of POI data points in the plot units, the functional classification of the current land plot units in the study area is carried out based on the word frequency-inverse text frequency index method.

4. The method for rapidly constructing a city digital base based on multi-source remote sensing data according to claim 1, characterized in that: The step S4 comprises: S4-1. Slice the preprocessed high-resolution orthophoto remote sensing image according to a fixed size to obtain an image slice data set; S4-2. Input the image slice data set into the instance segmentation model based on deep learning to extract the building, and obtain the building extraction result set corresponding to the image slice set; S4-3. Vectorize the building extraction result set to obtain the building top outline vector data of the study area; S4-4. Regularize the graph of the building top outline vector data in the study area to obtain regularized building top outline vector data in the study area; S4-5, correcting the regularized building top outline vector data of the study area, correcting the building top outline to the building bottom footprint, and obtaining the building outline vector data of the study area.

5. The method for rapidly constructing a city digital base based on multi-source remote sensing data according to claim 4, characterized in that: The deep learning-based instance segmentation model in S4-2 is based on the Mask R-CNN model and is improved based on the characteristics of remote sensing images to obtain an improved instance segmentation model RS-Mask R-CNN; the specific improvement process is: simultaneously inputting the RGB three-channel data and elevation channel data of the image slice in the model input stage, and adding a dual attention module after the feature extraction backbone network, wherein the dual attention module includes a position attention branch and a channel attention branch.

6. The method for rapidly constructing a city digital base based on multi-source remote sensing data according to claim 4, characterized in that: In step S5, the buildings in the study area are classified by building functions by setting classification rules to obtain building function classification data in the study area. The specific process is as follows: Referring to the POI classification rules, the system of urban building function classification with 3 major categories and 12 minor categories was established through induction and merging. According to the urban building function classification system, the POI data categories were reclassified into categories that can be used to mark building functions, and different weights were set for each type of POI points based on expert scores. The building outline data of the study area were superimposed and analyzed with the reclassified POI data. According to the number, type and weight of different types of POI points in the building outline, the building function classification of the building outline data of the study area was carried out based on the word frequency-inverse text frequency index method.

7. The method for rapidly constructing a city digital base based on multi-source remote sensing data according to claim 6, characterized in that: The word frequency-inverse text frequency index method has the following formula: TF-IDF i,j =TF i,j ×IDF i Where i represents a word, j represents a document; n i,j represents the frequency of occurrence of i in document j; ∑ k n k,j represents the total number of occurrences of all words in document j, where k represents the index of all terms in document j; TF i,j Represents the frequency of term i in document j; IDF i The inverse document frequency D of term i represents the total number of documents; {j:t i ∈d j } represents the total number of documents containing word i, that is, n i,j ≠0 number of documents; TF-IDF i,j Represents the TF-IDF value of term i in document j.

8. The method for rapidly constructing a city digital base based on multi-source remote sensing data according to claim 1, characterized in that: The classification rules set in step S3 and step S5 are to perform functional classification using the word frequency-inverse text frequency index method. The specific classification steps are as follows: Collect POI data in the study area, and each POI data includes the type and location information of the POI; divide the study area into several plot units or building outlines and record them as spatial units; according to the division of spatial units, record the number of each type of POI in each spatial unit, and for each POI type in each spatial unit, calculate the word frequency TF according to the formula i,j And the corresponding inverse document frequency IDF i , thus obtaining TF-IDF i,j ; According to the expert scores, different weights are set for each type of POI, and the corresponding TF-IDF values ​​are weighted; for each spatial unit, the weighted TF-IDF values ​​of all POI types are aggregated to obtain a comprehensive score; according to the comprehensive score, spatial units with equal comprehensive scores are classified into one category, and traversed to all spatial units to achieve functional classification.

9. The method for rapidly constructing a city digital base based on multi-source remote sensing data according to claim 1, characterized in that: The step S6 comprises: S6-1. Based on the high-resolution stereo image pair obtained in S1-2, the digital surface model DSM data of the study area is extracted by using the image dense matching algorithm; S6-2. Filter the digital surface model DSM data of the study area obtained in step S6-1 to obtain the digital elevation model DEM data of the study area; S6-3. Differentiate the extracted digital surface model DSM data with the digital elevation model DEM data to obtain normalized digital surface model nDSM data representing the height of various surface objects in the study area; S6-4. Fit the building outline vector data of the study area obtained in step S4 with the normalized digital surface model nDSM data representing the height of surface objects in the above-mentioned study area, and perform surface analysis. According to the height information indicated by the pixels located in the building outline vector data in the normalized digital surface model nDSM data, the building height information in the study area is obtained.

Images