Building remote sensing information processing system and method based on feature fusion

Abstract

The invention discloses a building remote sensing information processing system and method based on feature fusion, and relates to the technical field of information processing.The processing method comprises the following steps that image information of all buildings is collected based on satellite equipment, interest point data is recognized through third-party data, and the interest point data is obtained; performing preprocessing to generate a plurality of images and effective interest point data; performing inversion on the height of the earth surface feature to obtain digital surface model data, performing data processing and operation to obtain normalized digital surface model data, performing superposition processing in combination with a plurality of images, and extracting to obtain a multi-dimensional feature image; constructing an instance segmentation model to obtain building contour data; based on a preset roof style, roof style characteristics of each building are obtained; presetting a plurality of types of feature data, and analyzing and calculating the various types of feature data of each building; and obtaining an adjacent relationship between the buildings, constructing a distribution diagram containing the buildings, and obtaining a building function classification result.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing, and in particular to a building remote sensing information processing system and method based on feature fusion. Background Art

[0002] With the acceleration of the urbanization process in China, urban construction is changing with each passing day. More and more people are migrating to cities, and the urban structure has undergone major changes. Therefore, it has become crucial to understand its structure and dynamic characteristics; for the acquisition of urban building data, the traditional method is through manual surveys or planning approval data. However, due to the lack of historical data, as well as the time-consuming, laborious and untimely update of manual surveys, it is difficult to meet the needs of urban management departments for the dynamic acquisition and update of urban building data during the rapid urban renewal process;

[0003] With the continuous development of remote sensing technology, a large amount of multi-source remote sensing data with multiple platforms, multiple modalities and dynamic updates has been generated, providing data support for building extraction based on remote sensing data; at the same time, artificial intelligence technologies represented by deep learning have developed rapidly and have been widely applied in the field of intelligent processing of remote sensing images;

[0004] However, existing image segmentation models only utilize the visual features of high-resolution remote sensing images, and other features of multi-source remote sensing data have not been effectively utilized; existing building extraction and building function classification methods do not fully utilize the advantages of multi-source heterogeneity of remote sensing data. How to fully exploit the information of multi-source and multi-modal remote sensing data and improve the accuracy of building extraction and the accuracy of building function classification is still an urgent problem to be solved in current research. Summary of the Invention

[0005] The purpose of the present invention is to provide a building remote sensing information processing system and method based on feature fusion to solve the problems raised in the prior art.

[0006] To achieve the above object, the present invention provides the following technical solution: A building remote sensing information processing method based on feature fusion, the processing method includes the following steps:

[0007] Step S100: Collect image information of each building based on a number of satellite devices, and identify the interest point data of each building through third-party data; preprocess the collected image information and interest point data to generate a number of images and valid interest point data respectively;

[0008] Step S200: Invert the height of surface ground objects according to the number of images to obtain digital surface model data, perform data processing and arithmetic operations on the digital surface model data to obtain normalized digital surface model data; perform superposition processing on the normalized digital surface model data and the number of images, and extract a number of multi-dimensional feature images;

[0009] Step S300: Construct an instance segmentation model to extract building information from any multi-dimensional feature image, process the extracted building information to obtain the building contour data of each building; based on the preset roof styles, obtain the roof style features of each building;

[0010] Step S400: Preset several types of feature data, perform spatial association and statistical analysis on the building contour data of each building and various types of data, and perform analysis and calculation on any type of feature data of each building;

[0011] Step S500: Obtain the adjacent relationship between buildings, obtain the comprehensive features of each building based on various types of feature data, and construct a distribution map including each building; construct a graph neural network to classify each building according to its building function to obtain the building function classification result.

[0012] Further, step S100 includes the following steps:

[0013] Step S101: Obtain high-resolution stereo images from high-resolution stereo satellites, and perform rectification and cropping on the obtained stereo images through preprocessing to generate high-resolution stereo image pairs; the preprocessing of the stereo image pairs includes, but is not limited to, radiometric correction, atmospheric correction, geometric correction, cropping, etc.;

[0014] Step S102: Obtain high-resolution night light satellite images from high-resolution night light satellites, and perform rectification and cropping on the night light satellite images through preprocessing to generate night light images; the preprocessing of the night light images includes, but is not limited to, radiometric correction, atmospheric correction, geometric correction, cropping, etc.;

[0015] Step S103: Collect the geographical information of each building in the city monitored by the satellite device through a map application, set the obtained geographical information data as point-of-interest data, and generate a number of valid point-of-interest data after data correction for the point-of-interest data; among them, the point-of-interest data is geographical information data that abstracts the urban functions in life into a spatial point form. In addition to including spatial information such as the longitude and latitude coordinates and addresses of geographical entities, it also connects various attribute information, such as the name, category, and administrative division of the point of interest; the point-of-interest data can empower research such as spatio-temporal behavior, urban planning, and geographical information;

[0016] Since there are various problems in the collected original point-of-interest data, such as classification errors and missing coordinate information, data correction and other preprocessing are required. Data correction specifically refers to performing classification correction on the misclassified points, deleting the points with missing information, and deleting the points of interest with unclear functional features and unable to clearly represent the building functional features.

[0017] Further, step S200 includes the following steps:

[0018] Step S201: Select a pair of stereo image pairs that form the same region from all high - resolution stereo image pairs. The stereo image pair contains two images. Extract digital surface model data from the stereo image pair through a binocular stereo matching algorithm.

[0019] Step S202: Filter the extracted digital surface model data, remove surface features and ground objects in the digital surface model data, and retain the ground height to obtain digital elevation model data; Subtract the digital elevation model data from the digital surface model data to obtain normalized digital surface model data.

[0020] Step S203: Obtain each ortho - multispectral image, resample by referring to the ortho - multispectral image, and correct the normalized digital surface model data; Superimpose the corrected normalized digital surface model data with the ortho - multispectral image to obtain a multi - dimensional feature image containing multi - spectral channels and height channels.

[0021] Further, step S300 includes the following steps:

[0022] Step S301: Preset a fixed image cutting size, perform image segmentation on any multi - dimensional feature image to obtain an image slice data set of the multi - dimensional feature image; Input the image slice data set into an instance segmentation model to extract the buildings contained in each image slice data, and obtain a building extraction result set corresponding to the image slice data set.

[0023] In terms of the instance segmentation model, by constructing an instance segmentation model, in feature extraction, combine the convolutional neural network and the attention module of the Transformer model, comprehensively utilize the local feature extraction ability of the convolutional neural network and the global spatial context modeling ability of the Transformer model to improve the image feature extraction ability and effectively improve the accuracy of building contour extraction.

[0024] Step S302: Merge the building extraction result set to obtain the building extraction result of the region presented in the multi - dimensional feature image; Vectorize the building extraction result to obtain building contour data, and regularize the building graphics contained in the building contour data to obtain regularized building contour data.

[0025] Step S303: Extract various roof styles from the building contour data. According to the formula:

[0026] ;

[0027] where, Confidencep→i is the confidence that the p-th building in the area belongs to the i-th roof style, C i () is an arrangement and summarization function that summarizes the confidence that the p-th building belongs to each roof style; obtaining the roof style feature F of the p-th building in the area p .

[0028] Furthermore, the step S400 includes the following steps:

[0029] Step S401: Extract the building contour image of any building from the building contour data, obtain the building contour image of the p-th building, and set the pixel value of the j-th pixel point in the building contour image of the p-th building as DN j , according to the formula:

[0030] ;

[0031] Among them, DESC() is a descending sorting function that sorts each pixel in the building contour image of the p-th building in descending order according to the pixel value. n is a positive number and n ∈ (0, 100), which is the percentage number intercepted after sorting all pixels. Mean() is an average value function that calculates the average value of the intercepted first n% of pixel values; calculating to obtain the building height feature H of the p-th building p ;

[0032] Step S402: Preset several types of valid interest point data and several types of building function categories, construct the mapping relationship between valid interest points and building functions, according to the formula:

[0033] ;

[0034] Among them, f is a mapping function, type poi is the category of valid interest point data, type building is the category of building functions; for any type of valid interest point data, match the corresponding type of building function, and set corresponding weights for each type of valid interest point data; reclassify the valid interest point categories into categories that can be used to mark building functions; for any category of valid interest point data, there is a unique corresponding category of building functions

[0035] Step S403: Obtain the regularized building contour data of the p-th building, extract the building contour of the building from the regularized building contour data, buffer the building contour of the p-th building with a preset buffer distance to obtain buffered building contour data; spatially connect the buffered building contour data of the p-th building with various types of valid interest point data to obtain various types of valid interest point data associated with the building contour of the p-th building; set the number of types of valid interest point data associated with the building contour of the p-th building as mp , where the number of the k-th type of valid point-of-interest data is (u k ) p , according to the formula:

[0036] ;

[0037] where, (w k ) p is the weight value of the k-th type of valid point-of-interest data, C k () is the permutation aggregation function, which aggregates the weight ratios of various types of valid point-of-interest data of the p-th building; the social function feature U p of the p-th building is calculated;

[0038] Step S404: Obtain the night light image of the p-th building, perform spatial association analysis and area statistics analysis with the regularized building contour data, obtain the building contour image included in the night light image of the p-th building within the building contour, and obtain that the number of pixels in the building contour image is r. Set the pixel value of the j1-th pixel among them as DN j1 , according to the formula:

[0039] ;

[0040] where, S p is the contour area of the p-th building; the night light index L p of the p-th building is calculated; the night light index is used to characterize the human activity characteristics of the building.

[0041] Further, step S500 includes the following steps:

[0042] Step S501: Arbitrarily select the p-th building, and respectively extract the roof style feature F p , building height feature H p , social function feature U p and night light index L p of the p-th building. Perform a concatenation operation on the extracted various features. According to the formula:

[0043] ;

[0044] where, CONTACT() is the concatenation operation; the comprehensive feature Z p of the p-th building is obtained; the concatenation can help enhance the distinguishability between buildings and is conducive to the subsequent building classification;

[0045] Step S502: Obtain the centroid position of any building, construct a constrained Delaunay triangulation with the center positions of each building as nodes, and calculate the correlation between each building through the Delaunay triangulation; if there are two buildings that touch the same triangular edge, establish a proximity relationship between the nodes corresponding to the two buildings, and generate an edge E between the two nodes.

[0046] Step S503: Use the nodes corresponding to each building as graph nodes V, and the edge E between two nodes to construct an undirected graph G(V, E). Among them, set the feature of any graph node V to the comprehensive feature Z of the corresponding building.

[0047] Step S504: Construct a graph attention network model to classify the building functions of each building in the undirected graph to obtain the building function classification result; use the attention mechanism to automatically learn and optimize the connection relationship between nodes, and improve the calculation efficiency by parallel computing the attention values between node pairs; the graph attention mechanism assigns different weights to each neighbor node, enabling the model to better process nodes with different weights.

[0048] To better implement the above method, a building remote sensing information processing system is also proposed. The processing system includes a data acquisition and processing module, a data image analysis module, a building information extraction module, a feature data processing module, and a building function classification module.

[0049] The data acquisition and processing module is used to collect the image information and building data of each building through a high-resolution stereo satellite, preprocess the collected image information and building data, and generate several images and valid data.

[0050] The data image analysis module is used to invert the height of the surface features according to the several images to obtain digital surface model data, perform data processing and arithmetic operations on the digital surface model data to obtain normalized digital surface model data; perform superposition processing on the normalized digital surface model data and the several images to extract several multi-dimensional feature images.

[0051] The building information extraction module is used to construct an instance segmentation model to extract building information from any multi-dimensional feature image, process the extracted building information to obtain the building contour data of each building; based on the preset roof style, obtain the roof style features of each building.

[0052] The feature data processing module is used to preset several types of feature data, perform spatial association and statistical analysis on the building contour data of each building and various types of data, and perform analysis and calculation on any type of feature data of each building.

[0053] The building function classification module is used to obtain the adjacent relationship between buildings, obtain the comprehensive features of each building based on various feature data, and construct a distribution map including each building; construct a graph neural network to classify each building according to the building function to obtain the building function classification result.

[0054] Furthermore, the data image analysis module includes a model data generation unit and a feature image extraction unit;

[0055] The model data generation unit is used to invert the height of the surface features from the several images to obtain digital surface model data, perform data processing and arithmetic operations on the digital surface model data to obtain normalized digital surface model data; the feature image extraction unit is used to perform superposition processing on the normalized digital surface model data and the several images, and extract several multi-dimensional feature images.

[0056] Furthermore, the building information extraction module includes a contour data processing unit and a roof style analysis unit;

[0057] The contour data processing unit is used to construct an instance segmentation model to extract building information from any multi-dimensional feature image, and process the extracted building information to obtain the building contour data of each building; the roof style analysis unit is used to obtain the roof style features of each building based on the preset roof styles.

[0058] Furthermore, the building function classification module includes a building distribution analysis unit and a classification result generation unit;

[0059] The building distribution analysis unit is used to obtain the adjacent relationship between buildings, obtain the comprehensive features of each building based on various feature data, and construct a distribution map including each building; the classification result generation unit is used to construct a graph neural network to classify each building according to the building function to obtain the building function classification result.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] 1. The present invention comprehensively utilizes multi-source and multi-modal remote sensing data, makes full use of the advantages of multi-source and heterogeneous remote sensing data, maximally excavates multi-source remote sensing information, and extracts multi-dimensional features such as visual features, morphological features, height features, socio-economic features, and human activity features of buildings, which can effectively improve the accuracy of building contour extraction and the accuracy of building function classification;

[0062] 2. The present invention can efficiently and automatically extract urban building information, provide data support and technical guarantee for urban planning and management, has the advantages of low production cost, high efficiency, and can realize periodic update, and has broad application prospects. Description of the Drawings

[0063] Figure 1 Schematic diagram of steps for a method of processing building remote sensing information based on feature fusion;

[0064] Figure 2 Schematic diagram of the structure of a system for processing building remote sensing information based on feature fusion;

[0065] Figure 3 Schematic diagram of the process for inverting the height of surface features;

[0066] Figure 4 Schematic diagram of the process for extracting building information;

[0067] Figure 5 Schematic diagram of the structure of a graph attention network model for building function classification;

[0068] Figure 6 Technical roadmap of a system and method for processing building remote sensing information based on feature fusion. Detailed implementation manners

[0069] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0070] Embodiment: As Figures 1 to 6 shown, the present invention provides a method for processing building remote sensing information based on feature fusion. The processing method includes the following steps:

[0071] Step S100: Collect image information of each building based on a number of satellite devices, and identify the interest point data of each building through third-party data; preprocess the collected image information and interest point data to generate a number of images and effective interest point data respectively;

[0072] Among them, step S100 includes the following steps:

[0073] Step S101: Obtain high-resolution stereo images from high-resolution stereo satellites, and correct and crop the obtained stereo images through preprocessing to generate high-resolution stereo image pairs;

[0074] Step S102: Obtain high-resolution night light satellite images from high-resolution night light satellites, and correct and crop the night light satellite images through preprocessing to generate night light images;

[0075] Step S103: Collect the geographical information of each building in the city monitored by the satellite device through a map application, set the obtained geographical information data as point-of-interest data, and generate a number of valid point-of-interest data after data correction for the point-of-interest data.

[0076] Step S200: Invert the height of surface features from the several images to obtain digital surface model data, perform data processing and arithmetic operations on the digital surface model data to obtain normalized digital surface model data; perform overlay processing on the normalized digital surface model data and the several images to extract several multi-dimensional feature images;

[0077] Among them, step S200 includes the following steps:

[0078] Step S201: Select a pair of stereo image pairs that form the same area from all high-resolution stereo image pairs. The stereo image pair contains two images, and extract digital surface model data from the stereo image pair through a binocular stereo matching algorithm;

[0079] Step S202: Perform filtering processing on the extracted digital surface model data, remove surface features in the digital surface model data and retain the ground height to obtain digital elevation model data; subtract the digital surface model data from the digital elevation model data to obtain normalized digital surface model data;

[0080] Step S203: Obtain each ortho-multispectral image, resample by referring to the ortho-multispectral image to correct the normalized digital surface model data; perform overlay processing on the corrected normalized digital surface model data and the ortho-multispectral image to obtain a multi-dimensional feature image containing a multi-spectral channel and a height channel.

[0081] Step S300: Construct an instance segmentation model to extract building information from any multi-dimensional feature image, process the extracted building information to obtain building contour data for each building; based on a preset roof style, obtain roof style features for each building;

[0082] Among them, step S300 includes the following steps:

[0083] Step S301: Preset a fixed image cutting size, perform image segmentation on any multi-dimensional feature image to obtain a set of image slice data for the multi-dimensional feature image; input the set of image slice data into the instance segmentation model to extract the buildings contained in each image slice data, and obtain a set of building extraction results corresponding to the set of image slice data;

[0084] Step S302: Merge the building extraction result sets to obtain the building extraction result of the area presented in the multi-dimensional feature image; vectorize the building extraction result to obtain building contour data, and regularize the building graphics included in the building contour data to obtain regularized building contour data;

[0085] Step S303: Extract various roof styles from the building contour data. According to the formula:

[0086] ;

[0087] where Confidence p→i is the confidence that the p-th building in the area belongs to the i-th roof style, and C i () is a permutation aggregation function that aggregates the confidences that the p-th building belongs to each roof style; obtain the roof style feature F p

[0088] of the p-th building in the area.

[0089] Among them, step S400 includes the following steps:

[0090] Step S401: Extract the building contour image of any building from the building contour data, obtain the building contour image of the p-th building, and set the pixel value of the j-th pixel point in the building contour image of the p-th building as DN j , according to the formula:

[0091] ;

[0092] where DESC() is a descending sorting function that sorts each pixel in the building contour image of the p-th building in descending order of pixel value, n is a positive number and n ∈ (0, 100), is the percentage number intercepted after sorting all pixels, and Mean() is an average value function that calculates the average value of the intercepted first n% of pixel values; calculate the building height feature H p ;

[0093] Step S402: Preset several types of effective interest point data and several types of building function categories, construct a mapping relationship between the effective interest points and the building functions, and according to the formula:

[0094] ;

[0095] where f is a mapping function, type poiFor the category of valid point-of-interest data, type building For the category of building functions; for any category of valid point-of-interest data, match the corresponding category of building functions, and set corresponding weights for each category of valid point-of-interest data;

[0096] Step S403: Obtain the regularized building contour data of the p-th building, extract the building contour of the building from the regularized building contour data, buffer the building contour of the p-th building with a preset buffer distance to obtain buffered building contour data; spatially connect the buffered building contour data of the p-th building with each category of valid point-of-interest data to obtain each category of valid point-of-interest data associated with the building contour of the p-th building; set the number of types of valid point-of-interest data associated with the building contour of the p-th building to be m p , where the number of the k-th category of valid point-of-interest data is (u k ) p , according to the formula:

[0097] ;

[0098] where, (w k ) p is the weight value of the k-th category of valid point-of-interest data, C k () is the permutation aggregation function, which aggregates the weight ratios of each category of valid point-of-interest data of the p-th building; calculate the social function feature U of the p-th building p ;

[0099] Step S404: Obtain the night-time light image of the p-th building, conduct spatial association analysis and area statistics analysis with the regularized building contour data to obtain the building contour image contained within the building contour in the night-time light image of the p-th building, obtain the number of pixels in the building contour image as r, and set the pixel value of the j1-th pixel among them to be DN j1 , according to the formula:

[0100] ;

[0101] where, S p is the contour area of the p-th building; calculate the night-time light index L of the p-th building p .

[0102] Step S500: Obtain the adjacent relationship between buildings, obtain the comprehensive features of each building based on various feature data, and construct a distribution map containing each building; construct a graph neural network to classify each building according to the building function to obtain the building function classification result;

[0103] Among them, Step S500 includes the following steps:

[0104] Step S501: Arbitrarily select the p-th building, and respectively extract the roof style feature F p , building height feature H p , social function feature U p and night light index L p . Perform a concatenation operation on the extracted various features. According to the formula:

[0105] ;

[0106] where CONTACT() is the concatenation operation; obtain the comprehensive feature Z of the p-th building p ;

[0107] Step S502: Obtain the centroid position of any building, construct a constrained Delaunay triangulation with the center positions of each building as nodes, and calculate the correlation between each building through the Delaunay triangulation; if there are two buildings that touch the same triangular edge, establish a proximity relationship between the nodes corresponding to the two buildings, and generate an edge E between the two nodes;

[0108] Step S503: Use the nodes corresponding to each building as graph nodes V, and the edge E between two nodes to construct an undirected graph G(V, E), where the feature of any graph node V is set as the comprehensive feature Z of the corresponding building;

[0109] Step S504: Construct a graph attention network model, classify the building functions of each building in the undirected graph, and obtain the building function classification result;

[0110] Example: As Figure 5 shown, where is the input feature vector; is the output high-dimensional feature vector; is the importance of the feature relationship between nodes; is the weight matrix; contcat is the connection operation; avg is the operation of taking the mean; softmax is the normalized exponential function; i ∈ 1, 2, 3…N, N is the number of graph nodes, which refers to the number of building nodes in the present invention; F is the number of features, and in the present invention, the number of features is 4.

[0111] A building remote sensing information processing system, the processing system includes a data acquisition and processing module, a data image analysis module, a building information extraction module, a feature data processing module, and a building function classification module;

[0112] The data acquisition and processing module is used to collect the image information and building data of each building through a high-resolution stereo satellite, preprocess the collected image information and building data, and generate a number of images and valid data;

[0113] A data image analysis module, which is used to invert the height of surface features based on the several images to obtain digital surface model data, perform data processing and arithmetic operations on the digital surface model data to obtain normalized digital surface model data; perform superposition processing on the normalized digital surface model data and the several images, and extract several multi-dimensional feature images;

[0114] A building information extraction module, which is used to construct an instance segmentation model to extract building information from any multi-dimensional feature image, process the extracted building information to obtain the building contour data of each building; based on the preset roof styles, obtain the roof style features of each building;

[0115] A feature data processing module, which is used to preset several types of feature data, perform spatial association and statistical analysis on the building contour data of each building and various types of data, and perform analysis and calculation on any type of feature data of each building;

[0116] A building function classification module, which is used to obtain the adjacent relationship between buildings, obtain the comprehensive features of each building based on various types of feature data, and construct a distribution map including each building; construct a graph neural network to classify each building according to its building function to obtain the building function classification result.

[0117] Among them, the data image analysis module includes a model data generation unit and a feature image extraction unit;

[0118] The model data generation unit is used to invert the height of surface features based on the several images to obtain digital surface model data, perform data processing and arithmetic operations on the digital surface model data to obtain normalized digital surface model data; the feature image extraction unit is used to perform superposition processing on the normalized digital surface model data and the several images, and extract several multi-dimensional feature images.

[0119] Among them, the building information extraction module includes a contour data processing unit and a roof style analysis unit;

[0120] The contour data processing unit is used to construct an instance segmentation model to extract building information from any multi-dimensional feature image, process the extracted building information to obtain the building contour data of each building; the roof style analysis unit is used to based on the preset roof styles, obtain the roof style features of each building.

[0121] Among them, the building function classification module includes a building distribution analysis unit and a classification result generation unit;

[0122] The building distribution analysis unit is used to obtain the adjacent relationships between buildings, obtain the comprehensive features of each building based on various types of feature data, and construct a distribution map including each building; the classification result generation unit is used to construct a graph neural network to classify each building according to the building function and obtain the building function classification result.

[0123] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to embrace all changes falling within the meaning and scope of the equivalent elements of the claims in the present invention. Any reference signs in the claims should not be construed as limiting the claimed invention.

Claims

1. A method for processing building remote sensing information based on feature fusion, characterized in that: The processing method comprises the following steps: Step S100: collecting image information of each building based on a number of satellite devices, and identifying the point of interest data of each building through third-party data; pre-processing the collected image information and point of interest data to generate a number of images and valid point of interest data respectively; Step S200: inverting the height of the surface object according to the plurality of images to obtain digital surface model data, performing data processing and calculation operations on the digital surface model data to obtain normalized digital surface model data; superimposing the normalized digital surface model data and the plurality of images to extract a plurality of multi-dimensional feature images; Step S300: construct an instance segmentation model to extract building information from any multi-dimensional feature image, process the extracted building information to obtain building outline data of each building; based on the preset roof style, obtain the roof style features of each building; Step S400: Preset several types of feature data, spatially associate and statistically analyze the building outline data of each building with various types of data, and analyze and calculate any type of feature data of each building; Step S500: Obtain the neighbor relationship between each building, obtain the comprehensive characteristics of each building based on various feature data, and construct a distribution map containing each building; construct a graph neural network to classify each building according to its function to obtain a building function classification result.

2. The method for processing building remote sensing information based on feature fusion according to claim 1, characterized in that: The step S100 includes the following steps: Step S101: obtaining high-resolution stereo images from high-resolution stereo satellites, and performing correction and cropping on the obtained stereo images through preprocessing to generate high-resolution stereo image pairs; Step S102: obtaining a high-resolution night light satellite image from a high-resolution night light satellite, and performing correction and cropping on the night light satellite image through preprocessing to generate a night light image; Step S103: The geographic information of each building in the city monitored by the satellite device is collected through a map application, the obtained geographic information data is set as point of interest data, and the point of interest data is corrected to generate a number of valid point of interest data.

3. The method for processing building remote sensing information based on feature fusion according to claim 2 is characterized in that: The step S200 includes the following steps: Step S201: selecting a pair of stereo image pairs constituting the same area from all high-resolution stereo image pairs, wherein the stereo image pair includes two images, and extracting digital surface model data from the stereo image pair by using a binocular stereo matching algorithm; Step S202: filtering the extracted digital surface model data, removing the surface objects in the digital surface model data and retaining the ground height to obtain digital elevation model data; subtracting the digital surface model data from the digital elevation model data to obtain normalized digital surface model data; Step S203: Acquire each orthophoto multispectral image, and correct the normalized digital surface model data by resampling with reference to the orthophoto multispectral image; superimpose the corrected normalized digital surface model data with the orthophoto multispectral image to obtain a multidimensional feature image including a multispectral channel and a height channel.

4. The method for processing building remote sensing information based on feature fusion according to claim 3 is characterized in that: The step S300 includes the following steps: Step S301: Preset a fixed image cutting size, perform image segmentation on any multi-dimensional feature image, and obtain an image slice data set of the multi-dimensional feature image; input the image slice data set into the instance segmentation model to extract the buildings contained in each image slice data, and obtain a building extraction result set corresponding to the image slice data set; Step S302: merging the building extraction result sets to obtain building extraction results of the area presented in the multi-dimensional feature image; vectorizing the building extraction results to obtain building outline data, and regularizing the building graphics contained in the building outline data to obtain regularized building outline data; Step S303: extract various roof styles from the building outline data according to the formula: ； Among them, Confidence p→i is the confidence that the p-th building in the region belongs to the i-th type of roof style, C i () is a permutation summary function, which summarizes the confidence that the p-th building belongs to each type of roof style; the roof style feature F of the p-th building in the area is obtained p .

5. The method for processing building remote sensing information based on feature fusion according to claim 4 is characterized in that: The step S400 includes the following steps: Step S401: extract the building outline image of any building from the building outline data, obtain the building outline image of the p-th building, and set the pixel value of the j-th pixel point in the building outline image of the p-th building to DN j , according to the formula: ； Among them, DESC() is a descending sorting function, which sorts each pixel in the building outline image of the p-th building in descending order according to the pixel value size, n is a positive number and n∈(0,100), which is the percentage intercepted after sorting all pixels, and Mean() is an average function, which calculates the average value of the first n% of the intercepted pixel values; the building height feature H of the p-th building is calculated. p ; Step S402: Preset several types of valid interest point data and several types of building function categories, and construct a mapping relationship between valid interest points and building functions according to the formula: ； Among them, f is the mapping function, type poi is the category of valid interest point data, type building is the category of building functions; for any type of valid POI data, a corresponding type of building function is matched, and corresponding weights are set for each type of valid POI data; Step S403: Obtain the regularized building outline data of the p-th building, extract the building outline of the building from the regularized building outline data, buffer the building outline of the p-th building by a preset buffer distance to obtain buffered building outline data; spatially connect the buffered building outline data of the p-th building with various types of valid interest point data to obtain various types of valid interest point data associated with the building outline of the p-th building; set the number of types of valid interest point data associated with the building outline of the p-th building to m p , where the number of valid interest point data of the kth category is (u k ) p , according to the formula: ； Among them, (w k ) p is the weight value of the k-th category of valid interest point data, C k () is the permutation summary function, which summarizes the weight proportions of various valid points of interest data of the p-th building; the social function characteristics U of the p-th building are calculated. p ; Step S404: Obtain the nighttime light image of the pth building, perform spatial correlation analysis and surface statistical analysis with the regularized building outline data, obtain the building outline image contained in the nighttime light image of the pth building within the building outline, obtain the number of pixels in the building outline image as r, and set the pixel value of the j1th pixel as DN j1 , according to the formula: ； Among them, S p is the outline area of ​​the pth building; calculate the night light index L of the pth building p .

6. The method for processing building remote sensing information based on feature fusion according to claim 5, characterized in that: The step S500 includes the following steps: Step S501: randomly select the p-th building and extract the roof style feature F of the p-th building p , Building height characteristics H p 、Social Function Characteristics p and night light index L p , perform serial operations on the extracted features, according to the formula: ； Among them, CONTACT() is a serial operation; the comprehensive feature Z of the pth building is obtained p ; Step S502: Obtain the centroid position of any building, construct a constrained Delaunay triangulation network with the center position of each building as a node, and calculate the correlation between each building through the Delaunay triangulation network; if there are two buildings that touch the same triangle edge, establish a neighbor relationship between the nodes corresponding to the two buildings, and generate a graph edge E between the two nodes; Step S503: Taking the nodes corresponding to each building as graph nodes V and the graph edges E between two nodes, an undirected graph G(V,E) is constructed, wherein the feature of any graph node V is set as the comprehensive feature Z of the corresponding building; Step S504: construct a graph attention network model, classify the building functions of each building in the undirected graph, and obtain the building function classification results.

7. A building remote sensing information processing system, used to execute a building remote sensing information processing method based on feature fusion according to any one of claims 1 to 6, characterized in that: The processing system includes a data acquisition processing module, a data image analysis module, a building information extraction module, a feature data processing module and a building function classification module; The data acquisition and processing module is used to collect image information and building data of each building through high-resolution stereo satellites, pre-process the collected image information and building data, and generate a number of images and valid data; The data image analysis module is used to invert the height of the surface objects according to the plurality of images to obtain digital surface model data, perform data processing and calculation operations on the digital surface model data to obtain normalized digital surface model data; perform superposition processing on the normalized digital surface model data and the plurality of images to extract a plurality of multi-dimensional feature images; The building information extraction module is used to construct an instance segmentation model to extract building information from any multi-dimensional feature image, and process the extracted building information to obtain building outline data of each building; Based on the preset roof style, the roof style characteristics of each building are obtained; The characteristic data processing module is used to preset several types of characteristic data, spatially associate and statistically analyze the building outline data of each building with various types of data, and analyze and calculate any type of characteristic data of each building; The building function classification module is used to obtain the adjacent relationship between buildings, obtain the comprehensive characteristics of each building based on various feature data, and construct a distribution map containing each building; construct a graph neural network to classify each building according to its function to obtain a building function classification result.

8. A building remote sensing information processing system according to claim 6, characterized in that: The data image analysis module includes a model data generation unit and a feature image extraction unit; The model data generating unit is used to invert the height of the surface objects according to the plurality of images to obtain digital surface model data, and perform data processing and calculation operations on the digital surface model data to obtain normalized digital surface model data; the feature image extracting unit is used to perform superposition processing on the normalized digital surface model data and the plurality of images to extract a plurality of multi-dimensional feature images.

9. A building remote sensing information processing system according to claim 6, characterized in that: The building information extraction module includes a profile data processing unit and a roof style analysis unit; The outline data processing unit is used to construct an instance segmentation model to extract building information from any multi-dimensional feature image, and process the extracted building information to obtain the building outline data of each building; the roof style analysis unit is used to obtain the roof style characteristics of each building based on a preset roof style.

10. A building remote sensing information processing system according to claim 6, characterized in that: The building function classification module includes a building distribution analysis unit and a classification result generation unit; The building distribution analysis unit is used to obtain the adjacent relationship between each building, obtain the comprehensive characteristics of each building based on various feature data, and construct a distribution map containing each building; the classification result generation unit is used to construct a graph neural network to classify each building according to its function and obtain a building function classification result.

Images