A method and system for rapid extraction of building targets from integrated surveying and mapping data

Abstract

This invention discloses a method and system for rapid extraction of building targets from integrated surveying and mapping data, relating to the technical field of building target extraction. The method includes: integrating multi-source parallel acquisition data to construct an integrated surveying and mapping database for the target area; acquiring first-dimensional integrated surveying and mapping data of the target building; extracting high-dimensional feature vectors representing the exterior materials of the target building; inputting the high-dimensional feature vectors into a spectral-texture co-enhancement model for data extraction impact analysis; if a first influence coefficient is greater than a preset influence coefficient, activating a spectral-texture co-enhancement network for data enhancement, acquiring second-dimensional integrated surveying and mapping data of the target building, and generating an extraction model for the target building. This invention solves the technical problem in existing technologies where it is difficult to effectively identify target buildings composed of different building materials based on their reflectivity, leading to low building recognition rates and poor extraction accuracy. It efficiently generates accurate extraction models, achieving the technical effect of improving the building target recognition rate and accuracy.

Description

Technical Field

[0001] This application relates to the technical field of building target extraction, specifically to a method and system for rapid extraction of building targets from integrated surveying and mapping data. Background Technology

[0002] Different building materials, such as glass, metal, and brick, have drastically different light reflection characteristics, resulting in a wide and complex range of grayscale values ​​and tones in optical images. For example, glass may produce specular reflection, causing some areas to be overly bright and lose detail; metals have unique reflectance spectral characteristics that can be easily confused with other materials, making it difficult to effectively distinguish building targets made of different materials, leading to low recognition rates. Furthermore, the textures formed by different materials vary significantly in roughness and pattern complexity, often failing to fully extract and utilize the rich information contained in surface textures to aid in identification. Existing methods struggle to accurately capture these subtle differences, thus affecting the accuracy of building target recognition.

[0003] Therefore, current technologies suffer from the problem of difficulty in effectively identifying target buildings made of different building materials based on their reflective properties, resulting in low building recognition rates and poor extraction accuracy. Summary of the Invention

[0004] This application provides a method and system for rapid extraction of building targets by integrating surveying and mapping data. It solves the technical problem in the prior art that it is difficult to effectively identify target buildings composed of different building materials based on the material reflectivity, resulting in low building recognition rate and poor extraction accuracy. It efficiently generates accurate extraction models, thereby achieving the technical effect of improving the building target recognition rate and accuracy.

[0005] This application provides a method for rapid extraction of building targets from integrated surveying and mapping data. The method includes: integrating multi-source parallel acquisition data to construct a surveying and mapping integrated database of a target area, the target area including the target building to be extracted; acquiring first target surveying and mapping integrated data of the target building in the surveying and mapping integrated database; extracting high-dimensional feature vectors representing the exterior materials of the target building based on the first target surveying and mapping integrated data, the high-dimensional feature vectors being composed of spectral feature vectors and surface texture feature vectors; inputting the high-dimensional feature vectors into a spectral-texture co-enhancement model for data extraction impact analysis, and outputting a first impact coefficient; if the first impact coefficient is greater than a preset impact coefficient, activating a spectral-texture co-enhancement network to perform data enhancement on the first target surveying and mapping integrated data, acquiring second target surveying and mapping integrated data, and generating an extraction model of the target building based on the second target surveying and mapping integrated data.

[0006] In a possible implementation, the rapid building target extraction method based on the integrated mapping data further performs the following processing: the spectral-texture co-enhancement model determines the exterior material type of the target building based on the high-dimensional feature vector; real-time illumination conditions of the multi-source parallel acquisition data are obtained, including real-time illumination intensity and real-time illumination angle; a mapping function relationship is established between the exterior material type sample and the illumination condition sample to characterize the degree of influence on data extraction; the exterior material type and the real-time illumination conditions are analyzed based on the mapping function relationship, and a first influence coefficient is output.

[0007] In a possible implementation, the method for rapid extraction of building targets from integrated surveying and mapping data further performs the following processing: activating a spectral-texture collaborative enhancement network to identify boundary defects in the first target integrated surveying and mapping data and outputting defect integrated surveying and mapping data; obtaining matching lighting conditions with an influence coefficient less than a preset influence coefficient based on the exterior material type; and using the matching integrated surveying and mapping sample data corresponding to the matching lighting conditions to perform defect data enhancement on the defect integrated surveying and mapping data to obtain second target integrated surveying and mapping data.

[0008] In a possible implementation, the method for rapid extraction of building targets from integrated mapping data further performs the following processing: acquiring high-dimensional feature vector samples, which include spectral feature vector samples and surface texture feature samples of different building materials under different lighting conditions. The spectral feature vector samples include band reflectance of spectral images, and the surface texture feature vector samples include contrast, energy, and correlation parameters calculated by GLCM; acquiring label sample data, which includes material categories and corresponding labeled integrated mapping sample data; and training a model based on the high-dimensional feature vector samples and the label sample data to obtain a converged spectral-texture co-enhancement network.

[0009] In a possible implementation, the method for rapid extraction of building targets from integrated surveying data further performs the following processing: initializing a generative adversarial network, including initializing a dual-branch generator and a discriminator; wherein, the dual-branch generator includes a spectral branch and a texture branch, the spectral branch simulates the band reflectivity changes of different building materials under different lighting conditions through a fully connected layer, and the texture branch simulates the surface texture changes of different building materials under different lighting conditions through a deconvolution layer; after inputting the high-dimensional feature vector samples into the dual-branch generator, a generative high-dimensional feature vector is output; the discriminator receives the generative high-dimensional feature vector output by the dual-branch generator and performs adversarial loss feedback training with the high-dimensional feature vector samples to obtain a trained generative adversarial network; and the trained generative adversarial network is invoked to supplement the high-dimensional feature vector samples.

[0010] In a possible implementation, the method for rapid extraction of building targets from integrated surveying and mapping data further performs the following processing: if there are multiple target buildings to be extracted, acquire multiple integrated surveying and mapping data corresponding to the multiple target buildings; extract multiple high-dimensional feature vectors representing the facade materials of the multiple target buildings based on the multiple integrated surveying and mapping data; input the multiple high-dimensional feature vectors into the spectral-texture collaborative enhancement model to obtain multiple extraction models corresponding to the multiple target buildings.

[0011] In a possible implementation, the method for rapid extraction of building targets from integrated surveying and mapping data further performs the following processing: the spectral-texture collaborative enhancement network further includes a dynamic weight adjustment module, which dynamically adjusts the fusion weights of spectral features and texture features according to the exterior material type and the real-time lighting conditions; and performs defect data weight enhancement on the defect surveying and mapping integrated data according to the fusion weights to obtain the second target surveying and mapping integrated data.

[0012] This application also provides a rapid building target extraction system based on integrated surveying and mapping data. The system includes: a surveying and mapping integrated database construction module, used to integrate multi-source parallel acquisition data to construct a surveying and mapping integrated database of a target area, the target area including the target building to be extracted; a first target surveying and mapping integrated data acquisition module, used to acquire first target surveying and mapping integrated data of the target building in the surveying and mapping integrated database; an exterior facade feature vector extraction module, used to extract high-dimensional feature vectors representing the exterior facade materials of the target building based on the first target surveying and mapping integrated data, the high-dimensional feature vectors being composed of spectral feature vectors and surface texture feature vectors; a data extraction impact analysis module, used to input the high-dimensional feature vectors into a spectral-texture co-enhancement model for data extraction impact analysis, and output a first impact coefficient; and a target building extraction model generation module, used to activate a spectral-texture co-enhancement network to perform data enhancement on the first target surveying and mapping integrated data if the first impact coefficient is greater than a preset impact coefficient, to acquire second target surveying and mapping integrated data, and to generate an extraction model of the target building based on the second target surveying and mapping integrated data.

[0013] This application proposes a method and system for rapid extraction of building targets using integrated surveying and mapping data. The method integrates multi-source parallel acquisition data to construct an integrated surveying and mapping database for the target area; acquires integrated surveying and mapping data for the first target; extracts high-dimensional feature vectors representing the exterior materials of the target building; inputs these high-dimensional feature vectors into a spectral-texture co-enhancement model for data extraction impact analysis; if the first impact coefficient is greater than a preset impact coefficient, activates the spectral-texture co-enhancement network for data enhancement, acquires integrated surveying and mapping data for the second target, and generates an extraction model for the target building. This solves the technical problem in existing technologies where it is difficult to effectively identify target buildings composed of different building materials based on their reflectivity, leading to low building recognition rates and poor extraction accuracy. It efficiently generates accurate extraction models, achieving the technical effect of improving the building target recognition rate and accuracy. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1 A schematic diagram of the method for rapid extraction of building targets from integrated surveying data provided in this application embodiment.

[0016] Figure 2 A schematic diagram of the structure of a rapid building target extraction system integrating surveying data provided in this application embodiment.

[0017] Figure labeling: 10, Surveying and Mapping Integrated Database Construction Module; 20, First Target Surveying and Mapping Integrated Data Acquisition Module; 30, Facade Feature Vector Extraction Module; 40, Data Extraction Impact Analysis Module; 50, Target Building Extraction Model Generation Module. Detailed Implementation

[0018] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0021] This application provides a method for rapid extraction of building targets from integrated surveying data, such as... Figure 1 As shown, the method includes:

[0022] Step S100: Integrate multi-source parallel acquisition data to construct a mapping integrated database for the target area, wherein the target area includes the target buildings to be extracted.

[0023] Preferably, data from the target area within the same or similar time period is collected through multiple methods to reflect the status of buildings in the target area under the same time benchmark. This avoids data inconsistencies caused by changes in building status (such as construction progress, decoration changes, etc.) due to time differences. The target area includes the target buildings to be extracted. Specifically, optical remote sensing data is acquired using optical sensors mounted on satellites or aircraft to provide large-area image information of the target area, covering the overall layout, general outline, and relationship with the surrounding environment of the buildings, such as clearly showing the arrangement of different buildings. Terrestrial laser scanning (TLS) and airborne laser scanning (ALS) acquire laser scanning data of the target buildings. TLS can perform high-precision scanning of buildings at close range, accurately obtaining the three-dimensional coordinate information of the building surface and generating detailed data. Point cloud data is used to acquire complex facade structures and decorative details of buildings, accurately measuring the size, shape, and surface undulations of buildings. ALS is used to quickly acquire 3D information of terrain and buildings in large areas, including the height of buildings within the target area, the spatial relationship between terrain and buildings, etc. Oblique photogrammetry data is obtained by taking images of the target area from multiple angles, presenting multiple sides of the building, providing rich texture information for building realistic 3D models, making the reconstructed building models more realistic and accurate, and can be used to intuitively display the appearance of buildings and assist in identifying detailed features and structures of buildings. Surveyors use instruments such as total stations and levels to measure the target building on-site, obtaining on-site measurement data, including key dimensional parameters of the target building, such as the length, width, and height of the building, and the position and size of doors and windows.

[0024] Preferably, the acquired multi-source data undergoes format conversion and preprocessing to integrate it into a single database framework. For example, different formats of laser scanning point cloud data and optical image data are uniformly converted into a format suitable for database storage. Noise reduction and calibration preprocessing operations are then performed. Based on spatial reference information such as geographic coordinates, these data from different sources are precisely registered to ensure complete spatial matching. For instance, the location of a building in the optical remote sensing image is precisely aligned with the corresponding building location in the laser scanning point cloud data. A suitable database management system (DBMS), such as PostgreSQL, is selected to construct a mapping integrated database. Considering the characteristics and storage requirements of different data types, corresponding data tables and fields are set up. For example, for optical remote sensing data, a data table containing fields such as image file path, capture time, and band information can be created; for laser scanning point cloud data, a table structure storing point cloud coordinates and reflection intensity information is set up to achieve efficient storage and fast querying of multi-source parallel acquisition data.

[0025] Step S200: Obtain the first target surveying and mapping integration data of the target building in the surveying and mapping integration database.

[0026] Preferably, the first target mapping integrated data of the target building is selected from the mapping integrated database. Specifically, the target building is accurately located based on the geographic coordinates and building attributes (building name, number, use, etc.) in the mapping integrated data. Then, relevant multi-source mapping data is extracted from different data tables according to the correlation. For example, images of the target building area are extracted from the optical remote sensing data table, from which information such as building appearance and surrounding environment can be obtained; point cloud data of the building is retrieved from the laser scanning point cloud data table, which can accurately present the three-dimensional shape and structural details of the building; multi-angle images are obtained from the oblique photogrammetry data storage location to provide rich texture for comprehensive analysis of the building appearance; then, the multi-source mapping data (optical images, point cloud data, oblique photogrammetry data, etc.) are spatiotemporally aligned to generate the first target mapping integrated data of the target building. In this way, the high-dimensional feature vector representing the exterior material of the target building can be accurately extracted to accurately describe the material properties and appearance texture of the target building.

[0027] Step S300: Extract a high-dimensional feature vector representing the exterior material of the target building based on the first target mapping integrated data. The high-dimensional feature vector consists of a spectral feature vector and a surface texture feature vector.

[0028] Preferably, the spectral feature vector and surface texture feature vector of the target building are extracted from the integrated mapping data of the first target, and the spectral feature vector and surface texture feature vector are combined to form a high-dimensional feature vector characterizing the facade material. This integrates information from both spectral and textural aspects, and can more comprehensively and accurately describe the characteristics of the facade material of the target building. Specifically, spectral features refer to the reflection, absorption and transmission characteristics of the facade material of the target building to different wavelengths of light. The optical remote sensing data in the integrated mapping data of the first target contains rich spectral information, and reflectance values ​​of different bands are extracted from it, such as in the visible light band (e.g., blue light, green light, red light) and near-infrared band. The reflectance values ​​of these bands constitute the basic elements of the spectral feature vector. Then, the spectral-related feature parameters, such as spectral slope, spectral curvature, and band ratio, are calculated, further enriching the information of the spectral feature vector. This helps to distinguish different types of facade materials. For example, the reflectance of metal materials and concrete materials differs greatly in the near-infrared band. By analyzing the characteristics of these bands, it can be preliminarily determined whether the facade contains metal components.

[0029] Preferably, surface texture features reflect the roughness, texture direction, pattern, and other characteristics of the target building facade. Laser scanning data, oblique photography data, and high-resolution optical images in the first target mapping integrated data are all used to extract surface texture features. Specifically, methods based on gray-level co-occurrence matrix (GLCM) and local binary mode (LBP) are used to extract texture features from image data. For example, the gray-level co-occurrence matrix describes texture features by statistically analyzing the frequency of occurrence of pixel pairs with different gray values ​​in the image, which can reflect information such as texture roughness, contrast, and directionality. The density distribution and surface curvature of the point cloud calculated from the laser scanning point cloud data can also be used as part of the texture features. For example, decorative lines or uneven surfaces on the building facade will cause changes in point cloud density and curvature, and the corresponding texture features can be extracted by analyzing these changes.

[0030] Step S400: Input the high-dimensional feature vector into the spectral-texture co-enhancement model for data extraction and impact analysis, and output the first impact coefficient.

[0031] Preferably, the spectral-texture synergistic enhancement model uses the extracted high-dimensional feature vector containing both spectral and surface texture feature vectors as input to simulate and analyze how spectral and texture information interact and synergistically affect the extraction effect on the target building facade material data. Specifically, the spectral-texture synergistic enhancement model fuses the spectral and texture features in the input high-dimensional feature vector to explore the potential relationship between them and their combined impact on accurately extracting facade material data. For example, the model may analyze whether certain combinations of spectral and texture features can more effectively identify a specific facade material, or determine whether certain feature combinations contain redundant information; then, the input high-dimensional feature vector is combined with a set... The standard is compared to assess the degree of difference between the current high-dimensional feature vector and the ideal situation, and then the impact on data extraction is analyzed. Finally, the first influence coefficient is output to represent the degree of influence of the current high-dimensional feature vector on the data extraction of the target building facade material. Generally speaking, if the first influence coefficient is close to 1, it means that the high-dimensional feature vector is well matched with the ideal feature pattern, which has a positive promoting effect on data extraction and can accurately reflect the characteristics of the target building facade material, which helps to improve the accuracy and reliability of data extraction. If the coefficient is much less than 1, it means that there is a large difference between the high-dimensional feature vector and the ideal pattern, and there may be some interference factors or incomplete features, resulting in blurred boundaries or textures, which has a certain negative impact on data extraction.

[0032] Furthermore, step S400 also includes step S410, whereby the spectral-texture co-enhancement model determines the exterior material type of the target building based on the high-dimensional feature vector; step S420, whereby the real-time illumination conditions of the multi-source parallel acquisition data are obtained, including real-time illumination intensity and real-time illumination angle; and step S430, whereby a mapping function relationship characterizing the degree of influence of data extraction is established between the exterior material type sample and the illumination condition sample, and the exterior material type and the real-time illumination conditions are analyzed based on the mapping function relationship to output a first influence coefficient.

[0033] Preferably, since different facade materials have unique characteristics in spectral reflectance and surface texture, the spectral-texture co-enhancement model analyzes high-dimensional feature vectors to determine the facade material type of the target building. For example, glass has specific reflectance peaks in its spectrum, and its surface texture is usually smooth; while the spectral reflectance distribution and texture characteristics of stone are different from those of glass. The spectral-texture co-enhancement model determines the facade material type of the target building, such as glass, stone, metal, and concrete, by comparing high-dimensional feature vectors with feature templates of various facade materials. During the multi-source parallel data acquisition process, real-time lighting conditions are recorded simultaneously, including real-time light intensity and real-time light angle. Light intensity refers to the energy of visible light received per unit area, and light angle refers to the incident angle of light relative to the facade of the target building. Lighting conditions have a significant impact on the spectral reflectance characteristics of building facade materials. For example, under strong direct sunlight, the reflectance of the material may increase, causing changes in spectral characteristics. Different light angles will also produce different shadow and highlight effects on the material surface, affecting the performance of texture features.

[0034] Preferably, a large number of exterior facade material type samples (such as glass, stone, metal, concrete, etc.) and corresponding lighting condition samples are collected through field measurements, remote sensing imagery, and laboratory simulations. Each sample data is labeled to clarify its exterior facade material type and corresponding lighting conditions (real-time light intensity and real-time light angle), and the degree of their influence on data extraction is analyzed. A mapping function relationship characterizing the degree of influence on data extraction is established, describing the influence of different exterior facade material types on the accuracy of target building data extraction under different lighting conditions. Specifically, features related to exterior facade material type and lighting conditions are extracted; for exterior facade materials, spectral feature vectors and surface features are extracted. Texture feature vectors are generated; for lighting conditions, real-time light intensity and angle are extracted, and the angle between the lighting direction and the building orientation, as well as the rate of change of light intensity over different time periods, can also be calculated; features with significant correlation to the degree of data extraction are screened through correlation analysis and principal component analysis, redundant features are removed, and data dimensionality is reduced; then, a suitable machine learning model is selected to establish a mapping function relationship, such as training the model (e.g., linear regression model, decision tree model, support vector machine, neural network, etc.) using sample data, so that the model can learn the inherent mapping relationship between facade material type samples and lighting condition samples, in order to minimize the error between the prediction results and the actual annotations. Finally, after evaluation and verification, a mapping function relationship representing the degree of influence of facade material type samples and lighting condition samples on data extraction is obtained, which can accurately output the corresponding first influence coefficient based on the input facade material type and real-time lighting conditions, describing the degree of influence on data extraction under that condition.

[0035] Step S500: If the first influence coefficient is greater than the preset influence coefficient, activate the spectral-texture collaborative enhancement network to perform data enhancement on the first target mapping integrated data, obtain the second target mapping integrated data, and generate the extraction model of the target building based on the second target mapping integrated data.

[0036] Preferably, the preset influence coefficient is a threshold pre-set based on experience and the requirements for data extraction accuracy. When the first influence coefficient is greater than the preset influence coefficient, the current first target mapping integrated data will have significant interference or uncertainty when used to extract target building information, which may lead to inaccurate extraction results. Therefore, data enhancement processing is required. Then, the spectral-texture collaborative enhancement network is activated to perform data enhancement and reconstruction of the first target mapping integrated data. Specifically, for the spectral features in the data, the accuracy of spectral reflectance is adjusted, and the differences between the spectral features of different materials are enhanced, etc., to enhance spectral information. For example, for spectral features that are not obvious enough... By adjusting the shape of the spectral curves and increasing the weight of specific bands, the network enhances the spectral characteristics of building materials, making them easier to identify. Regarding texture features, the network optimizes the surface texture of building facades. For example, for blurry texture images, image sharpening and edge enhancement are used to make texture details clearer. For incomplete texture information, interpolation and completion are used to repair and improve it, thereby increasing the recognizability of texture features. The spectral-texture synergistic enhancement network further enhances the synergistic effect of spectral and texture information, thereby obtaining integrated mapping data for the second target, which can more comprehensively and accurately reflect the facade features of the target building. Then, by utilizing the spectral and texture information in the second target mapping integrated data, a machine learning classification model is constructed and trained to determine the extraction model of the target building. Based on the input second target mapping integrated data, the model can accurately identify and extract relevant information of the target building, such as the building's outline, facade material type, and spatial structure. For example, a deep learning convolutional neural network (CNN) can be used, with the second target mapping integrated data as input. After training and learning the network, the network parameters can be adjusted so that the network can output accurate target building extraction results, ultimately generating the target building extraction model.

[0037] Furthermore, step S500 also includes step S510, activating the spectral-texture collaborative enhancement network to identify boundary defects in the first target mapping integration data and output defect mapping integration data; step S520, obtaining matching lighting conditions with an influence coefficient less than a preset influence coefficient according to the exterior material type; step S530, calling the matching mapping integration sample data corresponding to the matching lighting conditions to perform defect data enhancement on the defect mapping integration data and obtain the second target mapping integration data.

[0038] Preferably, after the spectral-texture co-enhancement network is activated, it performs in-depth analysis on the integrated mapping data of the first target, identifies and marks boundary defects. Among them, boundary defects in building target mapping may include blurred or incomplete edges of building facades, or unclear features at material junctions. By analyzing spectral features, the activated spectral-texture co-enhancement network can detect abnormal changes in spectral reflectance in certain areas, which may indicate the presence of boundary defects. Alternatively, it can detect breaks or discontinuities in surface texture at the boundaries by examining texture features. These defects are then marked to clarify the location and extent of the boundary defects, and finally, defect mapping integrated data containing boundary defect markings is output. Since different facade materials exhibit different spectral and textural characteristics under different lighting conditions, matching lighting conditions with an influence coefficient lower than the preset influence coefficient are obtained based on the facade material type of the target building. Specifically, when the influence coefficient corresponding to the lighting condition is less than the threshold, it indicates that the data extraction interference under this lighting condition is smaller and more conducive to accurate analysis. For example, for glass facades, under a certain specific light intensity and angle, its spectral reflectance and textural characteristics can be more clearly identified, and the corresponding influence coefficient is less than the preset value. Such lighting conditions are then identified as matching lighting conditions.

[0039] Preferably, after obtaining the matching lighting conditions, the corresponding matching mapping integrated sample data is retrieved from the database. This matching mapping integrated sample data is then applied to the defect mapping integrated data, where the defect data is enhanced. Specifically, this includes correcting and supplementing the spectral characteristics of the defect area to better match the normal spectral performance of the facade material under matching lighting conditions; and repairing and improving texture features, filling in missing parts of the texture or making the texture clearer and more coherent. Through data enhancement and reconstruction of the boundary defect data, the second target mapping integrated data is finally obtained, improving the quality and usability of the data to more accurately extract relevant information about the target building.

[0040] Further, step S510 also includes step S511, obtaining high-dimensional feature vector samples, the high-dimensional feature vector samples including spectral feature vector samples and surface texture feature samples of different building materials under different lighting conditions, the spectral feature vector samples including band reflectance of spectral images, and the surface texture feature vector samples including contrast, energy, and correlation parameters calculated by GLCM; step S512, obtaining label sample data, the label sample data including material categories and corresponding labeled mapping integrated sample data; step S513, performing model training based on the high-dimensional feature vector samples and the label sample data to obtain a spectral-texture co-enhancement network trained to convergence.

[0041] Preferably, a spectrometer or device with multispectral imaging capabilities is used to collect data on different building materials under different lighting conditions, measuring and recording the reflectance information of the building materials in multiple bands to form spectral images. Reflectance data for each band is extracted from the collected spectral images, with the reflectance value of each band serving as one dimension of a feature vector. These reflectance data collectively constitute a spectral feature vector sample. A high-resolution camera is used to capture surface images of different building materials under different lighting conditions. The acquired color images are converted to grayscale images to reduce the dimensionality of the image data while highlighting the texture information of the image. For grayscale images, the gray-level co-occurrence matrix (GLCM) is calculated to describe the spatial distribution of gray levels. The frequency of pixel pairs with specific gray values ​​that are separated by a certain distance and direction is statistically analyzed. By changing the distance and direction parameters of pixel pairs, different GLCMs can be obtained. Parameters such as contrast, energy, and correlation are calculated from the GLCM. Contrast reflects the sharpness and roughness of textures in the image, energy represents the uniformity of gray-level distribution, and correlation measures the linear correlation of local gray levels in the image. These parameters serve as elements of the surface texture feature vector, collectively forming the surface texture feature vector sample. Finally, the spectral feature vector sample and the surface texture feature sample are combined to generate high-dimensional feature vector samples of different building materials under different lighting conditions.

[0042] Preferably, the building materials used in different parts of the target building are clearly identified, such as steel, wood, glass, concrete, and stone. A comprehensive survey of the target building is then conducted, including its geometric dimensions, shape, location information, and detailed surface features of the materials. The collected survey data is then associated and labeled with the corresponding material categories to form integrated survey data with corresponding labels. Finally, the material categories and the integrated survey data with corresponding labels are combined to form tagged sample data. The prepared high-dimensional feature vector samples and label sample data are input into the spectral-texture co-enhancement network for training. During training, the network attempts to predict the corresponding material category based on the input feature vector samples and compares the prediction results with the true values ​​in the label sample data. The network performance is evaluated by calculating the error between the predicted and true values. Based on the error, the network uses optimization algorithms (such as stochastic gradient descent) to adjust the parameters (such as weights and biases) in the network, so that the error gradually decreases. When the error decreases to a certain extent, or when the error no longer decreases significantly in continuous training iterations, the model is considered to have converged, and the spectral-texture co-enhancement network is obtained. It can perform accurate spectral-texture co-analysis on new and unseen data, such as accurately identifying the characteristics of different building materials under various lighting conditions and predicting their material categories.

[0043] Furthermore, step S511 also includes step A, initializing the generative adversarial network, including initializing a dual-branch generator and a discriminator; step B, wherein the dual-branch generator includes a spectral branch and a texture branch, the spectral branch simulates the band reflectivity changes of different building materials under different lighting conditions through a fully connected layer, and the texture branch simulates the surface texture changes of different building materials under different lighting conditions through a deconvolution layer; step C, after inputting the high-dimensional feature vector samples into the dual-branch generator, a generative high-dimensional feature vector is output, and the discriminator receives the generative high-dimensional feature vector output by the dual-branch generator and performs adversarial loss feedback training with the high-dimensional feature vector samples to obtain a trained generative adversarial network; step D, the trained generative adversarial network is invoked to supplement the high-dimensional feature vector samples.

[0044] Preferably, the input of the spectral-texture co-enhancement network is connected to the generative adversarial network (GAN). This means that the data processed by the GAN (high-dimensional feature vector samples supplemented by samples) will serve as the input data for the spectral-texture co-enhancement network, forming a data transfer and interaction relationship. The GAN provides richer input data for the spectral-texture co-enhancement network, which helps improve its performance. Specifically, when initializing the GAN, its core components, the dual-branch generator and discriminator, are initialized. The dual-branch generator includes a spectral branch and a texture branch. The spectral branch uses fully connected layers to simulate the band reflectance changes of different building materials under different lighting conditions. By weighted summation of input features, it learns and captures the patterns and characteristics of the spectral reflectance of different building materials under various lighting conditions, thereby generating simulated band reflectance data. The texture branch uses deconvolution layers to simulate the surface texture changes of different building materials under different lighting conditions. The deconvolution layer has an upsampling function, which can recover high-resolution texture information from low-dimensional features. Through training, it learns the characteristics of the surface texture of different building materials under different lighting conditions, and then generates corresponding surface texture data.

[0045] Preferably, high-dimensional feature vector samples are input into a dual-branch generator. The spectral branch and texture branch process the input separately, and then the output is merged to produce a generative high-dimensional feature vector containing spectral and texture feature information simulated by the generator. Specifically, the discriminator receives the generative high-dimensional feature vector and the original high-dimensional feature vector samples, distinguishing whether the input feature vector comes from real high-dimensional feature vector samples (i.e., real data) or is generated by the generator (i.e., generated data). During training, the generator and discriminator engage in adversarial competition. The generator strives to generate generative high-dimensional feature vectors that can fool the discriminator, while the discriminator continuously improves its ability to distinguish between real data and generated data.

[0046] Preferably, by calculating adversarial loss (such as cross-entropy loss), the parameters (such as weights and biases) of the generator and discriminator are adjusted based on the loss value. After multiple rounds of iterative training, when the quality of the data generated by the generator is high enough that the discriminator has difficulty distinguishing between real and generated data, the trained generative adversarial network is obtained. Finally, the generative adversarial network uses its generative ability to supplement high-dimensional feature vector samples. That is, the generator generates new samples similar to the original high-dimensional feature vector samples based on the spectral and texture feature patterns of building materials learned during training. For example, when the number of samples of certain building materials in the original dataset is small or samples under certain lighting conditions are missing, the generated new samples can fill these gaps, which helps to improve the training effect and performance of the spectral-texture co-enhancement network.

[0047] Furthermore, step S510 also includes step S514, where the spectral-texture collaborative enhancement network further includes a dynamic weight adjustment module, which dynamically adjusts the fusion weights of spectral features and texture features according to the exterior material type and the real-time lighting conditions; step S515, the defect mapping integrated data is weighted according to the fusion weights to obtain the second target mapping integrated data.

[0048] Preferably, the spectral-texture co-enhancement network also includes a dynamic weight adjustment module. Since the importance of spectral and texture features for accurately identifying and extracting building information varies depending on the facade material and the lighting conditions, the dynamic weight adjustment module is used to dynamically adjust the fusion weights of spectral and texture features according to the facade material type and real-time lighting conditions. For example, metal facades have high reflectivity, and spectral features play a dominant role in identifying highly reflective areas. By increasing the weight of spectral features, the spectral reflectance differences of metal materials under different lighting conditions can be captured more accurately, thereby improving the recognition accuracy of highly reflective areas. For glass facades, their transparency may cause blurred edges in the image. Texture features are more critical for solving the problem of blurred edges of transparent materials. By increasing the weight of texture features, the texture details of the glass surface can be better highlighted, helping to more clearly define the edges of the glass material.

[0049] Preferably, after determining the fusion weights of spectral and texture features, the defect mapping integrated data undergoes defect data weight enhancement. Specifically, based on different fusion weights, the spectral and texture features in the defect mapping integrated data are weighted separately. If the current material is metal and the spectral feature weight is high, the spectral feature portion of the data is given a greater weight to enhance its influence in the data; for the texture feature portion, it is processed according to the corresponding weight. By adjusting the weights of spectral and texture features, important features related to the current facade material and lighting conditions can be highlighted, further optimizing the defect mapping integrated data and obtaining the second target mapping integrated data. By comprehensively considering the weight changes of spectral and texture features under different facade materials and lighting conditions, the defect portion is enhanced more specifically. Compared with the defect mapping integrated data, its quality is higher and it can more accurately reflect the characteristics of the target building, thereby generating a more accurate target building extraction model, which helps to improve the accuracy and reliability of target building extraction.

[0050] Furthermore, step S500 also includes step S540, if there are multiple target buildings to be extracted, obtaining multiple target mapping integrated data corresponding to the multiple target buildings; step S550, extracting multiple high-dimensional feature vectors representing the facade materials of the multiple target buildings based on the multiple target mapping integrated data; step S560, inputting the multiple high-dimensional feature vectors into the spectrum-texture collaborative enhancement model to obtain multiple extraction models corresponding to the multiple target buildings.

[0051] Preferably, when there are multiple target buildings to be extracted within the target area, target mapping integrated data corresponding to each target building is acquired. The mapping integrated database contains data acquired in parallel from multiple sources (such as optical remote sensing data, laser scanning data, oblique photogrammetry data, etc.). The target mapping integrated data acquired for each target building can comprehensively reflect the characteristics of the building, such as its location, shape, and the spectral and texture characteristics of its facade materials. Then, high-dimensional feature vectors representing the facade materials of each target building are extracted from the target mapping integrated data. That is, spectral feature vectors are obtained by analyzing spectral information, and surface texture feature vectors are extracted from surface texture information using methods such as gray-level co-occurrence matrix. The two are then combined to form a high-dimensional feature vector. For multiple target buildings, multiple corresponding high-dimensional feature vectors are obtained, and each high-dimensional feature vector accurately describes the characteristics of the facade materials of the corresponding target building.

[0052] Preferably, the high-dimensional feature vectors of multiple target buildings are input into the spectral-texture co-enhancement model. The model analyzes and processes each high-dimensional feature vector, considering the synergistic effect of spectral and texture features and other relevant factors (such as lighting conditions) to determine the degree of influence on data extraction. It may also be necessary to activate the spectral-texture co-enhancement network to enhance the data. Finally, for each high-dimensional feature vector, a corresponding extraction model for the target building is generated. The extraction model can accurately identify and extract relevant information of the corresponding target building based on the input data, such as the building outline and the type of exterior facade material, thereby ensuring efficient and accurate processing of relevant information of multiple target buildings.

[0053] In the above text, refer to Figure 1 A method for rapid extraction of building targets from integrated surveying data according to embodiments of the present invention is described in detail. Next, reference will be made to... Figure 2 A rapid building target extraction system based on integrated surveying data is described according to an embodiment of the present invention.

[0054] The rapid building target extraction system based on integrated surveying data according to embodiments of the present invention addresses the technical problem in existing technologies where it is difficult to effectively identify target buildings composed of different building materials based on their reflective properties, resulting in low building recognition rates and poor extraction accuracy. The system efficiently generates accurate extraction models, thereby improving the technical effect of increasing the building target recognition rate and accuracy. Figure 2 As shown, the rapid building target extraction system based on integrated surveying and mapping data includes: a surveying and mapping integrated database construction module 10, a first target surveying and mapping integrated data acquisition module 20, an exterior facade feature vector extraction module 30, a data extraction impact analysis module 40, and a target building extraction model generation module 50.

[0055] The system includes: a mapping integrated database construction module 10, used to integrate multi-source parallel acquisition data to construct a mapping integrated database for a target area, the target area including the target building to be extracted; a first target mapping integrated data acquisition module 20, used to acquire the first target mapping integrated data of the target building in the mapping integrated database; an exterior feature vector extraction module 30, used to extract high-dimensional feature vectors representing the exterior materials of the target building based on the first target mapping integrated data, the high-dimensional feature vectors being composed of spectral feature vectors and surface texture feature vectors; a data extraction impact analysis module 40, used to input the high-dimensional feature vectors into a spectral-texture co-enhancement model for data extraction impact analysis, and output a first impact coefficient; and a target building extraction model generation module 50, used to activate a spectral-texture co-enhancement network to enhance the first target mapping integrated data if the first impact coefficient is greater than a preset impact coefficient, to acquire second target mapping integrated data, and to generate an extraction model of the target building based on the second target mapping integrated data.

[0056] The specific configuration of the data extraction impact analysis module 40 will be described in detail below. The data extraction impact analysis module 40 further includes: the spectral-texture collaborative enhancement model determining the exterior material type of the target building based on the high-dimensional feature vector; acquiring the real-time illumination conditions of the multi-source parallel acquisition data, including real-time illumination intensity and real-time illumination angle; establishing a mapping function relationship between the exterior material type samples and the illumination condition samples to characterize the degree of data extraction impact; analyzing the exterior material type and the real-time illumination conditions based on the mapping function relationship; and outputting a first impact coefficient.

[0057] The specific configuration of the target building extraction model generation module 50 will be described in detail below. The target building extraction model generation module 50 further includes: activating a spectral-texture collaborative enhancement network to identify boundary defects in the first target mapping integrated data and output defect mapping integrated data; obtaining matching lighting conditions with an influence coefficient less than a preset influence coefficient based on the facade material type; and using matching mapping integrated sample data corresponding to the matching lighting conditions to enhance the defect mapping integrated data to obtain second target mapping integrated data.

[0058] The specific configuration of the target building extraction model generation module 50 will be described in detail below. The target building extraction model generation module 50 further includes: acquiring high-dimensional feature vector samples, which include spectral feature vector samples and surface texture feature samples of different building materials under different lighting conditions; the spectral feature vector samples include band reflectance of spectral images, and the surface texture feature vector samples include contrast, energy, and correlation parameters calculated by GLCM; acquiring label sample data, which includes material categories and corresponding labeled mapping integrated sample data; and training a model based on the high-dimensional feature vector samples and the label sample data to obtain a converged spectral-texture co-enhancement network.

[0059] The specific configuration of the target building extraction model generation module 50 will be described in detail below. The target building extraction model generation module 50 further includes: initializing a generative adversarial network, including initializing a dual-branch generator and a discriminator; wherein the dual-branch generator includes a spectral branch and a texture branch, the spectral branch simulating the band reflectance changes of different building materials under different lighting conditions through a fully connected layer, and the texture branch simulating the surface texture changes of different building materials under different lighting conditions through a deconvolution layer; after inputting the high-dimensional feature vector samples into the dual-branch generator, a generative high-dimensional feature vector is output; the discriminator receives the generative high-dimensional feature vector output by the dual-branch generator and performs adversarial loss feedback training with the high-dimensional feature vector samples to obtain a trained generative adversarial network; and the trained generative adversarial network is invoked to supplement the high-dimensional feature vector samples.

[0060] The specific configuration of the mapping integrated database construction module 10 will be described in detail below. The mapping integrated database construction module 10 further includes: if there are multiple target buildings to be extracted, acquiring multiple target mapping integrated data corresponding to the multiple target buildings; extracting multiple high-dimensional feature vectors representing the exterior materials of the multiple target buildings based on the multiple target mapping integrated data; and inputting the multiple high-dimensional feature vectors into the spectral-texture collaborative enhancement model to obtain multiple extraction models corresponding to the multiple target buildings.

[0061] The specific configuration of the target building extraction model generation module 50 will be described in detail below. The target building extraction model generation module 50 further includes: the spectral-texture collaborative enhancement network also includes a dynamic weight adjustment module, which dynamically adjusts the fusion weights of spectral features and texture features according to the exterior material type and the real-time lighting conditions; and performs defect data weight enhancement on the defect mapping integrated data according to the fusion weights to obtain the second target mapping integrated data.

[0062] The building target rapid extraction system based on integrated surveying data provided in this embodiment of the invention can execute the building target rapid extraction method based on integrated surveying data provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0063] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0064] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for rapid extraction of building targets from integrated surveying and mapping data, characterized in that, The method includes: Integrate multi-source parallel acquisition data to construct a mapping integration database for the target area, which includes the target buildings to be extracted; Obtain the first target surveying and mapping integration data of the target building in the surveying and mapping integration database; Based on the first target mapping integrated data, extract the high-dimensional feature vector representing the exterior material of the target building. The high-dimensional feature vector is composed of spectral feature vector and surface texture feature vector. The high-dimensional feature vector is input into the spectral-texture co-enhancement model for data extraction and impact analysis, and the first impact coefficient is output. The spectral-texture co-enhancement model determines the exterior material type of the target building based on the high-dimensional feature vector. The real-time illumination conditions of the multi-source parallel acquisition data are obtained, including real-time illumination intensity and real-time illumination angle. Establish a mapping function relationship between the facade material type sample and the lighting condition sample to characterize the degree of influence of data extraction. Analyze the facade material type and the real-time lighting conditions according to the mapping function relationship and output the first influence coefficient. If the first influence coefficient is greater than the preset influence coefficient, the spectral-texture collaborative enhancement network is activated to perform data enhancement on the first target mapping integrated data, obtain the second target mapping integrated data, and generate the extraction model of the target building based on the second target mapping integrated data; Among them, the activated spectral-texture collaborative enhancement network is used to identify boundary defects in the first target mapping integration data and output defect mapping integration data; Based on the type of exterior facade material, obtain matching lighting conditions with an influence coefficient less than a preset influence coefficient; The defect mapping integrated data is enhanced by calling the matching mapping integrated sample data corresponding to the matching illumination conditions to obtain the second target mapping integrated data.

2. The method for rapid extraction of building targets from integrated surveying and mapping data as described in claim 1, characterized in that, Methods for training spectral-texture co-enhancement networks include: High-dimensional feature vector samples are obtained, including spectral feature vector samples and surface texture feature samples of different building materials under different lighting conditions. The spectral feature vector samples include the band reflectance of the spectral image, and the surface texture feature vector samples include contrast, energy, and correlation parameters calculated by GLCM. Acquire label sample data, which includes material categories and corresponding labeled integrated survey and mapping sample data; The model is trained based on the high-dimensional feature vector samples and the label sample data to obtain a spectral-texture co-enhancement network that has been trained to convergence.

3. The method for rapid extraction of building targets from integrated surveying and mapping data as described in claim 2, characterized in that, The input of the spectral-texture co-enhancement network is connected to a generative adversarial network, and the method includes: Initialize the generative adversarial network, including initializing the bi-branch generator and discriminator; The dual-branch generator includes a spectral branch and a texture branch. The spectral branch simulates the band reflectivity changes of different building materials under different lighting conditions through a fully connected layer, and the texture branch simulates the surface texture changes of different building materials under different lighting conditions through a deconvolution layer. After the high-dimensional feature vector sample is input into the dual-branch generator, a generative high-dimensional feature vector is output. The discriminator receives the generative high-dimensional feature vector output by the dual-branch generator and performs adversarial loss feedback training with the high-dimensional feature vector sample to obtain the trained generative adversarial network. The trained generative adversarial network is invoked to supplement the high-dimensional feature vector samples.

4. The method for rapid extraction of building targets from integrated surveying and mapping data as described in claim 1, characterized in that, The methods for constructing an integrated mapping database for the target area also include: If there are multiple target buildings to be extracted, obtain integrated mapping data for multiple target buildings; Based on the integrated mapping data of the multiple targets, extract multiple high-dimensional feature vectors representing the exterior facade materials of the multiple target buildings; The multiple high-dimensional feature vectors are input into the spectral-texture co-enhancement model to obtain multiple extraction models corresponding to the multiple target buildings.

5. The method for rapid extraction of building targets from integrated surveying and mapping data as described in claim 1, characterized in that, The spectral-texture synergistic enhancement network also includes a dynamic weight adjustment module, which dynamically adjusts the fusion weights of spectral features and texture features according to the exterior material type and the real-time lighting conditions. Based on the fusion weight, the defect mapping integrated data is weighted to enhance the defect data weight, thereby obtaining the second target mapping integrated data.

6. A rapid building target extraction system based on integrated surveying and mapping data, characterized in that, The system is used to implement the rapid extraction method for building targets from integrated surveying and mapping data as described in any one of claims 1 to 5. The system comprises: The mapping integrated database construction module is used to integrate multi-source parallel acquisition data and construct a mapping integrated database for a target area, wherein the target area includes the target buildings to be extracted. The first target mapping integrated data acquisition module is used to acquire the first target mapping integrated data of the target building in the mapping integrated database. The facade feature vector extraction module is used to extract high-dimensional feature vectors representing the facade materials of the target building based on the first target mapping integrated data. The high-dimensional feature vectors are composed of spectral feature vectors and surface texture feature vectors. The data extraction impact analysis module is used to input the high-dimensional feature vector into the spectral-texture co-enhancement model for data extraction impact analysis and output a first impact coefficient. The spectral-texture co-enhancement model determines the exterior material type of the target building based on the high-dimensional feature vector, obtains the real-time lighting conditions of the multi-source parallel acquisition data, including real-time lighting intensity and real-time lighting angle, establishes a mapping function relationship between the exterior material type sample and the lighting condition sample to characterize the degree of data extraction impact, analyzes the exterior material type and the real-time lighting conditions based on the mapping function relationship, and outputs a first impact coefficient. The target building extraction model generation module is used to activate a spectral-texture collaborative enhancement network to perform data enhancement on the first target mapping integrated data if the first influence coefficient is greater than a preset influence coefficient, obtain second target mapping integrated data, and generate an extraction model of the target building based on the second target mapping integrated data. Specifically, activating the spectral-texture collaborative enhancement network is used to identify boundary defects in the first target mapping integrated data and output defect mapping integrated data. Based on the exterior material type, matching lighting conditions with influence coefficients less than a preset influence coefficient are obtained. Matching mapping integrated sample data corresponding to the matching lighting conditions is then called to perform defect data enhancement on the defect mapping integrated data to obtain the second target mapping integrated data.

Images