Building identification and modeling method and device based on satellite remote sensing image

Abstract

The invention discloses a building identification and modeling method and device based on a satellite remote sensing image, and belongs to the technical field of image processing, and the method comprises the steps: carrying out the preprocessing of a target satellite remote sensing image, dynamically adjusting the resolution of an input image, retaining key detail information, and carrying out the storage and block management of historical satellite remote sensing images. The method comprises the following steps: establishing a cross-scale recognition model based on historical satellite remote sensing images, performing image recognition on the satellite remote sensing images by using the cross-scale recognition model, obtaining building geometric features in the images, performing semantic extraction on the building geometric features, and generating a building three-dimensional model according to extracted semantic attributes. The incidence relation between the building contour and the geometric morphology is dug to realize deep fusion of the features, the comprehensiveness and accuracy of feature expression are improved, and the reliability and practicability of a modeling result are enhanced.

Description

Technical Field

[0001] This application belongs to the field of image processing technology, and more specifically, relates to a method and apparatus for identifying and modeling buildings based on satellite remote sensing images. Background Technology

[0002] With the rapid development of satellite remote sensing technology, high-resolution satellite remote sensing images are playing an increasingly crucial role in many fields such as urban planning, disaster emergency response, real estate registration, and smart city construction. Among these, the accurate identification and 3D modeling of buildings, as important artificial features on the Earth's surface, is one of the core tasks of remote sensing image interpretation. Achieving automated identification and high-precision modeling of buildings through satellite remote sensing images can effectively improve work efficiency in related fields, reduce labor costs, and provide reliable data support for decision-making.

[0003] Currently, building recognition and modeling technology based on satellite remote sensing images has become a research hotspot in the field of remote sensing technology applications, and related technologies are constantly emerging. However, in practical applications, it is difficult to simultaneously and accurately capture the global geometric shape and local edge texture features of buildings, and the feature expression capability is insufficient. It still faces problems such as insufficient image resolution adaptability, incomplete feature extraction, limited recognition accuracy, and lack of modeling reliability. Summary of the Invention

[0004] To address the aforementioned problems and technical deficiencies, this application adopts the following technical solution: a method for identifying and modeling buildings based on satellite remote sensing images, comprising the following steps: Preprocess the target satellite remote sensing image, dynamically adjust the input image resolution, and retain key detail information; Historical satellite remote sensing images are stored and managed in blocks, and cross-scale recognition models are established based on historical satellite remote sensing images; A cross-scale recognition model is used to perform image recognition on satellite remote sensing images to obtain the geometric features of buildings in the images; Semantic extraction is performed on the geometric features of the building, and a 3D model of the building is generated based on the extracted semantic attributes.

[0005] Preferably, the dynamic resolution adjustment involves constructing a resolution mapping table, using corresponding resolution sizes for images with different details, and then setting a feedback mechanism. If the detail retention rate is lower than a threshold after adjustment, the interpolation parameters are re-optimized.

[0006] Preferably, the cross-scale recognition model includes a multi-scale feature pyramid encoder and a geometric prior constraint decoder; The encoder is configured with two feature extraction branch channels: a geometric information branch and a detail-aware branch. Geometric information branch extracts the geometric contours or shapes of buildings in the global image; The detail-aware branch extracts local edges and textures of buildings in the image; The decoder extracts the relationship between the building's outline and its geometric shape.

[0007] Furthermore, the image recognition includes: For each independent temporal phase of the image, perform an initial classification of buildings and non-buildings; By fusing multi-temporal classification and extraction results, a complete building mask is generated; Differential feature fusion is used to capture changes in buildings across different time phases; Vectorized building boundaries are generated by combining building masks and building variations, and then edge constraints are applied. A cross-scale recognition model is used to extract geometric features from the building boundaries after edge constraints.

[0008] Furthermore, the semantic extraction includes: Construct a mapping rule base to store the mapping relationship between geometric features and building types; Determining building size and number of floors based on edge texture; The building type is determined based on its outline shape and relationships; The geometric loss for modeling is calculated using the fusion loss function.

[0009] Furthermore, the formula for calculating the modeling geometric loss is as follows: L=λ1×L geo + λ2×L sem + λ3×L reg Where L is the modeling geometric loss function, L geo For geometric loss, the upper mean square error is used to calculate the error between the geometric characteristics and the measured values, L. sem Semantic loss is calculated using cross-entropy loss to determine the error between the predicted type and the actual type. reg λ1, λ2, and λ3 are the regularization loss and the constraint model parameters, respectively. λ1, λ2, and λ3 are the corresponding weight coefficients, and λ1 + λ2 + λ3 = 1.

[0010] Furthermore, the 3D model of the building is generated by using a digital surface model to drive the workflow, combining the building size, number of floors, type and geometric loss to perform parameter calculations, and then generating the 3D model of the building based on the parameter calculation results.

[0011] A device for identifying and modeling buildings based on satellite remote sensing images, the device comprising a service processor and a distributed memory, the service processor being connected to the memory, the distributed memory storing a service self-management program configured to store machine-readable instructions, the service processor executing the service self-management program, the instructions, when executed by the processor, to implement the building identification and modeling method based on satellite remote sensing images as described above.

[0012] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the content of the method for identifying and modeling buildings based on satellite remote sensing images as described above.

[0013] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the content of the method for identifying and modeling buildings based on satellite remote sensing images as described above.

[0014] Compared to existing technologies, the beneficial effects of this application are as follows: (1) This application constructs a resolution mapping table to match the appropriate resolution size for images with different levels of detail richness, and introduces a feedback mechanism to optimize the interpolation parameters, thereby solving the problem of detail loss or redundancy caused by fixed resolution processing. This ensures the complete preservation of key detail information and improves the pertinence and efficiency of image preprocessing. (2) This application designs a multi-scale feature pyramid encoder and a geometric prior constraint decoder to accurately extract the global geometric contour morphology and local edge texture features of the building, respectively, and explores the relationship between the building contour and geometric morphology to achieve deep feature fusion, thereby improving the comprehensiveness and accuracy of feature expression; (3) This application generates a complete building mask by performing initial classification on each independent temporal phase and fusing the results of multiple temporal phases. It also uses differential feature fusion to capture building change information, combines the mask and change information to generate vectorized boundaries and perform edge constraints, constructs a mapping rule library to achieve semantic extraction, realizes deep fusion of geometric features and semantic information, improves the matching degree between the three-dimensional model and the actual building, and enhances the reliability and practicality of the modeling results. Attached Figure Description

[0015] In the attached diagram: Figure 1 This is a schematic diagram of the method steps in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments. Generally, the components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Example

[0017] like Figure 1 As shown, the method for identifying and modeling buildings based on satellite remote sensing images includes the following steps: Preprocess the target satellite remote sensing image, dynamically adjust the input image resolution, and retain key detail information; Dynamic resolution adjustment involves constructing a resolution mapping table, applying corresponding resolution sizes to images with different levels of detail, and then setting a feedback mechanism. If the detail retention rate is lower than the threshold after adjustment, the interpolation parameters are re-optimized.

[0018] Historical satellite remote sensing images are stored and managed in blocks, and cross-scale recognition models are established based on historical satellite remote sensing images; The cross-scale recognition model includes a multi-scale feature pyramid encoder and a geometric prior constraint decoder; The encoder is configured with two feature extraction branch channels: a geometric information branch and a detail-aware branch. Geometric information branch extracts the geometric contours or shapes of buildings in the global image; The detail-aware branch extracts local edges and textures of buildings in the image; The decoder extracts the relationship between the building's outline and its geometric shape.

[0019] A cross-scale recognition model is used to perform image recognition on satellite remote sensing images to obtain the geometric features of buildings in the images; Image recognition includes: For each independent temporal phase of the image, perform an initial classification of buildings and non-buildings; By fusing multi-temporal classification and extraction results, a complete building mask is generated; Differential feature fusion is used to capture changes in buildings across different time phases; Vectorized building boundaries are generated by combining building masks and building variations, and then edge constraints are applied. A cross-scale recognition model is used to extract geometric features from the building boundaries after edge constraints.

[0020] Semantic extraction is performed on the geometric features of the building, and a 3D model of the building is generated based on the extracted semantic attributes.

[0021] Semantic extraction includes: Construct a mapping rule base to store the mapping relationship between geometric features and building types; Determining building size and number of floors based on edge texture; The building type is determined based on its outline shape and relationships; The geometric loss for modeling is calculated using the fusion loss function.

[0022] The formula for calculating the geometric loss in modeling is as follows: L=λ1×L geo + λ2×L sem + λ3×L reg Where L is the modeling geometric loss function, L geo For geometric loss, the upper mean square error is used to calculate the error between the geometric characteristics and the measured values, L. sem Semantic loss is calculated using cross-entropy loss to determine the error between the predicted type and the actual type. reg λ1, λ2, and λ3 are the regularization loss and the constraint model parameters, respectively. λ1, λ2, and λ3 are the corresponding weight coefficients, and λ1 + λ2 + λ3 = 1.

[0023] The 3D model of a building is created by using a digital surface model to drive the workflow, combining the building's dimensions, number of floors, type, and geometric loss to calculate parameters, and then building the 3D model of the building based on the parameter calculation results. Example

[0024] Key detail quantification employs an edge density detection algorithm and contour integrity assessment to construct a detail importance index D, with the formula as follows: D = α × (number of edge pixels / total number of pixels in the image) + β × (contour overlap) Where α and β are weighting coefficients, when D≥0.7 it is judged as a high detail image, when D≤0.3 it is a low detail image, and in between is a medium detail image.

[0025] Dynamic resolution adjustment involves building a resolution mapping table: The high-detail images are progressively upgraded in resolution from 10m×10m to 5m×5m to preserve architectural texture details; Medium-detail images are rendered at a fixed resolution of 7.5m × 7.5m. Low-detail images are rendered at a 10m×10m resolution, and noise is reduced through neighboring pixel fusion. A feedback mechanism is also set up so that if the detail retention rate is lower than 0.9 after adjustment, the interpolation parameters are automatically re-optimized.

[0026] The bottom layer of the multi-scale feature pyramid encoder is the feature input layer, which receives preprocessed images at different resolutions. The middle layer is a dual-branch feature extraction layer. The geometric information branch uses dilated convolution (dilation rate 2, 4, 6) to extract the global geometric contour and obtains multi-scale geometric features through hollow spatial pyramid pooling. The detail-aware branch uses an attention mechanism to focus on local edges and textures, and dynamically adjusts the convolution kernel size (switching between 3×3 and 5×5) to adapt to different detail densities.

[0027] The dual-branch outputs achieve adaptive fusion through a feature fusion gating unit, and the fusion weights are dynamically adjusted according to the detail importance index D. The top layer is the feature pyramid generation layer, which generates feature maps at four scales (1 / 4, 1 / 8, 1 / 16, and 1 / 32).

[0028] The geometric prior constraint decoder incorporates a building geometric prior library. During feature upsampling, the decoder compares the extracted features with the prior library templates through a geometric matching unit, calculates the matching degree, and corrects feature regions with a matching degree lower than 0.5.

[0029] Simultaneously, a contour-morphology association network is constructed, which learns the relationship between building contour nodes and geometric morphology parameters through graph neural networks, and outputs structured geometric features.

[0030] Single-phase classification employs an improved lightweight semantic segmentation network to perform binary classification of buildings and non-buildings in each phase preprocessed image, outputting a classification probability map. A hard sample mining mechanism is introduced to label ambiguous regions, which will be the focus of subsequent optimization.

[0031] Multi-temporal fusion adopts a weighted fusion strategy, with weights determined based on image quality assessment results and classification confidence. After fusion, an initial building mask is generated, and blurred regions are supplemented by interpolation of high-confidence regions in adjacent temporal phases.

[0032] Differential feature fusion uses the highest quality time phase as the reference time phase, selects the reference time phase and the target time phase, and constructs a differential feature extraction branch: Pixel-level difference: calculate the grayscale difference and texture difference between two time phases; Feature-level difference involves inputting two temporal phases into a cross-scale encoder, extracting high-level features, and then calculating the Euclidean distance. Differential features are fused through convolution and then input into a change detection network, which outputs a mask of the changed areas of buildings.

[0033] Vectorization and edge constraints employ a contour extraction algorithm to extract building boundaries from fused and varied masks, resulting in an initial vectorized contour.

[0034] Edge constraints include: Geometric constraints, based on a library of building geometry priors, correct protrusions or indentations in the outline that do not conform to common building forms; Accuracy constraints are implemented by calibrating the contour position using GPS-measured control points to control errors. The final output is a precisely vectorized building boundary, which serves as input for geometric feature extraction. Example

[0035] A device for identifying and modeling buildings based on satellite remote sensing images is provided. The device includes a service processor and a distributed memory. The service processor is connected to the memory. The distributed memory stores a service self-management program configured to store machine-readable instructions. The service processor executes the service self-management program. When the instructions are executed by the processor, they implement the method for identifying and modeling buildings based on satellite remote sensing images as described in Embodiment 1. Example

[0036] From a hardware perspective, this application provides an embodiment of an electronic device containing all or part of a method for identifying and modeling buildings based on satellite remote sensing images. The electronic device includes a service processor and a distributed memory. The service processor is connected to the memory, and the distributed memory stores a service self-management program configured to store machine-readable instructions. The service processor executes the service self-management program, and the instructions, when executed by the processor, implement the method for identifying and modeling buildings based on satellite remote sensing images as described above. Example

[0037] The embodiments of this application also provide a computer-readable storage medium capable of implementing the building identification and modeling method based on satellite remote sensing images, where the execution subject is a server or client as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all the contents of the building identification and modeling method based on satellite remote sensing images, where the execution subject is a server or client as described in the above embodiments.

[0038] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of this application, and these all fall within the protection scope of this application.

Claims

1. A method for identifying and modeling buildings based on satellite remote sensing images, characterized in that, Includes the following steps: Preprocess the target satellite remote sensing image, dynamically adjust the input image resolution, and retain key detail information; Historical satellite remote sensing images are stored and managed in blocks, and cross-scale recognition models are established based on historical satellite remote sensing images; A cross-scale recognition model is used to perform image recognition on satellite remote sensing images to obtain the geometric features of buildings in the images; Semantic extraction is performed on the geometric features of the building, and a 3D model of the building is generated based on the extracted semantic attributes.

2. The method for identifying and modeling buildings based on satellite remote sensing images according to claim 1, characterized in that, The dynamic resolution adjustment involves constructing a resolution mapping table, using corresponding resolution sizes for images with different details, and then setting a feedback mechanism. If the detail retention rate is lower than the threshold after adjustment, the interpolation parameters are re-optimized.

3. The method for identifying and modeling buildings based on satellite remote sensing images according to claim 1, characterized in that, The cross-scale recognition model includes a multi-scale feature pyramid encoder and a geometric prior constraint decoder. The encoder is configured with two feature extraction branch channels: a geometric information branch and a detail-aware branch. Geometric information branch extracts the geometric contours or shapes of buildings in the global image; The detail-aware branch extracts local edges and textures of buildings in the image; The decoder extracts the relationship between the building's outline and its geometric shape.

4. The method for identifying and modeling buildings based on satellite remote sensing images according to claim 3, characterized in that, The image recognition includes: For each independent temporal phase of the image, perform an initial classification of buildings and non-buildings; By fusing multi-temporal classification and extraction results, a complete building mask is generated; Differential feature fusion is used to capture changes in buildings across different time phases; Vectorized building boundaries are generated by combining building masks and building variations, and then edge constraints are applied. A cross-scale recognition model is used to extract geometric features from the building boundaries after edge constraints.

5. The method for identifying and modeling buildings based on satellite remote sensing images according to claim 3, characterized in that, The semantic extraction includes: Construct a mapping rule base to store the mapping relationship between geometric features and building types; Determining building size and number of floors based on edge texture; The building type is determined based on its outline shape and relationships; The geometric loss for modeling is calculated using the fusion loss function.

6. The method for identifying and modeling buildings based on satellite remote sensing images according to claim 5, characterized in that, The formula for calculating the modeling geometric loss is as follows: L=λ1×L geo + λ2×L sem + λ3×L reg Where L is the modeling geometric loss function, L geo For geometric loss, the upper mean square error is used to calculate the error between the geometric characteristics and the measured values, L. sem Semantic loss is calculated using cross-entropy loss to determine the error between the predicted type and the actual type. reg λ1, λ2, and λ3 are the regularization loss and the constraint model parameters, respectively. λ1, λ2, and λ3 are the corresponding weight coefficients, and λ1 + λ2 + λ3 = 1.

7. The method for identifying and modeling buildings based on satellite remote sensing images according to claim 6, characterized in that, The building's 3D model is generated by using a digital surface model to drive the workflow, combining building dimensions, number of floors, type, and geometric loss to perform parameter calculations, and then generating the building's 3D model based on the parameter calculation results.

8. A device for identifying and modeling buildings based on satellite remote sensing images, characterized in that, The device includes a service processor and a distributed memory. The service processor is connected to the memory, and the distributed memory stores a service self-management program configured to store machine-readable instructions. The service processor executes the service self-management program, and the instructions, when executed by the processor, implement the building identification and modeling method based on satellite remote sensing images as described in claim 1.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the content of the building identification and modeling method based on satellite remote sensing images as described in claim 1.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for identifying and modeling buildings based on satellite remote sensing images as described in claim 1.

Images