Method for spatializing building embodied carbon emission based on remote sensing data

Abstract

This invention relates to the field of building occult carbon emissions measurement technology, and in particular to a spatialized method for calculating building occult carbon emissions based on remote sensing data. The method includes: a deep learning network for building height estimation infers the predicted building height of each raster pixel using radar and optical remote sensing image features; analyzing the building coverage of each raster pixel based on building footprint vector data; calculating the building occult carbon emissions of each raster pixel based on the corresponding predicted building height, building coverage, building type, and building material type; and generating spatial distribution characteristics of building occult carbon emissions for the target measurement space based on the building occult carbon emissions of each raster pixel. This method aims to identify the spatial distribution patterns of building occult carbon emissions within large-scale, cross-regional urban building clusters, analyze the continuous spatial variation characteristics of occult carbon emissions, and provide guidance for spatial planning of urban building expansion and the formulation of carbon reduction strategies.

Description

Technical Field

[0001] This invention relates to the field of building hidden carbon emissions measurement technology, and in particular to a spatial calculation method for building hidden carbon emissions based on remote sensing data. Background Technology

[0002] Building embodied carbon emissions refer to the greenhouse gas emissions generated throughout a building's entire life cycle, from raw material extraction, building material production, transportation to the construction site, construction and installation, to later maintenance, demolition, and disposal. The building material production stage is a significant component of building life-cycle carbon emissions. With the continuous advancement of urbanization, the expansion of built-up areas, and the sustained growth of building volume, there is an urgent need to conduct systematic accounting and assessment of building embodied carbon emissions to support related carbon emission monitoring, statistics, and management.

[0003] Existing methods for calculating embodied carbon emissions from buildings are mostly based on statistical yearbooks, energy balance sheets, or building material production data, using administrative divisions as the basic unit for top-down total allocation. These methods fail to analyze the spatial distribution patterns of embodied carbon emissions across different blocks and plots within a city, and cannot directly reflect the continuous spatial variation characteristics of embodied carbon emissions from building complexes. Some studies have attempted to combine building outline vector data with building attribute ledgers for bottom-up calculations of embodied carbon emissions; however, these methods rely on detailed building survey data and structural design data, which are often difficult to obtain in large-scale, cross-regional urban building complex applications. Therefore, identifying the spatial distribution patterns of embodied carbon emissions within large-scale, cross-regional urban building complexes and analyzing the continuous spatial variation characteristics of embodied carbon emissions to provide guidance for spatial planning of urban building expansion and the formulation of carbon reduction strategies has become a pressing technical problem in this field. Summary of the Invention

[0004] This invention aims to provide a spatial method for calculating the implicit carbon emissions of buildings based on remote sensing data, in order to identify the spatial distribution patterns of implicit carbon emissions within large-scale, cross-regional urban building clusters, analyze the continuous spatial variation characteristics of implicit carbon emissions, and provide guidance for spatial planning of urban building expansion and the formulation of carbon reduction strategies.

[0005] To achieve the above objectives, the first aspect of the present invention provides a method for spatially calculating the implicit carbon emissions of buildings based on remote sensing data, comprising the following steps: Determine the target measurement space and the raster pixel resolution of the target measurement space; The radar image features and optical remote sensing image features of the target measurement space are obtained based on the grid pixel resolution. A deep learning network for building height estimation is trained, which then infers the predicted building height of each grid cell in the target measurement space based on the radar image features and the optical remote sensing image features. Obtain building footprint vector data of the target measurement space, and analyze the building coverage rate of each raster cell in the target measurement space based on the building footprint vector data; Obtain the building category and building material category of each raster cell within the target measurement space; For any grid cell within the target measurement space, the building's implicit carbon emissions are calculated based on the corresponding predicted building height, building coverage, building category, and building material category, thereby obtaining the building's implicit carbon emissions for each grid cell within the target measurement space. The spatial distribution characteristics of the building's hidden carbon emissions in the target measurement space are generated based on the building's hidden carbon emissions for each raster cell.

[0006] The aforementioned method for spatializing building carbon emissions based on remote sensing data acquires radar image features and optical remote sensing image features that match the resolution of the raster pixels, and trains a deep learning network for building height estimation to infer the predicted building height value for each raster pixel. Without relying on detailed building survey data and structural design data, it identifies the distribution of building heights within a large-scale, cross-regional target measurement area for spatializing building carbon emissions.

[0007] Because a single pixel may contain multiple land cover types, such as buildings, roads, and green spaces, directly treating pixels with non-zero building heights as completely covered by buildings would lead to an overestimation of building area, thus affecting the accuracy of implicit carbon emission calculations. Therefore, building coverage is determined through spatial overlay analysis of building footprint vector data and raster pixels, incorporating the actual building footprint into the calculation process to avoid overestimating the carbon emissions of raster pixels. Finally, considering the differences in material consumption intensity and carbon emission factors among different building types and building materials, the building category and building material category of each raster pixel are obtained. Based on this, combined with the predicted building height and building coverage, the implicit carbon emissions of each raster pixel are calculated. Based on the implicit carbon emissions of each raster pixel, the spatial distribution characteristics of implicit carbon emissions in the target calculation space are generated. This allows the raster pixel to be used as the smallest spatial unit to display the spatial differences and distribution patterns of implicit carbon emissions of building clusters within the target calculation area, providing guidance for spatial planning of urban building expansion and the formulation of carbon reduction strategies.

[0008] Further, the step of enabling the deep learning network for building height estimation to infer the predicted building height of each raster cell within the target measurement space using the radar image features and the optical remote sensing image features includes: The radar image features are input into the radar image feature encoder in the building height estimation deep learning network, and the optical remote sensing image features are input into the optical remote sensing image feature encoder in the building height estimation deep learning network; wherein, the radar image feature encoder and the optical remote sensing image feature encoder have the same structure and their parameters are independent of each other; The radar image feature encoder extracts the radar image feature multi-scale deep features of the radar image, and the optical remote sensing image feature encoder extracts the optical remote sensing image feature multi-scale deep features of the optical remote sensing image. Using the decoder of the deep learning network for building height estimation, the spatial resolution of the multi-scale deep features of the radar image and the multi-scale deep features of the optical remote sensing image is restored by upsampling in a stepwise manner, so as to obtain the multi-scale decoding features of the radar image and the multi-scale decoding features of the optical remote sensing image. The multi-scale deep features of the radar image, the multi-scale decoding features of the radar image, the multi-scale deep features of the optical remote sensing image, and the multi-scale decoding features of the optical remote sensing image are fused in a skip connection manner to obtain fused features; Based on the fused features, the predicted building height of each raster cell in the target measurement space is inferred.

[0009] In this implementation, the deep learning network for building height estimation employs a dual-encoder structure to process radar image features and optical remote sensing image features separately. The radar image feature encoder and the optical remote sensing image feature encoder have identical structures and independent parameters, allowing features from both modalities to be encoded in their respective independent parameter spaces. This avoids mutual interference from differences in the distribution of features from different sources, while structural consistency ensures alignment of encoding levels, which is beneficial for subsequent feature fusion. By encoding features from different remote sensing modalities separately, the impact of different imaging mechanisms and data distribution differences on the feature learning process can be reduced, thereby improving the stability and accuracy of building height feature representation. In the decoding stage, the decoder restores the spatial resolution of multi-scale deep features through progressive upsampling. While restoring spatial details, it gradually integrates high-level semantic information, ensuring that the output feature has the same spatial resolution as the input raster pixel resolution. A skip connection is used to fuse the multi-scale deep features output by the encoder with the multi-scale decoded features output by the decoder, introducing low-level spatial detail information while preserving high-level semantic information, compensating for high-frequency spatial detail information that may be lost during progressive upsampling. By inferring the building height prediction value for each raster cell based on fusion features, the building height estimation can simultaneously utilize the ability of radar imagery to express the three-dimensional geometry of the building and the ability of optical remote sensing imagery to express the surface texture of the building, thereby improving the accuracy of building height prediction.

[0010] Further, the step of obtaining the radar image features and optical remote sensing image features of the target measurement space based on the grid pixel resolution includes: Obtain the ground distance detection data of the target measurement space, and classify the ground distance detection data into ascending orbit images and descending orbit images; The number of ascending and descending orbit images is counted. If the number of ascending orbit images is greater than the number of descending orbit images, the ascending orbit images are used as unified orbit images. If the number of ascending orbit images is less than the number of descending orbit images, the descending orbit images are used as unified orbit images. If the number of ascending orbit images is equal to the number of descending orbit images, either the ascending orbit image or the descending orbit image is used as a unified orbit image. Extract the VV polarization (vertical transmit-vertical receive polarization) backscattering coefficient image and the VH polarization (vertical transmit-horizontal receive polarization) backscattering coefficient image within a preset time window from the unified orbit image; The VV polarization backscattering coefficient images within a preset time window are synthesized using the median synthesis method to obtain a synthesized VV polarization backscattering coefficient image. The VH polarization backscattering coefficient images within a preset time window are synthesized using the median synthesis method to obtain synthesized VH polarization backscattering coefficient images. The synthesized VV polarization backscattering coefficient image and the synthesized VH polarization backscattering coefficient image are used as radar image features.

[0011] In this implementation, ground distance detection data is classified into ascending orbit images and descending orbit images. A unified orbit image is selected based on quantity advantage, ensuring the consistency of radar imagery in observation geometry. This avoids the problem of incomparable backscattering coefficients caused by differences in incident angles across different orbital directions when acquiring satellite data, providing a unified observation benchmark for subsequent time-series synthesis. VV-polarized and VH-polarized backscattering coefficient images are extracted from the unified orbit imagery. VV-polarization is sensitive to the dihedral reflection and volume scattering characteristics of building structures, while VH-polarization is sensitive to the volume scattering and multiple scattering characteristics of building complexes. Using both together can more comprehensively characterize the backscattering coefficient image features of building targets. A median synthesis method is used to synthesize multiple images within a preset time window. Utilizing the median statistic's insensitivity to outliers, abnormal backscattering values ​​caused by instantaneous weather conditions and sensor noise are reduced, thereby improving the accuracy of the synthesized radar image features.

[0012] Further, the step of obtaining the radar image features and optical remote sensing image features of the target measurement space based on the grid pixel resolution includes: Acquire the surface reflectance data of the target measurement space, the surface reflectance data including visible light band, near-infrared band, short-wave infrared band and image quality control band; According to the image quality control band, mark the invalid and valid pixels in all the raster pixels in the target measurement space; A binary cloud mask is constructed based on the invalid pixels and the valid pixels. The visible light band, the near-infrared band and the short-wave infrared band are processed using the binary cloud mask to obtain the visible light band, the near-infrared band and the short-wave infrared band of the cloud mask. A number of cloud mask visible light bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the cloud mask visible light bands to obtain the synthesized cloud mask visible light bands. Several cloud mask near-infrared bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the several cloud mask near-infrared bands to obtain the synthesized cloud mask near-infrared bands. Several cloud mask shortwave infrared bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the several cloud mask shortwave infrared bands to obtain the synthesized cloud mask shortwave infrared bands. The visible light band, near-infrared band, and short-wave infrared band of the synthetic cloud mask are unified to the grid pixel resolution as the optical remote sensing image feature.

[0013] It should be noted that the visible light band includes the blue band, green band, and red band.

[0014] In this implementation, surface reflectance data including visible light, near-infrared, short-wave infrared, and image quality control bands are acquired. Invalid and valid pixels are marked using the image quality control bands, and a binary cloud mask is constructed to process the visible light, near-infrared, and short-wave infrared bands. Pixels contaminated by clouds and cloud shadows are removed from the surface reflectance data, avoiding interference from cloud contamination and ensuring that all spectral data used in subsequent synthesis are valid surface observations. The median synthesis method is used to statistically synthesize the spectral bands of multiple cloud mask images within a preset time window. This eliminates the influence of residual cloud contamination while reducing the impact of outliers caused by atmospheric condition fluctuations and sensor radiometric calibration deviations, resulting in time-stable synthesized spectral characteristics. By unifying the visible light band, near-infrared band, and short-wave infrared band of the synthetic cloud mask to the grid pixel resolution as optical remote sensing image features, the consistency of spatial resolution of each band is ensured, and the consistency of spatial resolution between optical remote sensing image features and radar image features is also ensured. This provides spatially aligned data conditions for the parallel input of the dual encoder network and subsequent feature fusion.

[0015] Further, the step of acquiring the building footprint vector data of the target measurement space, and analyzing the building coverage of each raster cell within the target measurement space based on the building footprint vector data, includes: The building footprint vector data is spatially overlaid with each raster cell in the target measurement space to determine the building footprint polygon contained within the range of each raster cell; For any of the aforementioned raster cells: Calculate the total area of ​​intersection between the building footprint polygon contained within the raster cell and the raster cell; The cell area of ​​the raster cell is obtained, and the ratio of the sum of the intersecting areas to the cell area is analyzed to obtain the building coverage of the raster cell.

[0016] In this implementation, the building footprint vector data is spatially overlaid with each raster cell within the target measurement space to determine the building footprint polygons contained within each raster cell, establishing a spatial relationship between the vector building outline and the raster cells. For any raster cell, the sum of the intersection areas of all building footprint polygons within that raster cell and the raster cell itself is calculated to quantify the actual spatial extent occupied by the building within that cell. Then, by analyzing the ratio of the sum of the intersection areas to the cell area, the building coverage rate is obtained. This building coverage rate is used as a correction coefficient for subsequent implicit carbon emission calculations, ensuring that the cell-scale carbon emission calculations reflect the actual building footprint proportions, avoiding the overestimation of carbon emissions by treating the entire raster cell area as building area, and improving the accuracy of spatialized calculations of implicit carbon emissions.

[0017] Further, for any raster cell within the target measurement space, the building's implicit carbon emissions are calculated based on the corresponding predicted building height, building coverage, building category, and building material category, thereby obtaining the building's implicit carbon emissions for each raster cell within the target measurement space, including: For any of the aforementioned raster cells: Under the given grid cell, obtain the unit volume material strength parameter of each building material category corresponding to the building category, and obtain the carbon emission factor of each building material category; For any building material category in the raster cell, the building-inherent carbon emissions of the building material category in the raster cell are calculated based on the cell area, building coverage, predicted building height, unit volume material strength parameter of the building material category, and carbon emission factor. The building-implicit carbon emissions of this raster cell are generated based on the building-implicit carbon emissions of all the aforementioned building material categories in this raster cell.

[0018] In this implementation, the unit volume material strength parameter corresponding to the building category for each building material category is obtained. A material consumption correlation is established between building categories and building material categories to reflect the differences in material usage across structural systems and building envelopes for different building types, avoiding calculation biases caused by using a uniform material consumption coefficient. Furthermore, the carbon emission factor for each building material category is obtained to incorporate the differences in carbon emission intensity during production and transportation stages of different building materials into the calculation process. For any building material category, the unit volume material strength parameter reflects the amount of that building material category consumed per unit building volume under that building category, while the carbon emission factor reflects the greenhouse gas emissions generated per unit amount of that building material category during the production process. Based on the pixel area, building coverage, predicted building height, material strength parameters per unit volume, and carbon emission factor of the raster pixel, the implicit carbon emissions of each building material category in the raster pixel are calculated item by item. The product of the pixel area and the building coverage represents the actual building area within the pixel, which is then multiplied by the predicted building height to obtain the building volume. The building volume is multiplied by the material strength parameters per unit volume to obtain the usage of the building material category, which is then multiplied by the carbon emission factor to obtain the implicit carbon emissions of the building material category in the raster pixel.

[0019] The hidden carbon emissions of a grid cell are generated based on the building material categories within that grid cell. This allows for the display of spatial differences and distribution patterns of hidden carbon emissions from building clusters within the target measurement area, using the grid cell as the smallest spatial unit, thus enabling the spatialized measurement of hidden carbon emissions from buildings.

[0020] Furthermore, the deep learning network for training building height estimation includes: Define the training space; Acquire spaceborne lidar data in the training space and convert the spaceborne lidar data into building height labels for each grid cell in the training space; Construct a deep learning network structure for estimating building height to be trained; The deep learning network for estimating building height is trained using the mean squared error function as the loss function and the building height label of each grid cell in the training space as the supervision data.

[0021] In this implementation, a training space is defined and spaceborne LiDAR data for that space is acquired. High-precision building height reference information is obtained using the active ranging principle of spaceborne LiDAR, thus providing supervised labels for deep learning network training. This overcomes the difficulty of obtaining true building height data in large-scale, cross-regional urban building cluster applications. The spaceborne LiDAR data is transformed into building height labels for each grid cell in the training space, ensuring that the label data's spatial organization is consistent with the network's input and output to facilitate supervised learning. A deep learning network structure for estimating building height is constructed and trained using the mean squared error function (MSE) as the loss function. The MSE quantifies the deviation between the predicted value and the label pixel-by-pixel, driving the network parameters to optimize towards minimizing the height prediction error. This enables the trained network to effectively learn the spatial distribution patterns of building heights from radar image features and optical remote sensing image features, gaining the ability to infer the spatial building height of targets.

[0022] Further, the step of acquiring spaceborne lidar data in the training space and converting the spaceborne lidar data into building height labels for each grid cell in the training space includes: Extract effective spot data from the spaceborne lidar data; Obtain the target percentile relative height of each effective light spot in the effective light spot data; wherein, the target percentile relative height represents the height corresponding to when the cumulative echo energy of the effective light spot from the ground reaches a preset target percentage; Establish the spatial correspondence between the effective light spot data and each grid cell in the training space; Identify the effective light spot contained in each grid cell within the training space based on the spatial correspondence; For any of the aforementioned raster cells: If the number of effective light spots in the raster cell is zero, then the building height label of the raster cell is assigned a value of zero; If the number of valid light spots in the raster cell is not zero, then the average relative height of the target percentile of all valid light spots in the raster cell is used as the building height label of the raster cell.

[0023] In this implementation, effective spot data is extracted from spaceborne lidar data to ensure that the laser ranging data involved in tag generation meets the requirements for building height characterization in terms of quality and spatial location. The target percentile relative height of each effective spot is obtained, and the height corresponding to the accumulated echo energy from the ground upwards reaching a preset target percentage is used as the characterization of building height. This reduces building height errors caused by antenna noise, vegetation scattering, or signal tails when using the highest echo height. A spatial correspondence is established between effective spot data and each grid cell in the training space, and the effective spots contained in each grid cell are identified based on this spatial correspondence. For grid cells with zero effective spots, the building height label is assigned to zero; for grid cells with non-zero effective spots, the average target percentile relative height of all effective spots is taken as the building height label. This reduces the impact of random errors in individual spot measurements and improves the accuracy of the label values.

[0024] Further, the extraction of effective spot data from the spaceborne lidar data includes: Extract the spaceborne lidar spot from the spaceborne lidar data, and then extract the data quality identifier, data degradation identifier, laser detection sensitivity parameter, terrain slope and laser spot center point for each spaceborne lidar spot. The spaceborne lidar spot that meets the following conditions is extracted as the effective spot data: The data quality identifier is equal to 1; The data degradation flag is equal to 0; The laser detection sensitivity parameter is greater than 0.9; The terrain slope is less than 10°; The center point of the laser spot is located within the outline of the building footprint vector data in the target measurement space.

[0025] In this implementation, a data quality flag of 1 is used as a filtering condition to exclude light spots with data recording errors or processing abnormalities; a data degradation flag of 0 is used as a filtering condition to exclude light spots whose ranging accuracy is reduced due to sensor performance degradation or signal saturation; a laser detection sensitivity parameter greater than 0.9 is used as a filtering condition to ensure that the selected light spots have sufficient signal-to-noise ratio and echo signal strength, guaranteeing the reliability of height measurement; a terrain slope of less than 10° is used as a filtering condition to avoid the impact of laser echo waveform distortion and ground elevation estimation deviation caused by steep terrain on building height calculation; and the center point of the laser light spot being located within the outline of the building footprint vector data is used as a filtering condition to ensure that the extracted effective light spots indeed correspond to the building target area, excluding light spots falling in non-building areas such as roads, vegetation, and water bodies, ensuring that the building height labels used for network training have high confidence.

[0026] Furthermore, the building footprint vector data is obtained based on the building survey vector map of the target measurement space or an open-source building footprint dataset. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the spatial calculation method for building hidden carbon emissions based on remote sensing data provided in this embodiment of the invention. Figure 2 This is a diagram of the deep learning network structure for building height estimation provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the building height label manufacturing method provided in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the technical route of the spatial calculation method for building hidden carbon emissions based on remote sensing data provided in this embodiment of the invention. Detailed Implementation

[0028] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that the following detailed descriptions are exemplary and intended to provide further detailed explanation of the present invention. 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 in the specification is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings are used to distinguish different objects, not to describe a particular order.

[0029] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0030] Building embodied carbon emissions refer to the greenhouse gas emissions generated throughout a building's entire life cycle, from raw material extraction, building material production, transportation to the construction site, construction and installation, to later maintenance, demolition, and disposal. The building material production stage is a significant component of building life-cycle carbon emissions. With the continuous advancement of urbanization, the expansion of built-up areas, and the sustained growth of building volume, there is an urgent need to conduct systematic accounting and assessment of building embodied carbon emissions to support related carbon emission monitoring, statistics, and management.

[0031] Current methods for calculating embodied carbon emissions from buildings largely rely on macroeconomic data such as statistical yearbooks, energy balance sheets, and input-output tables, combined with activity data such as completed building area and building material production. They use material flow analysis to calculate the total consumption of building materials in a specific region over a certain period, and then combine this with life cycle assessment methods or emission factor methods to calculate the total embodied carbon emissions from buildings using material emission factors. Finally, by spatially aggregating embodied carbon emission data from different administrative regions, results are obtained at the city, regional, or national scales. However, these statistical data-based methods cannot analyze the spatial distribution patterns of embodied carbon emissions from buildings across different blocks or plots within a city, and cannot directly reflect the continuous spatial variation characteristics of embodied carbon emissions from building clusters. Some studies have attempted to combine building outline vector data with building attribute ledgers for bottom-up calculations of embodied carbon emissions from buildings; however, these methods rely on detailed building survey data and structural design data, which are often difficult to obtain in large-scale, cross-regional urban building cluster applications.

[0032] Specifically, statistical data-based methods for calculating building-embedded carbon emissions suffer from data release and update delays, making it difficult to meet the need for real-time monitoring of embedded carbon emissions. Furthermore, some methods for estimating building-embedded carbon emissions typically rely on large-scale, high-precision building structural information, such as building height or volume, as key input parameters. This leads to high acquisition costs and limited spatial coverage, making it difficult to obtain such data in large-scale, cross-regional urban building clusters. Therefore, identifying the spatial distribution patterns of building-embedded carbon emissions within large-scale, cross-regional urban building clusters and analyzing the continuous spatial variation characteristics of embedded carbon emissions to provide guidance for spatial planning of urban building expansion and the formulation of carbon reduction strategies has become a pressing technical problem in this field.

[0033] This invention aims to provide a spatial method for calculating the implicit carbon emissions of buildings based on remote sensing data, in order to identify the spatial distribution patterns of implicit carbon emissions within large-scale, cross-regional urban building clusters, analyze the continuous spatial variation characteristics of implicit carbon emissions, and provide guidance for spatial planning of urban building expansion and the formulation of carbon reduction strategies.

[0034] Please refer to Figure 1 To achieve the above objectives, the first aspect of the present invention provides a method for spatially calculating the implicit carbon emissions of buildings based on remote sensing data, comprising the following steps: S1. Determine the target measurement space and the raster pixel resolution of the target measurement space.

[0035] In one specific embodiment, the grid cell resolution is 10 meters, consistent with the nominal spatial resolution of the optical remote sensing image.

[0036] S2. Obtain the radar image features and optical remote sensing image features of the target measurement space based on the grid pixel resolution.

[0037] S3. Train a deep learning network for building height estimation, and then enable the deep learning network for building height estimation to infer the predicted building height of each grid cell in the target measurement space through the radar image features and the optical remote sensing image features.

[0038] S4. Obtain the building footprint vector data of the target measurement space, and analyze the building coverage rate of each raster cell in the target measurement space based on the building footprint vector data.

[0039] S5. Obtain the building category and building material category of each raster cell in the target measurement space.

[0040] S6. For any raster cell within the target measurement space, calculate the building's implicit carbon emissions for that raster cell based on the corresponding predicted building height, building coverage, building category, and building material category, thereby obtaining the building's implicit carbon emissions for each raster cell within the target measurement space.

[0041] S7. Generate the spatial distribution characteristics of the building's hidden carbon emissions in the target measurement space based on the building's hidden carbon emissions for each grid cell.

[0042] The aforementioned method for spatializing building carbon emissions based on remote sensing data acquires radar image features and optical remote sensing image features that match the resolution of the raster pixels, and trains a deep learning network for building height estimation to infer the predicted building height value for each raster pixel. Without relying on detailed building survey data and structural design data, it identifies the distribution of building heights within a large-scale, cross-regional target measurement area for spatializing building carbon emissions.

[0043] Because a single pixel may contain multiple land cover types, such as buildings, roads, and green spaces, directly treating pixels with non-zero building heights as completely covered by buildings would lead to an overestimation of building area, thus affecting the accuracy of implicit carbon emission calculations. Therefore, building coverage is determined through spatial overlay analysis of building footprint vector data and raster pixels, incorporating the actual building footprint into the calculation process to avoid overestimating the carbon emissions of raster pixels. Finally, considering the differences in material consumption intensity and carbon emission factors among different building types and building materials, the building category and building material category of each raster pixel are obtained. Based on this, combined with the predicted building height and building coverage, the implicit carbon emissions of each raster pixel are calculated. Based on the implicit carbon emissions of each raster pixel, the spatial distribution characteristics of implicit carbon emissions in the target calculation space are generated. This allows the raster pixel to be used as the smallest spatial unit to display the spatial differences and distribution patterns of implicit carbon emissions of building clusters within the target calculation area, providing guidance for spatial planning of urban building expansion and the formulation of carbon reduction strategies.

[0044] Further, the step of enabling the deep learning network for building height estimation to infer the predicted building height of each raster cell within the target measurement space using the radar image features and the optical remote sensing image features includes: The radar image features are input into the radar image feature encoder in the building height estimation deep learning network, and the optical remote sensing image features are input into the optical remote sensing image feature encoder in the building height estimation deep learning network; wherein, the radar image feature encoder and the optical remote sensing image feature encoder have the same structure and their parameters are independent of each other; The radar image feature encoder extracts the radar image feature multi-scale deep features of the radar image, and the optical remote sensing image feature encoder extracts the optical remote sensing image feature multi-scale deep features of the optical remote sensing image. Using the decoder of the deep learning network for building height estimation, the spatial resolution of the multi-scale deep features of the radar image and the multi-scale deep features of the optical remote sensing image is restored by upsampling in a stepwise manner, so as to obtain the multi-scale decoding features of the radar image and the multi-scale decoding features of the optical remote sensing image. The multi-scale deep features of the radar image, the multi-scale decoding features of the radar image, the multi-scale deep features of the optical remote sensing image, and the multi-scale decoding features of the optical remote sensing image are fused in a skip connection manner to obtain fused features; Based on the fused features, the predicted building height of each raster cell in the target measurement space is inferred.

[0045] In this implementation, the deep learning network for building height estimation employs a dual-encoder structure to process radar image features and optical remote sensing image features separately. The radar image feature encoder and the optical remote sensing image feature encoder have identical structures and independent parameters, allowing features from both modalities to be encoded in their respective independent parameter spaces. This avoids mutual interference from differences in the distribution of features from different sources, while structural consistency ensures alignment of encoding levels, which is beneficial for subsequent feature fusion. By encoding features from different remote sensing modalities separately, the impact of different imaging mechanisms and data distribution differences on the feature learning process can be reduced, thereby improving the stability and accuracy of building height feature representation. In the decoding stage, the decoder restores the spatial resolution of multi-scale deep features through progressive upsampling. While restoring spatial details, it gradually integrates high-level semantic information, ensuring that the output feature has the same spatial resolution as the input raster pixel resolution. A skip connection is used to fuse the multi-scale deep features output by the encoder with the multi-scale decoded features output by the decoder, introducing low-level spatial detail information while preserving high-level semantic information, compensating for high-frequency spatial detail information that may be lost during progressive upsampling. By inferring the building height prediction value for each raster cell based on fusion features, the building height estimation can simultaneously utilize the ability of radar imagery to express the three-dimensional geometry of the building and the ability of optical remote sensing imagery to express the surface texture of the building, thereby improving the accuracy of building height prediction.

[0046] like Figure 2 As shown, the deep learning network for building height estimation includes two feature encoders and one decoder. The feature encoder in the upper branch receives optical features, and the feature encoder in the lower branch receives radar features; wherein, Figure 2 The optical features mentioned are the optical remote sensing image features, the radar features are the radar image features, the feature encoder of the upper branch is the optical remote sensing image feature encoder, and the feature encoder of the lower branch is the radar image feature encoder. The two feature encoders have the same structure and their parameters are independent of each other. Figure 2 The diagram shows feature stitching operations performed along the channel dimension, and illustrates each decoding module with stacked convolutional layers, batch normalization layers, and upsampling layers. Each decoding module sequentially includes a convolutional layer, a batch normalization layer, and an upsampling layer. After progressive upsampling, the spatial resolution of the feature map is restored and the feature map size is increased progressively. The deep features at each scale output by the feature encoder are introduced into the corresponding scale decoding module in the decoder for stitching and fusion using a skip connection method. The decoder outputs a building height map, including the predicted building height value for each raster cell.

[0047] More specifically, the deep learning network for building height estimation takes optical remote sensing image features and radar image features as input, and achieves multi-source feature fusion through an encoder-decoder structure to output predicted building height values. In the encoding stage, optical remote sensing image features and radar image features are respectively input into two feature encoders with identical structures and independent parameters to extract deep features at different scales. By encoding features for different remote sensing modalities separately, the impact of different imaging mechanisms and data distribution differences on the feature learning process can be reduced, thereby improving the stability and accuracy of building height feature representation. In the decoding stage, the deep learning network for building height estimation uses a step-by-step upsampling method to restore the spatial resolution of the feature map, and fuses the output features of each scale in the encoder with the corresponding scale features in the decoder through skip connections. The fusion method involves concatenating features along the channel dimension to introduce low-level spatial detail information while preserving high-level semantic information. Each decoding module includes a convolutional layer, a batch normalization layer, and an upsampling layer, used for further modeling and reconstruction of the fused features. In the output stage, the deep learning network for building height estimation generates predicted building height values ​​in the form of pixel-level continuous values. To address the regression characteristics of the building height estimation task, the deep learning network for building height estimation uses the mean squared error function as the loss function during training. The formula for the loss function is as follows: In the formula, Indicates the number of samples. Indicates the first Mid-year The true value of each sample Indicates the first Mid-year The predicted value for each sample. Specifically, spaceborne LiDAR data from the training space is used to convert the spaceborne LiDAR data into a building height label for each grid cell in the training space, which serves as the true value for the sample.

[0048] Further, the step of obtaining the radar image features and optical remote sensing image features of the target measurement space based on the grid pixel resolution includes: Obtain the ground distance detection data of the target measurement space, and classify the ground distance detection data into ascending orbit images and descending orbit images; The number of ascending and descending orbit images is counted. If the number of ascending orbit images is greater than the number of descending orbit images, the ascending orbit images are used as unified orbit images. If the number of ascending orbit images is less than the number of descending orbit images, the descending orbit images are used as unified orbit images. If the number of ascending orbit images is equal to the number of descending orbit images, either the ascending orbit image or the descending orbit image is used as a unified orbit image. Extract the VV polarization (vertical transmit-vertical receive polarization) backscattering coefficient image and the VH polarization (vertical transmit-horizontal receive polarization) backscattering coefficient image within a preset time window from the unified orbit image; The VV polarization backscattering coefficient images within a preset time window are synthesized using the median synthesis method to obtain a synthesized VV polarization backscattering coefficient image. The VH polarization backscattering coefficient images within a preset time window are synthesized using the median synthesis method to obtain synthesized VH polarization backscattering coefficient images. The synthesized VV polarization backscattering coefficient image and the synthesized VH polarization backscattering coefficient image are used as radar image features.

[0049] In this implementation, ground distance detection data is classified into ascending orbit images and descending orbit images. A unified orbit image is selected based on quantity advantage, ensuring the consistency of radar imagery in observation geometry. This avoids the problem of incomparable backscattering coefficients caused by differences in incident angles across different orbital directions when acquiring satellite data, providing a unified observation benchmark for subsequent time-series synthesis. VV-polarized and VH-polarized backscattering coefficient images are extracted from the unified orbit imagery. VV-polarization is sensitive to the dihedral reflection and volume scattering characteristics of building structures, while VH-polarization is sensitive to the volume scattering and multiple scattering characteristics of building complexes. Using both together can more comprehensively characterize the backscattering coefficient image features of building targets. A median synthesis method is used to synthesize multiple images within a preset time window. Utilizing the median statistic's insensitivity to outliers, abnormal backscattering values ​​caused by instantaneous weather conditions and sensor noise are reduced, thereby improving the accuracy of the synthesized radar image features.

[0050] In one specific embodiment, the Sentinel-1 imaging satellite is used to acquire ground distance detection data of the target's measurement space.

[0051] It should be noted that electromagnetic wave transmission is divided into horizontal wave (H) and vertical wave (V), and the receiving method is also divided into horizontal wave (H) and vertical wave (V). When an imaging satellite transmits signals to the ground, the vibration direction of the radio electromagnetic waves used can be varied. In this invention, VV polarization (vertical transmission – vertical reception polarization) backscattering coefficient images and VH polarization (vertical transmission – horizontal reception polarization) backscattering coefficient images are used.

[0052] Further, the step of obtaining the radar image features and optical remote sensing image features of the target measurement space based on the grid pixel resolution includes: Acquire the surface reflectance data of the target measurement space, the surface reflectance data including visible light band, near-infrared band, short-wave infrared band and image quality control band; According to the image quality control band, mark the invalid and valid pixels in all the raster pixels in the target measurement space; A binary cloud mask is constructed based on the invalid pixels and the valid pixels. The visible light band, the near-infrared band and the short-wave infrared band are processed using the binary cloud mask to obtain the visible light band, the near-infrared band and the short-wave infrared band of the cloud mask. A number of cloud mask visible light bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the cloud mask visible light bands to obtain the synthesized cloud mask visible light bands. Several cloud mask near-infrared bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the several cloud mask near-infrared bands to obtain the synthesized cloud mask near-infrared bands. Several cloud mask shortwave infrared bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the several cloud mask shortwave infrared bands to obtain the synthesized cloud mask shortwave infrared bands. The visible light band, near-infrared band, and short-wave infrared band of the synthetic cloud mask are unified to the grid pixel resolution as the optical remote sensing image feature.

[0053] In one specific embodiment, the Sentinel-2 imaging satellite is used to acquire surface reflectance data of the target measurement space.

[0054] The image quality control band is a bit-coded band, where the 10th bit is used to identify clouds and the 11th bit is used to identify cirrus clouds. When the 10th or 11th bit is 1, the corresponding raster cell is determined to be an invalid cell affected by clouds or cirrus clouds, and the remaining raster cells are marked as valid cells. Based on this, a binary cloud mask is constructed, where invalid cells affected by clouds or cirrus clouds are assigned a value of 0, and valid cells are assigned a value of 1. This mask is applied to each spectral band to remove cells affected by clouds and cloud contamination.

[0055] It should be noted that the visible light band includes the blue band, green band, and red band.

[0056] In this implementation, surface reflectance data including visible light, near-infrared, short-wave infrared, and image quality control bands are acquired. Invalid and valid pixels are marked using the image quality control bands, and a binary cloud mask is constructed to process the visible light, near-infrared, and short-wave infrared bands. Pixels contaminated by clouds and cloud shadows are removed from the surface reflectance data, avoiding interference from cloud contamination and ensuring that all spectral data used in subsequent synthesis are valid surface observations. The median synthesis method is used to statistically synthesize the spectral bands of multiple cloud mask images within a preset time window. This eliminates the influence of residual cloud contamination while reducing the impact of outliers caused by atmospheric condition fluctuations and sensor radiometric calibration deviations, resulting in time-stable synthesized spectral characteristics. By unifying the visible light band, near-infrared band, and short-wave infrared band of the synthetic cloud mask to the grid pixel resolution as optical remote sensing image features, the consistency of spatial resolution of each band is ensured, and the consistency of spatial resolution between optical remote sensing image features and radar image features is also ensured. This provides spatially aligned data conditions for the parallel input of the dual encoder network and subsequent feature fusion.

[0057] In one specific embodiment, in order to unify the surface reflectance data of each band to the grid pixel resolution as an optical remote sensing image feature and ensure the consistency of spatial resolution of each band, a bilinear interpolation method is used to resample the shortwave infrared band to the grid pixel resolution, so as to ensure the consistency of spatial resolution of the visible light band, the near-infrared band and the shortwave infrared band.

[0058] Further, the step of acquiring the building footprint vector data of the target measurement space, and analyzing the building coverage of each raster cell within the target measurement space based on the building footprint vector data, includes: The building footprint vector data is spatially overlaid with each raster cell in the target measurement space to determine the building footprint polygon contained within the range of each raster cell; For any of the aforementioned raster cells: Calculate the total area of ​​intersection between the building footprint polygon contained within the raster cell and the raster cell; The cell area of ​​the raster cell is obtained, and the ratio of the sum of the intersecting areas to the cell area is analyzed to obtain the building coverage of the raster cell.

[0059] Specifically, the building coverage Calculate using the following formula: In the formula, This represents a 10-meter resolution raster pixel. This represents the pixel area of ​​the raster cell. This represents the polygon containing the building footprint within a cell.

[0060] In this implementation, the building footprint vector data is spatially overlaid with each raster cell within the target measurement space to determine the building footprint polygons contained within each raster cell, establishing a spatial relationship between the vector building outline and the raster cells. For any raster cell, the sum of the intersection areas of all building footprint polygons within that raster cell and the raster cell itself is calculated to quantify the actual spatial extent occupied by the building within that cell. Then, by analyzing the ratio of the sum of the intersection areas to the cell area, the building coverage rate is obtained. This building coverage rate is used as a correction coefficient for subsequent implicit carbon emission calculations, ensuring that the cell-scale carbon emission calculations reflect the actual building footprint proportions, avoiding the overestimation of carbon emissions by treating the entire raster cell area as building area, and improving the accuracy of spatialized calculations of implicit carbon emissions.

[0061] Further, for any raster cell within the target measurement space, the building's implicit carbon emissions are calculated based on the corresponding predicted building height, building coverage, building category, and building material category, thereby obtaining the building's implicit carbon emissions for each raster cell within the target measurement space, including: For any of the aforementioned raster cells: Under the given grid cell, obtain the unit volume material strength parameter of each building material category corresponding to the building category, and obtain the carbon emission factor of each building material category; For any building material category in the raster cell, the building-inherent carbon emissions of the building material category in the raster cell are calculated based on the cell area, building coverage, predicted building height, unit volume material strength parameter of the building material category, and carbon emission factor. The building-implicit carbon emissions of this raster cell are generated based on the building-implicit carbon emissions of all the aforementioned building material categories in this raster cell.

[0062] Specifically, the building's implicit carbon emissions are calculated using the following formula: in, This indicates the building coverage of the raster cell. This represents the predicted building height. Indicates the conversion factor. This represents the material strength parameter per unit area (unit: kg / m²) for building material category i within building category j; for any given pixel, there is one building category j and multiple materials i, such as steel, concrete, and wood; n represents the number of building material categories for this raster pixel. Indicates building material category Carbon emission factors.

[0063] Preferably, the conversion factor The value is 4.37, which is used to convert the material strength parameter per unit area into the material strength parameter per unit volume. Based on existing research, 4.37 is the preferred value.

[0064] In this implementation, the unit volume material strength parameter corresponding to the building category for each building material category is obtained. A material consumption correlation is established between building categories and building material categories to reflect the differences in material usage across structural systems and building envelopes for different building types, avoiding calculation biases caused by using a uniform material consumption coefficient. Furthermore, the carbon emission factor for each building material category is obtained to incorporate the differences in carbon emission intensity during production and transportation stages of different building materials into the calculation process. For any building material category, the unit volume material strength parameter reflects the amount of that building material category consumed per unit building volume under that building category, while the carbon emission factor reflects the greenhouse gas emissions generated per unit amount of that building material category during the production process. Based on the pixel area, building coverage, predicted building height, material strength parameters per unit volume, and carbon emission factor of the raster pixel, the implicit carbon emissions of each building material category in the raster pixel are calculated item by item. The product of the pixel area and the building coverage represents the actual building area within the pixel, which is then multiplied by the predicted building height to obtain the building volume. The building volume is multiplied by the material strength parameters per unit volume to obtain the usage of the building material category, which is then multiplied by the carbon emission factor to obtain the implicit carbon emissions of the building material category in the raster pixel.

[0065] The hidden carbon emissions of a grid cell are generated based on the building material categories within that grid cell. This allows for the display of spatial differences and distribution patterns of hidden carbon emissions from building clusters within the target measurement area, using the grid cell as the smallest spatial unit, thus enabling the spatialized measurement of hidden carbon emissions from buildings.

[0066] Furthermore, the deep learning network for training building height estimation includes: Define the training space; Acquire spaceborne lidar data in the training space and convert the spaceborne lidar data into building height labels for each grid cell in the training space; Construct a deep learning network structure for estimating building height to be trained; The deep learning network for estimating building height is trained using the mean squared error function as the loss function and the building height label of each grid cell in the training space as the supervision data.

[0067] In this implementation, a training space is defined and spaceborne LiDAR data for that space is acquired. High-precision building height reference information is obtained using the active ranging principle of spaceborne LiDAR, thus providing supervised labels for deep learning network training. This overcomes the difficulty of obtaining true building height data in large-scale, cross-regional urban building cluster applications. The spaceborne LiDAR data is transformed into building height labels for each grid cell in the training space, ensuring that the label data's spatial organization is consistent with the network's input and output to facilitate supervised learning. A deep learning network structure for estimating building height is constructed and trained using the mean squared error function (MSE) as the loss function. The MSE quantifies the deviation between the predicted value and the label pixel-by-pixel, driving the network parameters to optimize towards minimizing the height prediction error. This enables the trained network to effectively learn the spatial distribution patterns of building heights from radar image features and optical remote sensing image features, gaining the ability to infer the spatial building height of targets.

[0068] Further, the step of acquiring spaceborne lidar data in the training space and converting the spaceborne lidar data into building height labels for each grid cell in the training space includes: Extract effective spot data from the spaceborne lidar data; Obtain the target percentile relative height of each effective light spot in the effective light spot data; wherein, the target percentile relative height represents the height corresponding to when the cumulative echo energy of the effective light spot from the ground reaches a preset target percentage; Establish the spatial correspondence between the effective light spot data and each grid cell in the training space; Identify the effective light spot contained in each grid cell within the training space based on the spatial correspondence; For any of the aforementioned raster cells: If the number of effective light spots in the raster cell is zero, then the building height label of the raster cell is assigned a value of zero; If the number of valid light spots in the raster cell is not zero, then the average relative height of the target percentile of all valid light spots in the raster cell is used as the building height label of the raster cell.

[0069] In this implementation, effective spot data is extracted from spaceborne lidar data to ensure that the laser ranging data involved in tag generation meets the requirements for building height characterization in terms of quality and spatial location. The target percentile relative height of each effective spot is obtained, and the height corresponding to the accumulated echo energy from the ground upwards reaching a preset target percentage is used as the characterization of building height. This reduces building height errors caused by antenna noise, vegetation scattering, or signal tails when using the highest echo height. A spatial correspondence is established between effective spot data and each grid cell in the training space, and the effective spots contained in each grid cell are identified based on this spatial correspondence. For grid cells with zero effective spots, the building height label is assigned to zero; for grid cells with non-zero effective spots, the average target percentile relative height of all effective spots is taken as the building height label. This reduces the impact of random errors in individual spot measurements and improves the accuracy of the label values.

[0070] like Figure 3 As shown, in a preferred embodiment, the preset target percentage is 96%, that is, the target percentile relative height is the 96th percentile relative height, which represents the height corresponding to when the effective light spot accumulates 96% of the echo energy from the ground upwards, so as to reduce the building height error caused by antenna noise, vegetation scattering or signal tail when using the highest echo height.

[0071] Using the grid pixels of Sentinel-2 imagery as the training spatial reference, the center points of the spaceborne lidar spots are superimposed onto the corresponding grid pixels to establish a spatial correspondence between the spaceborne lidar spots and the grid pixels. For a grid pixel containing a single spaceborne lidar spot, the 96th percentile relative height value of that spot is taken as the reference building height. For a grid pixel containing multiple spaceborne lidar spots, the average of the 96th percentile relative height values ​​of all spots is taken as the reference building height. For areas without spaceborne lidar spots, the building height label for that grid pixel is assigned a value of 0. Based on the reference building heights of each grid pixel and the grid pixels with a value of 0, a 10-meter resolution building height label is created. Figure 3 The reference building height for a cell mentioned above is the building height label for the corresponding raster cell.

[0072] Further, the extraction of effective spot data from the spaceborne lidar data includes: Extract the spaceborne lidar spot from the spaceborne lidar data, and then extract the data quality identifier, data degradation identifier, laser detection sensitivity parameter, terrain slope and laser spot center point for each spaceborne lidar spot. The spaceborne lidar spot that meets the following conditions is extracted as the effective spot data: The data quality identifier is equal to 1; The data degradation flag is equal to 0; The laser detection sensitivity parameter is greater than 0.9; The terrain slope is less than 10°; The center point of the laser spot is located within the outline of the building footprint vector data in the target measurement space.

[0073] In this implementation, a data quality flag of 1 is used as a filtering condition to exclude light spots with data recording errors or processing abnormalities; a data degradation flag of 0 is used as a filtering condition to exclude light spots whose ranging accuracy is reduced due to sensor performance degradation or signal saturation; a laser detection sensitivity parameter greater than 0.9 is used as a filtering condition to ensure that the selected light spots have sufficient signal-to-noise ratio and echo signal strength, guaranteeing the reliability of height measurement; a terrain slope of less than 10° is used as a filtering condition to avoid the impact of laser echo waveform distortion and ground elevation estimation deviation caused by steep terrain on building height calculation; and the center point of the laser light spot being located within the outline of the building footprint vector data is used as a filtering condition to ensure that the extracted effective light spots indeed correspond to the building target area, excluding light spots falling in non-building areas such as roads, vegetation, and water bodies, ensuring that the building height labels used for network training have high confidence.

[0074] Furthermore, the building footprint vector data is obtained based on the building survey vector map of the target measurement space or an open-source building footprint dataset.

[0075] like Figure 4 As shown, in one specific embodiment, the above-mentioned spatial calculation method for building-hidden carbon emissions based on remote sensing data includes: After filtering the data quality of the spaceborne lidar data, the 96th percentile relative height was extracted, and a 10-meter resolution building height label was created. The resulting 10-meter resolution building height label was used as the label for training the deep learning network for building height estimation.

[0076] Radar features were obtained by preprocessing the Sentinel-1 ground distance detection data, and optical features were obtained by preprocessing the Sentinel-2 surface reflectance data. The Sentinel-1 ground distance detection data refers to the ground distance detection data acquired using the Sentinel-1 imaging satellite, and the Sentinel-2 surface reflectance data refers to the surface reflectance data acquired using the Sentinel-2 imaging satellite. Figure 4 The radar features mentioned above are the radar image features, and the optical features are the optical remote sensing image features.

[0077] The deep learning network for building height estimation outputs a large-scale, spatially continuous building height based on the input radar and optical features. The large-scale, spatially continuous building height is composed of the predicted building height value of each grid cell within the target measurement space. Figure 4 The 10-meter resolution building coverage rate is the building coverage rate, and the individual building footprint dataset is used to obtain the building footprint vector data. Figure 4 The building type obtained by distinguishing building types is the building category, the building material strength is the unit volume material strength parameter, the building material carbon emission factor is the carbon emission factor, and the building material parameters include the building material strength and the building material carbon emission factor.

[0078] The emission factor model is used to calculate the building height and the 10-meter resolution building coverage in the large-scale space continuously by grid-by-grid pixel, thereby obtaining the building-hidden carbon in the large-scale space continuously. The building-hidden carbon in the large-scale space continuously is composed of the building-hidden carbon emissions of each grid cell in the target measurement space, that is, the spatial distribution characteristics of building-hidden carbon emissions in the target measurement space are generated.

[0079] The spatial calculation method for building-hidden carbon emissions based on remote sensing data described in this invention has at least the following advantages compared with existing methods for estimating building-hidden carbon emissions based on statistical data: First, it enables the identification of the spatial distribution patterns of building-hidden carbon emissions within large-scale, cross-regional urban building clusters and analyzes the continuous spatial variation characteristics of hidden carbon emissions. Existing technologies mostly rely on statistical data at the administrative district scale or empirical parameters of building categories to calculate hidden carbon emissions, which is difficult to reflect the continuous spatial distribution characteristics of building-hidden carbon emissions. This invention, by introducing predicted building height, building coverage, and building material parameters into the raster pixel scale calculation, realizes the continuous spatial distribution calculation of building-hidden carbon emissions within a large-scale area and improves the spatial resolution of the calculation results. Second, the technical solution described in this invention is based on a unified rasterized data structure and parameterized calculation process, which can carry out automated spatial calculation of building-hidden carbon emissions for large-scale areas, and has good scalability and engineering application value.

[0080] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0081] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described; however, any combination of these technical features that does not contradict each other should be considered within the scope of this specification.

[0082] The embodiments described above are merely illustrative of several implementation methods 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 improvements and substitutions without departing from the concept of this application, and these improvements and substitutions should also be considered within the scope of protection of this invention. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for spatially calculating the implicit carbon emissions of buildings based on remote sensing data, characterized in that, include: Determine the target measurement space and the raster pixel resolution of the target measurement space; The radar image features and optical remote sensing image features of the target measurement space are obtained based on the grid pixel resolution. A deep learning network for building height estimation is trained, which then infers the predicted building height of each grid cell in the target measurement space based on the radar image features and the optical remote sensing image features. Obtain building footprint vector data of the target measurement space, and analyze the building coverage rate of each raster cell in the target measurement space based on the building footprint vector data; Obtain the building category and building material category of each raster cell within the target measurement space; For any grid cell within the target measurement space, the building's implicit carbon emissions are calculated based on the corresponding predicted building height, building coverage, building category, and building material category, thereby obtaining the building's implicit carbon emissions for each grid cell within the target measurement space. The spatial distribution characteristics of the building's hidden carbon emissions in the target measurement space are generated based on the building's hidden carbon emissions for each raster cell.

2. The method for spatially calculating building-embedded carbon emissions based on remote sensing data according to claim 1, characterized in that, The process of enabling the deep learning network for building height estimation to infer the predicted building height of each raster cell within the target measurement space using radar image features and optical remote sensing image features includes: The radar image features are input into the radar image feature encoder in the building height estimation deep learning network, and the optical remote sensing image features are input into the optical remote sensing image feature encoder in the building height estimation deep learning network; wherein, the radar image feature encoder and the optical remote sensing image feature encoder have the same structure and their parameters are independent of each other; The radar image feature encoder extracts the radar image feature multi-scale deep features of the radar image, and the optical remote sensing image feature encoder extracts the optical remote sensing image feature multi-scale deep features of the optical remote sensing image. Using the decoder of the deep learning network for building height estimation, the spatial resolution of the multi-scale deep features of the radar image and the multi-scale deep features of the optical remote sensing image is restored by upsampling in a stepwise manner, so as to obtain the multi-scale decoding features of the radar image and the multi-scale decoding features of the optical remote sensing image. The multi-scale deep features of the radar image, the multi-scale decoding features of the radar image, the multi-scale deep features of the optical remote sensing image, and the multi-scale decoding features of the optical remote sensing image are fused in a skip connection manner to obtain fused features; Based on the fused features, the predicted building height of each raster cell in the target measurement space is inferred.

3. The spatial calculation method for building-hidden carbon emissions based on remote sensing data according to claim 1, characterized in that, The step of obtaining the radar image features and optical remote sensing image features of the target measurement space based on the grid pixel resolution includes: Obtain the ground distance detection data of the target measurement space, and classify the ground distance detection data into ascending orbit images and descending orbit images; The number of ascending and descending orbit images is counted. If the number of ascending orbit images is greater than the number of descending orbit images, the ascending orbit images are used as unified orbit images. If the number of ascending orbit images is less than the number of descending orbit images, the descending orbit images are used as unified orbit images. If the number of ascending orbit images is equal to the number of descending orbit images, either the ascending orbit image or the descending orbit image is used as a unified orbit image. Extract the VV polarization backscattering coefficient image and the VH polarization backscattering coefficient image within a preset time window from the unified orbit image; The VV polarization backscattering coefficient images within a preset time window are synthesized using the median synthesis method to obtain a synthesized VV polarization backscattering coefficient image. The VH polarization backscattering coefficient images within a preset time window are synthesized using the median synthesis method to obtain synthesized VH polarization backscattering coefficient images. The synthesized VV polarization backscattering coefficient image and the synthesized VH polarization backscattering coefficient image are used as radar image features.

4. The spatial calculation method for building-embedded carbon emissions based on remote sensing data according to claim 1, characterized in that, The step of obtaining the radar image features and optical remote sensing image features of the target measurement space based on the grid pixel resolution includes: Acquire the surface reflectance data of the target measurement space, the surface reflectance data including visible light band, near-infrared band, short-wave infrared band and image quality control band; According to the image quality control band, mark the invalid and valid pixels in all the raster pixels in the target measurement space; A binary cloud mask is constructed based on the invalid pixels and the valid pixels. The visible light band, the near-infrared band and the short-wave infrared band are processed using the binary cloud mask to obtain the visible light band, the near-infrared band and the short-wave infrared band of the cloud mask. A number of cloud mask visible light bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the cloud mask visible light bands to obtain the synthesized cloud mask visible light bands. Several cloud mask near-infrared bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the several cloud mask near-infrared bands to obtain the synthesized cloud mask near-infrared bands. Several cloud mask shortwave infrared bands within a preset time window are obtained, and the median synthesis method is used to statistically synthesize the several cloud mask shortwave infrared bands to obtain the synthesized cloud mask shortwave infrared bands. The visible light band, near-infrared band, and short-wave infrared band of the synthetic cloud mask are unified to the grid pixel resolution as the optical remote sensing image feature.

5. The method for spatially calculating building-embedded carbon emissions based on remote sensing data according to claim 1, characterized in that, The step of acquiring the building footprint vector data of the target measurement space and analyzing the building coverage of each raster cell within the target measurement space based on the building footprint vector data includes: The building footprint vector data is spatially overlaid with each raster cell in the target measurement space to determine the building footprint polygon contained within the range of each raster cell; For any of the aforementioned raster cells: Calculate the total area of ​​intersection between the building footprint polygon contained within the raster cell and the raster cell; The cell area of ​​the raster cell is obtained, and the ratio of the sum of the intersecting areas to the cell area is analyzed to obtain the building coverage of the raster cell.

6. The spatial calculation method for building-embedded carbon emissions based on remote sensing data according to claim 5, characterized in that, For any raster pixel within the target measurement space, the building's implicit carbon emissions are calculated based on the corresponding predicted building height, building coverage, building category, and building material category, thereby obtaining the building's implicit carbon emissions for each raster pixel within the target measurement space, including: For any of the aforementioned raster cells: Under the given grid cell, obtain the unit volume material strength parameter of each building material category corresponding to the building category, and obtain the carbon emission factor of each building material category; For any building material category in the raster cell, the building-inherent carbon emissions of the building material category in the raster cell are calculated based on the cell area, building coverage, predicted building height, unit volume material strength parameter of the building material category, and carbon emission factor. The building-implicit carbon emissions of this raster cell are generated based on the building-implicit carbon emissions of all the aforementioned building material categories in this raster cell.

7. The method for spatially calculating the implicit carbon emissions of buildings based on remote sensing data according to claim 1, characterized in that, The training deep learning network for building height estimation includes: Define the training space; Acquire spaceborne lidar data in the training space and convert the spaceborne lidar data into building height labels for each grid cell in the training space; Construct a deep learning network structure for estimating building height to be trained; The deep learning network for estimating building height is trained using the mean squared error function as the loss function and the building height label of each grid cell in the training space as the supervision data.

8. The spatial calculation method for building-embedded carbon emissions based on remote sensing data according to claim 7, characterized in that, The step of acquiring spaceborne lidar data in the training space and converting the spaceborne lidar data into building height labels for each grid cell in the training space includes: Extract effective spot data from the spaceborne lidar data; Obtain the target percentile relative height of each effective light spot in the effective light spot data; wherein, the target percentile relative height represents the height corresponding to when the cumulative echo energy of the effective light spot from the ground reaches a preset target percentage; Establish the spatial correspondence between the effective light spot data and each grid cell in the training space; Identify the effective light spot contained in each grid cell within the training space based on the spatial correspondence; For any of the aforementioned raster cells: If the number of effective light spots in the raster cell is zero, then the building height label of the raster cell is assigned a value of zero; If the number of valid light spots in the raster cell is not zero, then the average relative height of the target percentile of all valid light spots in the raster cell is used as the building height label of the raster cell.

9. The spatial calculation method for building-embedded carbon emissions based on remote sensing data according to claim 8, characterized in that, The extraction of effective spot data from the spaceborne lidar data includes: Extract the spaceborne lidar spot from the spaceborne lidar data, and then extract the data quality identifier, data degradation identifier, laser detection sensitivity parameter, terrain slope and laser spot center point for each spaceborne lidar spot. The spaceborne lidar spot that meets the following conditions is extracted as the effective spot data: The data quality identifier is equal to 1; The data degradation flag is equal to 0; The laser detection sensitivity parameter is greater than 0.9; The terrain slope is less than 10°; The center point of the laser spot is located within the outline of the building footprint vector data in the target measurement space.

10. The method for spatially calculating the implicit carbon emissions of buildings based on remote sensing data according to claim 1, characterized in that, The building footprint vector data is obtained based on the building survey vector map of the target measurement space or an open-source building footprint dataset.

Images