A method and system for measuring urban green space carbon sequestration based on multi-source data fusion

Abstract

The present invention discloses a method and system for measuring urban green space carbon sinks based on multi-source data fusion, relating to the field of environmental monitoring technology. The method comprises the following steps: collecting and preprocessing multi-source data from urban green spaces; using a dual-channel generative adversarial network to complete the parameters of vegetation in building-blocked blind areas, obtaining complete green space leaf area index distribution data; using a ray tracing algorithm to simulate the path of light reflected from building glass curtain walls, obtaining a correction factor for photosynthetically active radiation received by the vegetation canopy, and identifying urban green space carbon sink distribution data through multi-scale data fusion; obtaining carbon sink errors based on the urban green space carbon sink distribution data, dynamically correcting them using a Bayesian optimization algorithm, and generating an urban green space carbon sink measurement report. The present invention utilizes a dual-channel generative adversarial network, utilizing a collaborative training mechanism between a U-Net generator and a PatchGAN discriminator, to achieve high-precision spatial continuity reconstruction of the leaf area index in building-blocked blind areas.

Description

Technical Field

[0001] The present invention relates to the field of environmental monitoring technology, and in particular to a method and system for measuring urban green space carbon sequestration based on multi-source data fusion. Background Art

[0002] Measuring carbon sinks in urban green spaces is a key technical area for urban ecological monitoring and carbon neutrality assessment. Current mainstream approaches are primarily based on a technical approach combining multi-source remote sensing data fusion with ground-based observations. These include using high-resolution remote sensing imagery to invert vegetation parameters and establishing carbon sink estimation models based on CO2 flux observation data from flux towers. For areas blocked by buildings, existing technologies often use the sun path projection method to determine the shadow range, combined with morphological image processing to extract effective vegetation areas. Regarding dynamic weight allocation, methods based on time series analysis and spatial clustering are widely used to model the spatiotemporal characteristics of carbon sinks, and some studies have introduced machine learning algorithms to improve parameter inversion accuracy.

[0003] Conventional methods have room for improvement in the following two aspects: First, existing weight allocation methods mostly use static or semi-static models, which cannot fully characterize the spatiotemporal heterogeneity of urban vegetation carbon sinks, especially the dynamic response under the influence of seasonal changes and urban microclimate; second, existing carbon sink measurement systems rarely involve the enhancing effect of reflected light from building glass curtain walls on vegetation photosynthesis, resulting in systematic deviations in the estimation of canopy photosynthetic active radiation. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an urban green space carbon sink measurement method based on multi-source data fusion to solve the problems of insufficient spatiotemporal dynamic weight allocation and imperfect building reflected light effect modeling in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a method for measuring urban green space carbon sinks based on multi-source data fusion, which includes collecting multi-source data of urban green spaces and performing preprocessing; using a dual-channel generative adversarial network to complete the parameters of vegetation in blind areas blocked by buildings, and obtaining complete green space leaf area index distribution data; based on the complete green space leaf area index distribution data, using a spatiotemporal perception dynamic weighted network to perform dynamic weight allocation, and generate a carbon sink data fusion weight matrix; using a ray tracing algorithm to simulate the light reflection path of the building glass curtain wall, obtain the photosynthetic active radiation correction coefficient received by the vegetation canopy, and identify the urban green space carbon sink distribution data through multi-scale data fusion; obtaining the carbon sink error based on the urban green space carbon sink distribution data, and dynamically correcting it through a Bayesian optimization algorithm to generate an urban green space carbon sink measurement report.

[0008] As a preferred solution of the urban green space carbon sequestration measurement method based on multi-source data fusion described in the present invention, the dual-channel generative adversarial network is used to complete the parameters of vegetation in building blind areas to obtain complete green space leaf area index distribution data. The steps are as follows:

[0009] A multi-scale morphological segmentation algorithm combined with building projection analysis is used to identify building occlusion blind spots.

[0010] A dual-channel generative adversarial network is constructed using the U-Net generator and the PatchGAN discriminator, and Xavier is used to randomly initialize the convolutional layer parameters of the U-Net generator.

[0011] The first channel uses the U-Net structure generator to obtain the leaf area index distribution data of the blind area based on the structural parameters of the unobstructed vegetation and environmental parameters around the blind area, and completes the parameters of the vegetation in the blind area blocked by the building.

[0012] The second channel uses the PatchGAN discriminator to identify the spatial distribution differences between the leaf area index distribution data of the blind area and the canopy density distribution characteristics of the blind area edge based on the canopy density distribution characteristics of the blind area edge, and dynamically optimizes the convolution layer parameters of the U-Net structure generator through back propagation;

[0013] The leaf area index distribution data of the blind area is fused with the initial green space leaf area index distribution data through the Kriging method to generate the complete green space leaf area index distribution data.

[0014] As a preferred solution of the urban green space carbon sink measurement method based on multi-source data fusion described in the present invention, the construction process of the spatiotemporal perception dynamic weighted network includes constructing the initial architecture of the spatiotemporal perception dynamic weighted network based on a bidirectional GRU layer, a 3D convolutional layer and a multi-head attention mechanism, and using a joint training set of spatiotemporal parameters to initialize hyperparameters, and finally forming a spatiotemporal perception dynamic weighted network.

[0015] As a preferred solution of the urban green space carbon sink measurement method based on multi-source data fusion according to the present invention, the steps of generating the carbon sink data fusion weight matrix are as follows:

[0016] The spatiotemporal characteristics of each pixel in the complete green space leaf area index distribution data are extracted through the spatiotemporal convolutional network;

[0017] A spatiotemporal-aware dynamic weighted network is used to dynamically assign weights to the spatiotemporal feature matrix of each pixel, and a non-linear interaction is performed through feature cascade and gating mechanism to generate a carbon sink data fusion weight matrix.

[0018] As a preferred solution of the urban green space carbon sequestration measurement method based on multi-source data fusion described in the present invention, the steps of using a ray tracing algorithm to simulate the path of light reflected from a building glass curtain wall and obtain the photosynthetically active radiation correction coefficient received by the vegetation canopy are as follows:

[0019] Based on the carbon sink data fusion weight matrix, the Monte Carlo ray tracing algorithm is used to simulate the path of reflected light from the building's glass curtain wall and generate a photosynthetically active radiation increment distribution map.

[0020] The photosynthetically active radiation correction coefficient was generated by performing a pixel-by-pixel ratio calculation on the photosynthetically active radiation increment distribution map and the canopy photosynthetically active radiation distribution data obtained by inverting urban vegetation spatial images.

[0021] As a preferred solution of the urban green space carbon sink measurement method based on multi-source data fusion of the present invention, wherein: the urban green space carbon sink distribution data is identified by multi-scale data fusion, the steps are as follows:

[0022] Convolutional feature extraction is performed on the photosynthetically active radiation correction coefficient and the complete green space leaf area index distribution data, and then fused through a gated feature modulation network to generate light-corrected leaf area index distribution data;

[0023] Gaussian pyramid decomposition was performed on the light-corrected leaf area index distribution data to obtain the spatial characteristic subbands at each scale. Multi-scale fusion was then performed through ridge regression to generate urban green space carbon sink distribution data.

[0024] As a preferred solution of the urban green space carbon sink measurement method based on multi-source data fusion described in the present invention, wherein: the carbon sink error is obtained and dynamically corrected by the Bayesian optimization algorithm to generate the urban green space carbon sink measurement report, the steps are as follows:

[0025] The Bootstrap resampling method was used to analyze the carbon sequestration error of each pixel;

[0026] The carbon sink data fusion weight matrix is used as a spatial prior, and a Gaussian process regression model is constructed in combination with multivariate auxiliary data to generate a spatial correlation prior distribution of carbon sinks.

[0027] Taking the spatial correlation prior distribution of carbon sequestration as a spatial constraint, the Bayesian optimization algorithm is used to dynamically correct the carbon sequestration error of each pixel in the urban green space carbon sequestration distribution data.

[0028] Through Markov Chain Monte Carlo, the posterior probability sampling of the corrected urban green space carbon sink distribution data was performed to generate a grid-type carbon sink probability distribution map;

[0029] Through the probability density driven spatial integration method, the mean value of each pixel in the raster carbon sink probability distribution map is spatially integrated to calculate the total carbon sink of urban green spaces and generate an urban green space carbon sink measurement report.

[0030] In a second aspect, the present invention provides an urban green space carbon sequestration measurement system based on multi-source data fusion, comprising: a data acquisition module for collecting multi-source data of urban green spaces and performing pre-processing;

[0031] The parameter completion module is used to complete the parameters of vegetation in building-blocked blind areas using a dual-channel generative adversarial network to obtain complete green space leaf area index distribution data;

[0032] The weight allocation module is used to perform dynamic weight allocation based on the complete green space leaf area index distribution data using a spatiotemporal perception dynamic weighted network and generate a carbon sink data fusion weight matrix;

[0033] The carbon sink identification module uses a ray tracing algorithm to simulate the path of light reflected from building glass curtain walls, obtains the photosynthetically active radiation correction factor received by the vegetation canopy, and identifies the carbon sink distribution data of urban green spaces through multi-scale data fusion;

[0034] The carbon sink distribution generation module is used to obtain the carbon sink error based on the urban green space carbon sink distribution data, and dynamically correct it through the Bayesian optimization algorithm to generate an urban green space carbon sink measurement report.

[0035] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the urban green space carbon sequestration measurement method based on multi-source data fusion as described in the first aspect of the present invention.

[0036] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the urban green space carbon sequestration measurement method based on multi-source data fusion as described in the first aspect of the present invention.

[0037] The present invention achieves the following beneficial effects: through a dual-channel generative adversarial network, utilizing a collaborative training mechanism of a U-Net generator and a PatchGAN discriminator, combined with three-stage supervision using a cycle consistency loss function, it achieves high-precision spatial continuity reconstruction of leaf area index in building-occluded blind areas. Using a ray tracing algorithm, Monte Carlo simulation of the path of reflected light from building glass curtain walls, combined with multi-scale fusion of a gated feature modulation network, it accurately quantifies the enhancement effect of reflected light on vegetation photosynthetically active radiation, providing a key correction basis for carbon sequestration calculations in complex urban environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 This is a flow chart of the urban green space carbon sequestration measurement method based on multi-source data fusion.

[0040] Figure 2 Schematic diagram of the urban green space carbon sequestration measurement system based on multi-source data fusion.

[0041] Figure 3 Flowchart for completing building occlusion blind zone parameters.

[0042] Figure 4 Flowchart of ray tracing and multi-scale fusion. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0044] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0045] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0046] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for measuring urban green space carbon sequestration based on multi-source data fusion, comprising the following steps:

[0047] S1. Collect multi-source data of urban green space and perform pre-processing;

[0048] Urban green space multi-source data includes vegetation structure parameters, environmental parameters, urban vegetation spatial image data, and building three-dimensional data;

[0049] It should be noted that vegetation structure parameters such as vegetation height, canopy density and leaf area index are obtained through lidar scanning; environmental parameters are derived from solar radiation intensity, air temperature and humidity, and carbon dioxide concentration recorded by meteorological stations; spatial image data of urban vegetation use high-resolution multispectral remote sensing images to invert vegetation indices; and three-dimensional building data are obtained based on oblique photogrammetry technology.

[0050] Preprocessing includes data cleaning, format standardization, missing value filling and spatial registration.

[0051] Furthermore, data cleaning uses an outlier detection algorithm to remove outliers in vegetation structure parameters; format standardization converts environmental parameters into CSV format under the WGS84 coordinate system; missing value filling uses the Kriging interpolation method to complete the cloud-occluded areas in the spatial image data of urban vegetation; spatial registration uses a feature point matching algorithm to align the spatial resolution of the three-dimensional building data and the spatial image data of urban vegetation.

[0052] S2. Use a dual-channel generative adversarial network to complete the parameters of vegetation in building-blocked blind areas and obtain complete green space leaf area index distribution data;

[0053] Based on the spatial image data of urban vegetation and the three-dimensional data of buildings, a multi-scale morphological segmentation algorithm combined with the building projection analysis method is used to identify building occlusion blind spots;

[0054] Furthermore, based on spatial imagery data of urban vegetation and 3D building data, the team first simulated the sun's trajectory on the 3D building data using a solar trajectory projection analysis method to identify building shadow areas at different times. A multi-scale morphological segmentation algorithm was then used to extract vegetation coverage based on the spatial imagery of urban vegetation. Combined with the building projection analysis method, the building shadow areas were spatially overlaid with the vegetation coverage areas to identify vegetation areas completely obscured by buildings. An edge detection algorithm was then used to determine the precise boundaries of these blind spots. Finally, a spatial distribution vector map of building blind spots was generated, containing attribute information such as their location, area, and surrounding environmental characteristics.

[0055] Use U-Net generator and PatchGAN discriminator to build a dual-channel generative adversarial network;

[0056] Furthermore, the U-Net generator consists of an encoder and decoder. The encoder extracts multi-scale features through five layers of downsampling convolutions, while the decoder reconstructs features using skip connections and transposed convolutions. The PatchGAN discriminator uses a fully convolutional network structure and uses a sliding window approach to locally discriminate unobstructed vegetation structure parameters and environmental parameters around the blind spot. This ultimately forms a dual-channel generative adversarial network.

[0057] Based on the 3D building data and vegetation height data, a dual-channel generative adversarial network is trained using a cycle consistency loss function.

[0058] Furthermore, in each iteration, the dual-channel generative adversarial network receives 3D building data as input, the U-Net generator outputs predicted vegetation parameters, and the PatchGAN discriminator distinguishes the authenticity of the leaf area index distribution data in the blind area. The cycle consistency loss function consists of three parts: the adversarial loss uses the discriminator output to identify the distribution difference between the generated data and the real data; the cycle reconstruction loss re-inputs the blind area leaf area index distribution data into the U-Net generator to obtain the reconstruction error, and optimizes the dual-channel generative adversarial network based on the reconstruction error.

[0059] Based on the spatial image data of urban vegetation, the vegetation index inversion algorithm is used to extract the canopy density distribution characteristics at the edge of the blind area;

[0060] Furthermore, the edge features of the vegetation index image are first extracted by the Canny edge detection algorithm, and the vegetation boundary line with a single pixel width is obtained by combining morphological refinement processing; a buffer zone with a width of 3-5 pixels is established with the boundary of the building occlusion blind spot as the center, and the spatial distribution characteristics of the vegetation index are statistically analyzed within the buffer zone, including the mean, coefficient of variation and texture feature parameters; the Gaussian filtering algorithm is used to smooth the edge features to generate the canopy density distribution characteristics of the blind spot edge with continuous gradient characteristics. The canopy density distribution characteristics of the blind spot edge include quantitative parameters such as the spatial position, density gradient and morphological characteristics of the vegetation edge.

[0061] Xavier is used to randomly initialize the convolutional layer parameters of the U-Net generator;

[0062] Furthermore, Xavier automatically determines the appropriate distribution range for weight parameters based on the number of input and output channels in each network layer, ensuring that activation values at each layer maintain stable numerical properties. Weight parameters are randomly generated from a normal distribution within a specific range, and bias terms are uniformly set to zero. This initialization strategy effectively maintains the stability of gradient propagation during network training, avoiding numerical non-convergence in the early stages of training and providing a good initial state for subsequent network parameter optimization.

[0063] It should be noted that the cycle consistency loss function refers to a three-stage supervision mechanism that constructs a bidirectional generation path (such as forward generation and reverse reconstruction) through an adversarial training framework, identifies the differences between reconstructed data (such as surrounding vegetation parameter data generated by reverse reconstruction) and original input data (such as the true value of surrounding vegetation parameters obtained by field surveys), and combines spatial gradient constraints (such as vegetation canopy edge consistency detection based on the Sobel operator).

[0064] The first channel uses the U-Net structure generator to obtain the leaf area index distribution data of the blind area based on the structural parameters of the unobstructed vegetation and environmental parameters around the blind area, and completes the parameters of the vegetation in the blind area blocked by the building.

[0065] Furthermore, the unobstructed vegetation structure parameters and environmental parameters around the blind spot are subjected to multi-level feature extraction through the encoder part of the U-Net structure generator. The encoder gradually captures the hierarchical features of the unobstructed vegetation structure parameters around the blind spot through five-layer convolution operations; the decoder part realizes feature reconstruction through transposed convolution and jump connection, and fuses the multi-scale features extracted by the encoder with the spatial information of the corresponding level, and finally outputs the leaf area index distribution data of the building occlusion blind spot; the completed leaf area index distribution data maintains spatial continuity with the vegetation parameters around the blind spot, ensuring the ecological rationality of the vegetation parameter reconstruction results in the building occlusion blind spot.

[0066] The second channel uses the PatchGAN discriminator to identify the spatial distribution differences between the leaf area index distribution data of the blind area and the canopy density distribution characteristics of the blind area edge based on the canopy density distribution characteristics of the blind area edge, and dynamically optimizes the convolution layer parameters of the U-Net structure generator through back propagation;

[0067] Furthermore, the discriminator performs local feature matching on the blind area leaf area index distribution data output by the U-Net structure generator through a sliding window method; the PatchGAN discriminator uses a five-layer convolutional structure to extract spatial features step by step, identifying the spatial distribution differences between the blind area leaf area index distribution data and the canopy density distribution characteristics at the edge of the blind area; and quantifies the degree of spatial consistency through an adversarial loss function. The spatial distribution differences are transmitted to the U-Net structure generator through the back propagation algorithm, and the weight parameters of each convolutional layer of the generator are dynamically adjusted; the optimization process focuses on correcting the leaf area index distribution in the transition zone between the blind area and the surrounding area.

[0068] It should be noted that the convolution layer parameters of the U-Net structure generator include the downsampling convolution kernel weights of each encoder layer, the transposed convolution kernel weights of each decoder layer, the scaling and translation parameters of the batch normalization layer, and the jump connection weight coefficients between each layer.

[0069] The leaf area index distribution data of the blind area is fused with the initial green space leaf area index distribution data through the Kriging method to generate the complete green space leaf area index distribution data.

[0070] Furthermore, first, the spatial distribution characteristics of the blind spot leaf area index distribution data and the initial green space leaf area index distribution data were analyzed by the Kriging interpolation method to determine the degree of correlation between different locations; the interpolation parameters including the minimum influence distance and the maximum search range were set, and the blind spot leaf area index distribution data was used as the area to be supplemented, and the initial green space leaf area index distribution data was used as the reference area; the optimal estimate of each location was calculated by the spatial weighted average method, and the weight was determined by the data point spacing and distribution density; a progressive area expansion strategy was adopted, starting from the blind spot boundary and gradually calculating outward to ensure smooth connection of the transition area; the complete green space leaf area index distribution data was finally generated to achieve a natural transition between the blind spot and the surrounding area.

[0071] It should be noted that the initial green space leaf area index distribution data were obtained by analyzing vegetation indices (such as NDVI and EVI) through direct band reflectance analysis of multispectral satellite remote sensing images and statistical analysis.

[0072] S3. Based on the complete green space leaf area index distribution data, a spatiotemporal-aware dynamic weighted network is used to perform dynamic weight allocation and generate a carbon sink data fusion weight matrix;

[0073] The spatiotemporal features of each pixel in the complete green space leaf area index distribution data are extracted through the spatiotemporal convolutional network (STCN);

[0074] It should be noted that pixel refers to the smallest spatial unit in the raster image of the complete green space leaf area index distribution data; the spatiotemporal characteristics reflect the dynamic change laws of the complete green space leaf area index distribution data in the time dimension (such as seasonal change trends) and spatial dimension (such as urban area differences).

[0075] Based on the bidirectional GRU layer, 3D convolutional layer and multi-head attention mechanism, the initial architecture of the spatiotemporal-aware dynamic weighted network is constructed. The spatiotemporal parameter joint training set is used to initialize the hyperparameters, and finally the spatiotemporal-aware dynamic weighted network is constructed.

[0076] Furthermore, the bidirectional GRU layer is responsible for capturing the dynamic changes of the complete green space leaf area index distribution data in the time dimension, and extracting temporal features through forward and backward processing directions; the 3D convolution layer processes three-dimensional data blocks composed of spatial and temporal dimensions, and uses a 3×3×3 convolution kernel to extract local spatiotemporal features; the multi-head attention mechanism analyzes the correlation between different spatiotemporal features and calculates the dependency weights between features; the spatiotemporal parameter joint training set contains temporal leaf area index data, vegetation classification data and multivariate auxiliary data, which are used to initialize the network's learning rate, batch size and number of training rounds; and finally forms a spatiotemporal-aware dynamic weighted network.

[0077] The joint training set of spatiotemporal parameters includes time series leaf area index data, spatial benchmark data (vegetation classification data), and multivariate auxiliary data (CO2 flux, soil carbon content, and meteorological data);

[0078] The process for acquiring the joint training set of spatiotemporal parameters is as follows: high-resolution remote sensing imagery of urban green spaces is collected, and time-series leaf area index data is generated using an inversion algorithm. LiDAR point cloud data from the same area is simultaneously acquired using LiDAR. A random forest-based point cloud classification algorithm is used to distinguish vegetation from non-vegetation features, generating spatial benchmark data (i.e., vegetation classification data). This data is combined with CO2 flux data and soil carbon content data recorded by ground-based fixed observation stations, as well as meteorological data provided by meteorological stations, to form multivariate auxiliary data. All data undergoes spatiotemporal alignment to ensure that the time-series leaf area index data, spatial benchmark data, and multivariate auxiliary data are completely consistent in terms of timestamps (UTC) and spatial coordinates (WGS84), ultimately forming the joint training set of spatiotemporal parameters.

[0079] A spatiotemporal-aware dynamic weighted network is used to dynamically assign weights to the spatiotemporal feature matrix of each pixel, and a non-linear interaction is performed through feature cascade and gating mechanism to generate a carbon sink data fusion weight matrix.

[0080] Furthermore, the spatiotemporal feature matrix of each pixel in the complete green space leaf area index distribution data is first input into the bidirectional GRU layer to extract the forward and backward dependencies in the time series; the 3D convolution layer performs a local receptive field scan on the spatiotemporal feature matrix to capture the feature change patterns of spatial neighborhoods and temporal neighborhoods; the multi-head attention mechanism calculates the correlation weights between different spatiotemporal features, focusing on the feature interactions affected by vegetation growing season phase changes and urban microclimate; the feature cascade operation splices the temporal features output by the bidirectional GRU layer, the spatial features extracted by the 3D convolution layer, and the attention weights; the gating mechanism dynamically adjusts the contribution weights of multi-source features through the feature selection gate controlled by the Sigmoid function; and the final output carbon sink data fusion weight matrix accurately quantifies the relative importance of the spatiotemporal feature matrix to the vegetation carbon sequestration capacity.

[0081] It should be noted that the spatiotemporal feature matrix is a spatiotemporal feature tensor extracted by integrating multi-temporal remote sensing data (such as NDVI time series) and spatial auxiliary data (such as meteorological and topographic data) using an attention mechanism.

[0082] S4. Use ray tracing algorithms to simulate the path of light reflected from building glass curtain walls, obtain the photosynthetically active radiation correction factor received by the vegetation canopy, and identify the carbon sink distribution data of urban green spaces through multi-scale data fusion;

[0083] Based on the carbon sink data fusion weight matrix, the Monte Carlo ray tracing algorithm is used to simulate the path of reflected light from the building's glass curtain wall and generate a photosynthetically active radiation increment distribution map.

[0084] Furthermore, the glass curtain wall surface in the building's three-dimensional data is divided into multiple reflective units according to material properties, and each unit is set with specific reflectivity and scattering parameters based on the optical property data of the building material; the Monte Carlo ray tracing algorithm emits a large number of randomly sampled light rays from the sun's position, and the reflection path is calculated according to the Fresnel reflection law when the light rays encounter the glass curtain wall surface; the intersection position of the reflected light rays and the vegetation canopy is importance sampled through the carbon sink data fusion weight matrix, and high-weight areas receive more ray tracing calculations; the energy contribution of each reflected light ray is quantified according to the incident angle, distance attenuation and curtain wall reflectivity, and the reflected light energy distribution received at each position of the vegetation canopy is accumulated; the final output photosynthetic active radiation increment distribution map records the radiation enhancement caused by the building's reflected light in raster form.

[0085] The photosynthetically active radiation increment distribution map is compared with the canopy photosynthetically active radiation distribution data obtained by inverting urban vegetation spatial images, and the photosynthetically active radiation correction coefficient is generated. The expression is:

[0086] ;

[0087] in, It is a pixel Correction factor for photosynthetically active radiation, is the photosynthetically active radiation increment distribution image element The value of (building reflected light weight coefficient), is the canopy photosynthetically active radiation distribution image element The value of (vegetation inversion value);

[0088] It should be noted that the dimensions were unified through standardization and normalization before calculation;

[0089] Convolutional feature extraction is performed on the photosynthetically active radiation correction coefficient and the complete green space leaf area index distribution data, and then fused through a gated feature modulation network to generate light-corrected leaf area index distribution data;

[0090] Furthermore, a two-branch convolutional neural network is first used to extract the spatial distribution characteristics of the photosynthetically active radiation correction coefficient (such as the distribution of radiation hotspots formed by building reflected light) and the hierarchical characteristics of the complete green space leaf area index distribution data (such as local canopy structure). Then, based on the gating unit of the Sigmoid function, the spatial distribution characteristics are used as the control signal to dynamically adjust the response intensity of the hierarchical characteristics. Finally, the modulated hierarchical features are spatially reconstructed through the transposed convolution layer and fused to generate the light-corrected leaf area index distribution data.

[0091] Gaussian pyramid decomposition was performed on the light-corrected leaf area index distribution data to obtain the spatial characteristic subbands at each scale. Multi-scale fusion was then performed through ridge regression to generate urban green space carbon sink distribution data.

[0092] Furthermore, the Gaussian pyramid decomposition algorithm was first used to generate characteristic subbands containing five scale levels through step-by-step downsampling. Each subband retained vegetation characteristic information of different spatial frequencies. During the decomposition process, the light-corrected leaf area index distribution data was smoothed using a Gaussian kernel with a standard deviation of 1.6, and bilinear interpolation was used between adjacent scales to maintain geometric consistency. The characteristic subbands at each scale were aligned with the corresponding scale of the photosynthetically active radiation increment distribution map to ensure accurate spatial position correspondence. In the ridge regression fusion stage, the regularization parameter λ was set to 0.5, and the fusion weights of the characteristics at each scale were calculated through least squares optimization to finally generate the urban green space carbon sink distribution data.

[0093] S5. Obtain the carbon sink error based on the urban green space carbon sink distribution data, and dynamically correct it through the Bayesian optimization algorithm to generate an urban green space carbon sink measurement report.

[0094] Based on the urban green space carbon sequestration distribution data, combined with meteorological data and light-corrected leaf area index distribution data, the Bootstrap resampling method was used to analyze the carbon sequestration error of each pixel.

[0095] Furthermore, based on the carbon sink distribution data of urban green spaces, combined with the temperature, radiation and precipitation parameters in the meteorological data, as well as the light-corrected leaf area index distribution data, a spatial dataset was constructed; the Bootstrap resampling method was used to perform multiple random samplings on each pixel position to generate a probability distribution of carbon sink values; by analyzing the probability distribution characteristics, the error range of the estimated carbon sink value of each pixel was calculated; and the final output result was the carbon sink error recorded separately for each pixel.

[0096] The carbon sink data fusion weight matrix is used as the spatial prior, and a Gaussian process regression model is constructed in combination with multivariate auxiliary data to generate the spatial correlation prior distribution of carbon sinks.

[0097] Furthermore, the carbon sink data fusion weight matrix is used as a spatial weight constraint condition and spatially superimposed with the vegetation type, soil properties and terrain parameters in the multivariate auxiliary data; the kernel function adopts a linear combination of the radial basis function and the carbon sink data fusion weight matrix, and the length scale parameter is set to 30 meters according to the average size of the vegetation patch; the mean function is initialized by the linear regression result of the multivariate auxiliary data, the covariance matrix is determined by spatial autocorrelation analysis, and is trained using the conjugate gradient method to maximize the marginal likelihood function; the final generated carbon sink spatial correlation prior distribution.

[0098] Taking the spatial correlation prior distribution of carbon sequestration as a spatial constraint, the Bayesian optimization algorithm is used to dynamically correct the carbon sequestration error of each pixel in the urban green space carbon sequestration distribution data.

[0099] Furthermore, a Bayesian optimization algorithm is used to gradually correct errors using information from neighboring pixels. Each iteration adjusts the correction amplitude based on the carbon sink weight, ultimately correcting the carbon sink error for each pixel in the urban green space carbon sink distribution data. Spatial constraints, in particular, are conditions that restrict the data distribution using known spatial correlation patterns. For example, in vegetation carbon sink analysis, carbon sinks in adjacent pixels are often similar, and this spatial autocorrelation constitutes a spatial constraint.

[0100] It should be noted that spatial constraints refer to the use of the correlation of geographic spatial data (such as the similarity of attributes of adjacent areas). For example, in carbon sink correction, it is mandatory that the difference between the corrected pixel value and the weighted average value of its surrounding pixels does not exceed a standard range (such as ±15%) to maintain spatial continuity.

[0101] The Markov Chain Monte Carlo method is used to perform posterior probability sampling on the corrected urban green space carbon sink distribution data to generate a grid-type carbon sink probability distribution map.

[0102] Furthermore, the revised urban green space carbon sink distribution data is used as the initial state, and a Markov chain with spatial correlation as the transfer rule is constructed (for example, the revised carbon sink pixel value is used as the state node, and when setting the transfer probability matrix, the carbon sink value transfer probability between adjacent pixels is positively correlated with its spatial weight, thereby simulating the spatial dynamic evolution of the carbon sink through Markov Chain Monte Carlo (MCMC).), and the sampling step size is set; each iteration is based on the Metropolis-Hastings criterion, with the carbon sink spatial correlation prior distribution as the recommended distribution to obtain the new state acceptance probability; during the sampling process, the spatial autocorrelation characteristics between pixels are maintained (for example, the carbon sinks of adjacent vegetation pixels show similar change trends), and high-error areas are densely sampled; finally, a grid-type carbon sink probability distribution map is generated.

[0103] The total carbon sink of urban green space is calculated by spatial integration of the mean value of each pixel in the grid carbon sink probability distribution map through probability density driven spatial integration method. The expression is:

[0104] ;

[0105] in, is the total carbon sink of urban green spaces, is the total number of pixels involved in the calculation of total carbon sink, It is The average carbon sink of each pixel is is the dynamic weight coefficient driven by probability density (value range: 0.3~0.7), It is The standard deviation of carbon sequestration per pixel, is a very small constant, It is The area of a pixel.

[0106] Based on the carbon sink error, total carbon sink of urban green space, grid-type carbon sink probability distribution map and urban green space carbon sink distribution data, an urban green space carbon sink measurement report is generated.

[0107] This embodiment also provides an urban green space carbon sequestration measurement system based on multi-source data fusion, comprising: a data acquisition module for collecting multi-source data of urban green spaces and performing pre-processing;

[0108] The parameter completion module is used to complete the parameters of vegetation in building-blocked blind areas using a dual-channel generative adversarial network to obtain complete green space leaf area index distribution data;

[0109] The weight allocation module is used to perform dynamic weight allocation based on the complete green space leaf area index distribution data using a spatiotemporal perception dynamic weighted network and generate a carbon sink data fusion weight matrix;

[0110] The carbon sink identification module uses a ray tracing algorithm to simulate the path of light reflected from building glass curtain walls, obtains the photosynthetically active radiation correction factor received by the vegetation canopy, and identifies the carbon sink distribution data of urban green spaces through multi-scale data fusion;

[0111] The carbon sink distribution generation module is used to obtain the carbon sink error based on the urban green space carbon sink distribution data, and dynamically correct it through the Bayesian optimization algorithm to generate an urban green space carbon sink measurement report.

[0112] This embodiment also provides a computer device, which is suitable for the urban green space carbon sequestration measurement method based on multi-source data fusion, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the urban green space carbon sequestration measurement method based on multi-source data fusion proposed in the above embodiment.

[0113] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0114] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the urban green space carbon sink measurement method based on multi-source data fusion proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0115] In summary, this paper achieves high-precision spatial continuity reconstruction of leaf area index in building-occluded blind areas through a dual-channel generative adversarial network, utilizing a collaborative training mechanism between a U-Net generator and a PatchGAN discriminator, and combining three-stage supervision with a cycle consistency loss function. Using a ray tracing algorithm, Monte Carlo simulation of the path of reflected light from building glass curtain walls, and multi-scale fusion with a gated feature modulation network, the paper accurately quantifies the enhancement effect of reflected light on vegetation photosynthetically active radiation, providing a key correction basis for carbon sequestration calculations in complex urban environments.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for measuring urban green space carbon sequestration based on multi-source data fusion, characterized by: include, Collecting and preprocessing multi-source data of urban green spaces; the multi-source data of urban green spaces includes vegetation structure parameters, environmental parameters, spatial image data of urban vegetation, and three-dimensional building data; A dual-channel generative adversarial network is used to complete the parameters of vegetation in building-blocked blind areas and obtain complete green space leaf area index distribution data. Based on the complete green space leaf area index distribution data, a spatiotemporal-aware dynamic weighted network is used to perform dynamic weight allocation and generate a carbon sink data fusion weight matrix. A ray tracing algorithm was used to simulate the path of light reflected from building glass curtain walls, obtain the correction coefficient for photosynthetically active radiation received by the vegetation canopy, and identify the distribution of carbon sequestration data in urban green spaces through multi-scale data fusion. Obtain carbon sequestration errors based on urban green space carbon sequestration distribution data, dynamically correct them using the Bayesian optimization algorithm, and generate an urban green space carbon sequestration measurement report. The construction process of the spatiotemporal perception dynamic weighted network includes building an initial architecture of the spatiotemporal perception dynamic weighted network based on a bidirectional GRU layer, a 3D convolutional layer, and a multi-head attention mechanism, and using a spatiotemporal parameter joint training set to initialize hyperparameters, ultimately forming the spatiotemporal perception dynamic weighted network; The steps for generating the carbon sink data fusion weight matrix are as follows: The spatiotemporal characteristics of each pixel in the complete green space leaf area index distribution data are extracted through the spatiotemporal convolutional network; A spatiotemporal-aware dynamic weighted network is used to dynamically assign weights to the spatiotemporal feature matrix of each pixel, and a non-linear interaction is performed through feature cascade and gating mechanism to generate a carbon sink data fusion weight matrix. The steps of using the ray tracing algorithm to simulate the path of light reflected from the building glass curtain wall and obtain the photosynthetically active radiation correction coefficient received by the vegetation canopy are as follows: Based on the carbon sink data fusion weight matrix, the Monte Carlo ray tracing algorithm is used to simulate the path of reflected light from the building's glass curtain wall and generate a photosynthetically active radiation increment distribution map. Perform pixel-by-pixel ratio calculation on the photosynthetically active radiation increment distribution map and the canopy photosynthetically active radiation distribution data obtained by inverting urban vegetation spatial images to generate the photosynthetically active radiation correction coefficient; The steps of identifying urban green space carbon sink distribution data through multi-scale data fusion are as follows: Convolutional feature extraction is performed on the photosynthetically active radiation correction coefficient and the complete green space leaf area index distribution data, and then fused through a gated feature modulation network to generate light-corrected leaf area index distribution data; Gaussian pyramid decomposition was performed on the light-corrected leaf area index distribution data to obtain the spatial characteristic subbands at each scale. Multi-scale fusion was then performed through ridge regression to generate urban green space carbon sink distribution data.

2. The urban green space carbon sequestration measurement method based on multi-source data fusion according to claim 1 is characterized by: The dual-channel generative adversarial network is used to complete the parameters of vegetation in the building blind area and obtain the complete green space leaf area index distribution data. The steps are as follows: A multi-scale morphological segmentation algorithm combined with building projection analysis is used to identify building occlusion blind spots. A dual-channel generative adversarial network is constructed using the U-Net generator and the PatchGAN discriminator, and Xavier is used to randomly initialize the convolutional layer parameters of the U-Net generator. The first channel uses the U-Net structure generator to obtain the leaf area index distribution data of the blind area based on the structural parameters of the unobstructed vegetation and environmental parameters around the blind area, and completes the parameters of the vegetation in the blind area blocked by the building. The second channel uses the PatchGAN discriminator to identify the spatial distribution differences between the leaf area index distribution data of the blind area and the canopy density distribution characteristics of the blind area edge based on the canopy density distribution characteristics of the blind area edge, and dynamically optimizes the convolution layer parameters of the U-Net structure generator through back propagation; The leaf area index distribution data of the blind area is fused with the initial green space leaf area index distribution data through the Kriging method to generate the complete green space leaf area index distribution data.

3. The urban green space carbon sequestration measurement method based on multi-source data fusion according to claim 2 is characterized by: The steps of obtaining carbon sink error and dynamically correcting it through Bayesian optimization algorithm to generate urban green space carbon sink measurement report are as follows: The Bootstrap resampling method was used to calculate the carbon sequestration error of each pixel; The carbon sink data fusion weight matrix is used as a spatial prior, and a Gaussian process regression model is constructed in combination with multivariate auxiliary data to generate a spatial correlation prior distribution of carbon sinks. Taking the spatial correlation prior distribution of carbon sequestration as a spatial constraint, the Bayesian optimization algorithm is used to dynamically correct the carbon sequestration error of each pixel in the urban green space carbon sequestration distribution data. Through Markov Chain Monte Carlo, the posterior probability sampling of the corrected urban green space carbon sink distribution data was performed to generate a grid-type carbon sink probability distribution map; Through the probability density driven spatial integration method, the mean value of each pixel in the raster carbon sink probability distribution map is spatially integrated to calculate the total carbon sink of urban green spaces and generate an urban green space carbon sink measurement report.

4. An urban green space carbon sink measurement system based on multi-source data fusion, based on the urban green space carbon sink measurement method based on multi-source data fusion according to any one of claims 1 to 3, characterized in that: include, Data acquisition module, used to collect multi-source data of urban green spaces and perform pre-processing; The parameter completion module is used to complete the parameters of vegetation in building-blocked blind areas using a dual-channel generative adversarial network to obtain complete green space leaf area index distribution data; The weight allocation module is used to perform dynamic weight allocation based on the complete green space leaf area index distribution data using a spatiotemporal perception dynamic weighted network and generate a carbon sink data fusion weight matrix; The carbon sink identification module uses a ray tracing algorithm to simulate the path of light reflected from building glass curtain walls, obtains the photosynthetically active radiation correction factor received by the vegetation canopy, and identifies the carbon sink distribution data of urban green spaces through multi-scale data fusion; The carbon sink distribution generation module is used to obtain the carbon sink error based on the urban green space carbon sink distribution data, and dynamically correct it through the Bayesian optimization algorithm to generate an urban green space carbon sink measurement report.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the urban green space carbon sequestration measurement method based on multi-source data fusion according to any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the urban green space carbon sink measurement method based on multi-source data fusion according to any one of claims 1 to 3 are implemented.

Images