A multimodal urban data fusion analysis system based on digital twins

Abstract

The present invention relates to the technical field of data fusion analysis, and discloses a multimodal urban data fusion analysis system based on digital twins, comprising: acquiring construction images, laser PM data, and XRF spectrum data at fixed time intervals; performing a box counting method on the construction images to obtain fractal dimensions; obtaining a calcium element ratio using XRF spectrum data; extracting a bare soil image from the construction image, and calculating a dust potential index based on the bare soil image; extracting a mean value of the dust potential index within a historical time period, and determining an activity amplification factor in combination with the dust potential index; calculating an effective density based on the calcium element ratio in combination with a preset reference density and calcium element density; determining a shape correction coefficient based on the fractal dimension; inverting the mass concentration of the dust cloud based on the laser PM data in combination with the effective density, the activity amplification factor, and the shape correction coefficient; and obtaining a surface area concentration of the dust cloud by inverting the mass concentration of the dust cloud in combination with the fractal dimension and the effective density.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion analysis, and more specifically, to a multimodal urban data fusion analysis system based on digital twins. Background Art

[0002] During urban construction, dust pollution is increasingly impacting the environment and human health. Dust generated at construction sites often exhibits unique physical properties. For example, calcium-rich particles tend to form complex agglomerates, resulting in a morphology and density distribution that significantly deviates from conventional spherical particles. This type of dust can cause a sudden spike in PM2.5 concentrations in the surrounding area, causing air quality indexes to exceed standards and increasing the incidence of respiratory diseases in workers exposed over the long term. However, traditional monitoring methods are not sufficiently adaptable to these unique particles, making it difficult to accurately quantify the extent of pollution. Targeted technical improvements are urgently needed.

[0003] Defects of existing construction dust monitoring technology: The single-point laser scattering method relies on the idealized setting of uniform spherical particles. When faced with dust with high calcium content and complex morphology, the mass concentration measurement error is large. Summary of the Invention

[0004] The present invention provides a multimodal urban data fusion analysis system based on digital twins to solve the technical problems raised in the background technology.

[0005] The present invention provides a multimodal urban data fusion analysis system based on digital twins, comprising:

[0006] Data acquisition module, used to acquire construction images, laser PM data, and XRF spectrum data at fixed time intervals;

[0007] Fractal dimension module, used to perform box counting method on construction images to obtain fractal dimension;

[0008] Calcium ratio module, used to obtain calcium ratio using XRF spectrum data;

[0009] Bare soil potential module, used to extract bare soil images from construction images and calculate the dust potential index based on the bare soil images;

[0010] The activity coefficient module is used to extract the mean value of the dust potential index within the historical time period and determine the activity amplification coefficient based on the dust potential index;

[0011] Density correction module, used to calculate effective density based on calcium ratio, combined with preset reference density and calcium density;

[0012] A shape coefficient module, used to determine a shape correction coefficient based on the fractal dimension;

[0013] The mass concentration module is used to invert the mass concentration of dust clouds based on laser PM data by combining effective density, activity amplification factor, and shape correction factor;

[0014] The surface concentration module is used to obtain the surface area concentration of dust clouds by using the mass concentration of dust clouds, combined with fractal dimension and effective density inversion.

[0015] Furthermore, the box counting method is performed on the construction image to obtain the fractal dimension, including:

[0016] Convert all the time-series construction images into grayscale matrices;

[0017] Selecting the mean of a preset number of time-series grayscale matrices to construct a reference grayscale matrix;

[0018] Calculate the absolute difference matrix between each selected grayscale matrix and the reference grayscale matrix;

[0019] Compare the i-th element in the absolute difference matrix with the preset difference threshold. If the i-th element is greater than the preset difference threshold, mark the i-th element as 1, otherwise mark it as 0 to generate a foreground mask matrix.

[0020] Perform the 5-pixel kernel opening operation and the 9-pixel kernel closing operation on the foreground mask matrix to form a dust cloud binary matrix, where element 1 represents dust cloud and element 0 represents non-dust cloud;

[0021] Set scale sequence Pixels;

[0022] By scale Divide the dust cloud binary matrix and count the number of boxes covered by the dust cloud in the dust cloud binary matrix , forming a sequence ;

[0023] For sequence Perform least squares regression and obtain the regression slope as the fractal dimension.

[0024] Furthermore, the XRF spectrum data is used to obtain the calcium ratio, including:

[0025] The XRF spectrum data was differentially processed using a preset blank background spectrum to obtain a net spectrum;

[0026] For the characteristic peak of calcium, a fitting window is set; the central energy of the characteristic peak of calcium is is 3.69keV, the fitting window 0.12keV;

[0027] The area of ​​the characteristic peak of calcium element was extracted by least square fitting using Gaussian linear composite model;

[0028] Perform the same fitting on the characteristic peaks of all elements in the net spectrum and accumulate them to obtain the sum of the peak areas of all elements;

[0029] The calcium ratio was obtained by the total element peak area and the area of ​​the normalized calcium characteristic peak.

[0030] Furthermore, the Gaussian linear composite model was used to extract the area of ​​the characteristic peak of calcium element by least square fitting, including:

[0031] In the net spectrum, locate and intercept the energy range of the characteristic peak of calcium , and get the calcium peak energy region spectrum ;

[0032] Establish a Gaussian linear composite model as follows:

[0033]

[0034] in, represents the amplitude of the Gaussian peak to be fitted, represents the broadening of the Gaussian peak to be fitted, represents the slope of the linear term to be fitted, The intercept of the linear term to be fitted;

[0035] The fitted residuals of the Gaussian linear composite model are as follows:

[0036]

[0037] The LM and gradient descent hybrid iterative algorithm is used to minimize the fitting residual of the Gaussian linear composite model to obtain the amplitude of the fitted Gaussian peak, the width of the fitted Gaussian peak, the slope of the fitted linear term, and the intercept of the fitted linear term;

[0038] The calcium peak area is calculated using the integral analytical algorithm of the Gaussian distribution as follows:

[0039]

[0040] in, represents the calcium peak area, represents the amplitude of the fitted Gaussian peak, represents the broadening of the fitted Gaussian peak.

[0041] Furthermore, the bare soil image in the camera video stream is extracted, and the dust potential index is calculated based on the bare soil image, including:

[0042] The grayscale matrix of the construction image is input into the pre-trained semantic segmentation network to form a bare soil binary matrix. The element 1 of the bare soil binary matrix represents bare soil, and the element 0 represents non-bare soil.

[0043] Count the sum of the elements 1 in the bare soil binary matrix, and take the ratio of the sum to the number of elements in the bare soil binary matrix as the bare soil area ratio;

[0044] Obtain the near-infrared channel component of the construction image and convert it into a near-infrared channel grayscale matrix;

[0045] Based on element 1 of the bare soil binary matrix, the elements with the same coordinates in the near-infrared channel grayscale matrix are extracted and the element mean is calculated to obtain the average near-infrared grayscale of the bare soil;

[0046] Determine the soil moisture content range in the construction area and calculate the soil dryness as follows:

[0047]

[0048] in, Indicates the dryness of the soil. Indicates the upper limit of the soil moisture content range, Indicates the lower limit of the soil moisture content range, represents the average near-infrared grayscale of bare soil, Indicates normalization processing;

[0049] The product of soil dryness and the proportion of bare soil area is used as the dust potential index.

[0050] Furthermore, the mean value of the dust potential index in the historical time period is extracted, and the activity amplification coefficient is determined in combination with the dust potential index. The activity amplification coefficient is the ratio of the dust potential index to the mean value of the dust potential index.

[0051] Furthermore, based on the calcium ratio, the effective density is calculated by combining the preset reference density and the calcium density. The effective density is obtained by linear difference, as follows:

[0052]

[0053] in, represents the effective density, Indicates the preset reference density, represents the calcium element density, Indicates the calcium ratio.

[0054] Furthermore, the shape correction coefficient is determined based on the fractal dimension, including:

[0055] Read the preset fractal slope coefficient of dust ;

[0056] Saturation clipping is performed based on the fractal dimension as follows:

[0057] If the fractal dimension , , then the fractal dimension is maintained as ; Otherwise, the fractal dimension The value is or 3;

[0058] According to the fractal dimension Calculate the shape correction factor as follows:

[0059]

[0060] in, Represents the shape correction factor.

[0061] Furthermore, based on the laser PM data, the effective density, activity amplification factor, and shape correction factor are combined to invert the mass concentration of the dust cloud, including:

[0062] Collect ambient temperature and humidity;

[0063] The ambient temperature and humidity are normalized with the laser PM data, effective density, activity amplification factor, and shape correction factor and then spliced ​​into a fusion feature vector;

[0064] The fused feature vector is input into the pre-trained mass concentration inversion model to obtain the mass concentration of the dust cloud.

[0065] Furthermore, the mass concentration of the dust cloud is used, combined with the fractal dimension and effective density inversion to obtain the surface area concentration of the dust cloud. The surface area concentration is obtained by calculating the power law model as follows:

[0066]

[0067] in, represents the surface area concentration of the dust cloud, represents the concentration weight, Indicates the mass concentration of the dust cloud.

[0068] The beneficial effects of the present invention are as follows: by constructing a multimodal urban data analysis system that integrates construction images, laser PM data, and XRF spectral data, key structural and dynamic characteristics of dust particles, such as fractal dimensions, calcium enrichment ratio, and construction activity intensity, are fully coupled and modeled, achieving accurate inversion of the mass concentration and surface area concentration of dust on construction sites. By introducing a Gaussian linear composite model to extract calcium content through spectral peak fitting, combining the dust potential index obtained based on semantic segmentation and near-infrared analysis, and the morphology correction parameters obtained from fractal geometry analysis, the above parameters are further input into the nonlinear inversion model in the form of a fused feature vector and adaptive correction is performed, effectively overcoming the problem of traditional laser scattering methods' inaccurate recognition of non-spherical, porous agglomerated, and high-calcium particles. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a module diagram of a multimodal urban data fusion analysis system based on digital twins of the present invention. DETAILED DESCRIPTION

[0070] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0071] like Figure 1 As shown in FIG, a multimodal urban data fusion analysis system based on digital twins includes:

[0072] Data acquisition module, used to acquire construction images, laser PM data, and XRF spectrum data at fixed time intervals;

[0073] Fractal dimension module, used to perform box counting method on construction images to obtain fractal dimension;

[0074] Calcium ratio module, used to obtain calcium ratio using XRF spectrum data;

[0075] Bare soil potential module, used to extract bare soil images from construction images and calculate the dust potential index based on the bare soil images;

[0076] The activity coefficient module is used to extract the mean value of the dust potential index within the historical time period and determine the activity amplification coefficient based on the dust potential index;

[0077] Density correction module, used to calculate effective density based on calcium ratio, combined with preset reference density and calcium density;

[0078] A shape coefficient module, used to determine a shape correction coefficient based on the fractal dimension;

[0079] The mass concentration module is used to invert the mass concentration of dust clouds based on laser PM data by combining effective density, activity amplification factor, and shape correction factor;

[0080] The surface concentration module is used to obtain the surface area concentration of dust clouds by using the mass concentration of dust clouds, combined with fractal dimension and effective density inversion.

[0081] Construction images represent sequences of digital images captured continuously at regular intervals by high-definition cameras deployed on the construction site. Each frame of the digital image can be converted into a single-channel grayscale matrix or a multi-channel matrix (including visible light and near-infrared).

[0082] Laser PM data represents the concentration of inhalable particulate matter in the air collected by the laser scattering particulate matter sensor at fixed time intervals. It usually includes three channels: PM1, PM2.5 and PM10. The output unit is .

[0083] XRF spectrum data refers to the spectrum data obtained by excitation scanning of dust particle samples collected on the filter membrane using a portable or fixed energy dispersive X-ray fluorescence instrument.

[0084] In one embodiment of the present invention, performing a box counting method on a construction image to obtain a fractal dimension includes:

[0085] Convert all the time-series construction images into grayscale matrices;

[0086] The color information is removed through the grayscale matrix to avoid the interference of spectral contrast and shadow color on subsequent differential operations.

[0087] Selecting the mean of a preset number of time-series grayscale matrices to construct a reference grayscale matrix;

[0088] The construction site background (e.g., ground texture and static areas) remains largely unchanged across multiple frames and is directly used to generate a static scene reference. The resulting baseline grayscale matrix accurately reflects the background brightness in the absence of dust, providing a comparison standard for the subsequent absolute difference matrix.

[0089] Calculate the absolute difference matrix between each selected grayscale matrix and the reference grayscale matrix;

[0090] Dust movement causes grayscale changes within a short timeframe, while the background remains relatively stable. The absolute difference matrix highlights the grayscale jumps caused by dust particles, reducing the impact of lighting changes.

[0091] Compare the i-th element in the absolute difference matrix with the preset difference threshold. If the i-th element is greater than the preset difference threshold, mark the i-th element as 1, otherwise mark it as 0 to generate a foreground mask matrix.

[0092] Only when the grayscale change exceeds a threshold is it considered dust cloud motion; otherwise, it may be noise or subtle light jitter. The foreground mask matrix accurately distinguishes possible dust cloud pixels (marked 1) from background or noise (marked 0), laying the foundation for morphological processing.

[0093] Perform the 5-pixel kernel opening operation and the 9-pixel kernel closing operation on the foreground mask matrix to form a dust cloud binary matrix, where element 1 represents dust cloud and element 0 represents non-dust cloud;

[0094] Opening operations remove isolated noise points, while closing operations fill small holes in the dust cloud region. The resulting dust cloud binary matrix has better connectivity and fewer false positives, where element 1 represents a dust cloud and element 0 represents a non-dust cloud, ensuring the integrity of the region during meshing.

[0095] Set scale sequence Pixels;

[0096] Fractal dimensions require multiple scales to characterize area coverage. Geometric scaling ensures that box boundaries align with pixels, avoiding round-off errors in sizing operations. This provides a unified, divisible set of scales for subsequent multi-scale meshing.

[0097] By scale Divide the dust cloud binary matrix and count the number of boxes covered by the dust cloud in the dust cloud binary matrix , forming a sequence ;

[0098] The essence of fractal dimension is the power law relationship between coverage and scale, and the coverage capacity needs to be quantified at each scale. The sequence fully reflects the spatial distribution characteristics of dust clouds at different grid sizes and provides data for slope regression.

[0099] For sequence Least squares regression is performed and the regression slope is obtained as the fractal dimension, which represents the degree of spatial self-similarity of the dust cloud at multiple scales.

[0100] In the logarithmic coordinate system, the fractal dimension is equal to right The linear fitting slope of . The regression slope, as a single-value fractal dimension D, can be directly used in the subsequent calculation of the shape correction coefficient to achieve accurate compensation of the dust cloud morphology during multimodal fusion.

[0101] In one embodiment of the present invention, obtaining the calcium ratio using XRF spectrum data includes:

[0102] The XRF spectrum data was differentially processed using a preset blank background spectrum to obtain a net spectrum;

[0103] XRF spectral data contains instrument background and ambient scattering signals. If not removed, subsequent peak fitting will be highly biased. By performing a channel-by-channel difference between a preset blank background spectrum and the original spectrum, the net spectrum of the true sample signal is obtained, laying the foundation for accurate peak extraction.

[0104] For the characteristic peak of calcium, a fitting window is set; the central energy of the characteristic peak of calcium is is 3.69keV, the fitting window 0.12keV;

[0105] The characteristic peak of calcium is adjacent to the secondary peaks of various light elements, and the interference of adjacent peaks is avoided by limiting the energy range. 3.69keV, fitting window For the 0.12keV segment, only the characteristic peak of calcium and its local base are retained and calculated centrally.

[0106] The area of ​​the characteristic peak of calcium element was extracted by least square fitting using Gaussian linear composite model;

[0107] Because a single Gaussian model cannot eliminate the base ramp error, and a single linear base cannot express the peak shape details, a Gaussian-linear composite model was used to simultaneously describe the peak shape and base, and the least squares method was used to accurately fit the peak area of ​​the calcium characteristic peak.

[0108] Perform the same fitting on the characteristic peaks of all elements in the net spectrum and accumulate them to obtain the sum of the peak areas of all elements;

[0109] The absolute amount of the calcium peak area must be relative to the overall element response to characterize the ratio. All characteristic peaks of elements in the net spectrum are fitted according to the same model, and the areas of each peak are accumulated to obtain the sum of the peak areas of all elements as the normalization benchmark.

[0110] The calcium ratio was obtained by the total element peak area and the area of ​​the normalized calcium characteristic peak.

[0111] Directly using the calcium peak area makes it difficult to eliminate the effects of sampling volume, excitation intensity, etc. The final calcium ratio is obtained by normalizing the total element peak area with the area of ​​the calcium characteristic peak. The calcium ratio is used to indicate the relative content of calcium.

[0112] In one embodiment of the present invention, a Gaussian linear composite model is used to extract the area of ​​the characteristic peak of calcium by least square fitting, including:

[0113] In the net spectrum, locate and intercept the energy range of the characteristic peak of calcium , and get the calcium peak energy region spectrum ;

[0114] Establish a Gaussian linear composite model as follows:

[0115]

[0116] in, represents the amplitude of the Gaussian peak to be fitted, represents the broadening of the Gaussian peak to be fitted, represents the slope of the linear term to be fitted, The intercept of the linear term to be fitted;

[0117] The characteristic peak of calcium element is approximately normally distributed in energy dispersive X-ray fluorescence, and its intensity decay along the energy axis is proportional to the Gaussian function of the atomic transition probability and the detector energy resolution. Describe the peak shape.

[0118] The net spectrum actually measured contains not only the peak signal, but also a relatively flat but possibly slightly trending base (such as scattering tail, spectrum bottom drift). Describe the local background.

[0119] and The combination of the two can simultaneously fit the peak top and the base, ensuring that the peak amplitude and broadening parameters are highly consistent with the real physical process.

[0120] The fitted residuals of the Gaussian linear composite model are as follows:

[0121]

[0122] The LM and gradient descent hybrid iterative algorithm is used to minimize the fitting residual of the Gaussian linear composite model to obtain the amplitude of the fitted Gaussian peak, the width of the fitted Gaussian peak, the slope of the fitted linear term, and the intercept of the fitted linear term;

[0123] The sum of squares is of the form The residuals generate a smooth error surface in parameter space, penalizing positive and negative residuals equally. The sum-of-squares function is continuous and differentiable in parameter space, resulting in a smooth error surface, making it suitable for the Levenberg-Marquardt and gradient descent hybrid iterative algorithm to rapidly approximate the optimal solution. Furthermore, under the condition that the noise is approximately independent and identically distributed, minimizing the squared residual is equivalent to maximum likelihood estimation.

[0124] It should be noted that gradient descent alone is prone to local oscillation in deep slopes or high noise situations, and the LM algorithm alone converges slowly in low-count ranges. A hybrid iterative algorithm combining LM and gradient descent allows for rapid convergence in high-count peak regions while maintaining stable updates in low-count tail regions, ensuring optimal peak amplitude, peak spread, slope, and intercept.

[0125] The calcium peak area is calculated using the integral analytical algorithm of the Gaussian distribution as follows:

[0126]

[0127] in, represents the calcium peak area, represents the amplitude of the fitted Gaussian peak, represents the broadening of the fitted Gaussian peak.

[0128] Because numerical integration is susceptible to truncation and discretization errors, the analytical integration algorithm for the Gaussian distribution is directly based on the calculation of the fitted parameters, which is both accurate and efficient. The calcium peak area quantitatively reflects the relative content of calcium, representing the area under the fitted Gaussian peak.

[0129] In one embodiment of the present invention, extracting a bare soil image from a camera video stream and calculating a dust potential index based on the bare soil image includes:

[0130] The grayscale matrix of the construction image is input into the pre-trained semantic segmentation network to form a bare soil binary matrix. The element 1 of the bare soil binary matrix represents bare soil, and the element 0 represents non-bare soil.

[0131] Multiple sets of construction images were obtained and training samples were obtained by manually annotating the bare soil and non-bare soil areas in the construction images. The grayscale matrices of the construction images were used as training data, and the semantic segmentation network was pre-trained based on the training data and training samples.

[0132] The average near-infrared grayscale of bare soil is used to quantify the average reflectance level of bare soil in the near-infrared band.

[0133] Count the sum of the elements 1 in the bare soil binary matrix, and take the ratio of the sum to the number of elements in the bare soil binary matrix as the bare soil area ratio;

[0134] Obtain the near-infrared channel component of the construction image and convert it into a near-infrared channel grayscale matrix;

[0135] Based on element 1 of the bare soil binary matrix, the elements with the same coordinates in the near-infrared channel grayscale matrix are extracted and the element mean is calculated to obtain the average near-infrared grayscale of the bare soil;

[0136] The closer the bare soil average near-infrared grayscale value is to 1, the drier the bare soil is, and the lower the corresponding moisture content is. The closer it is to 0, the wetter the bare soil is, and the higher the corresponding moisture content is.

[0137] Determine the soil moisture content range in the construction area and calculate the soil dryness as follows:

[0138]

[0139] in, Indicates the dryness of the soil. Indicates the upper limit of the soil moisture content range, Indicates the lower limit of the soil moisture content range, represents the average near-infrared grayscale of bare soil, Indicates normalization processing;

[0140] Based on the preset benchmark parameters Normalization is performed. The drier the soil source, the The closer it is to 1. The preset benchmark parameters are based on statistics of several groups The maximum value in is obtained.

[0141] The product of soil dryness and the proportion of bare soil area is used as the dust potential index.

[0142] It is difficult to fully quantify the dust potential using a single parameter of soil source dryness and bare soil area ratio. Only by coupling the exposed area of ​​bare soil with the dryness can the size and activity of the dust source be reflected simultaneously.

[0143] The Dust Potential Index (DPI) comprehensively characterizes the potential for dust generation from bare soil. A higher DPI value indicates that the dust source is both wider and drier per unit field of view, making it easily disturbed and prone to generating dust.

[0144] In one embodiment of the present invention, the average value of the dust potential index in the historical time period is extracted, and the activity amplification factor is determined in combination with the dust potential index. The activity amplification factor is the ratio of the dust potential index to the average value of the dust potential index.

[0145] Construction activities are cyclical and phased, and the dust potential index of a single event may fluctuate due to short-term sudden disturbances. Using the average value over a historical time period can filter out occasional pulses and form a steady-state baseline dust potential index.

[0146] When the DPI is higher than the historical mean, it indicates an increase in disturbance intensity and a higher probability of dust emission; otherwise, it indicates a decrease. The activity amplification factor, derived by comparing the DPI to the mean, accurately reflects the multiplier effect of current activity intensity compared to the baseline. When the activity amplification factor is greater than 1, the current dust source intensity is higher than normal; when it is less than 1, the current dust source intensity is lower than normal.

[0147] In one embodiment of the present invention, the effective density is calculated based on the calcium ratio in combination with the preset reference density and the calcium density, and the effective density is obtained by linear difference, as follows:

[0148]

[0149] in, represents the effective density, Indicates the preset reference density, represents the calcium element density, Indicates the calcium ratio.

[0150] It should be noted that the preset reference density represents the baseline density of common dust particles, while the calcium density represents the maximum density of pure calcium minerals. The calcium ratio reflects the relative abundance of calcium in the dust particle's chemical composition and is a key weighting factor in determining overall particle density. Using the calcium ratio as an interpolation coefficient, the linear weighting between the baseline and calcium density is dynamically adjusted to achieve a mapping of composition to physical density.

[0151] The chemical ratios need to be converted to physical densities to correct for the single density bias assumed by the laser scattering method: The effective density represents the equivalent volume density of dust particles, and the unit is consistent with the preset reference density and calcium density.

[0152] In one embodiment of the present invention, determining the shape correction coefficient according to the fractal dimension includes:

[0153] Read the preset fractal slope coefficient of dust ;

[0154] When the fractal dimension deviates from 2, an empirical coefficient is needed to characterize the degree of amplification or contraction of the shape for scattering and mass conversion for each unit of deviation.

[0155] Saturation clipping is performed based on the fractal dimension as follows:

[0156] If the fractal dimension , , then the fractal dimension is maintained as ; Otherwise, the fractal dimension The value is or 3;

[0157] In construction site dust, the theoretical fractal dimension of fractal agglomerated particles should be The range beyond this range may be due to computational noise or extreme frame errors, and if directly used for linear mapping, it will produce excessive correction.

[0158] Keep the fractal dimension Within range: When When , take the original value to reflect the actual shape deviation; if , then the fractal dimension The value is 3; if , then the fractal dimension The value is 3.

[0159] According to the fractal dimension Calculate the shape correction factor as follows:

[0160]

[0161] in, Represents the shape correction factor.

[0162] It is necessary to map the fractal dimension into the correction coefficient of scattering and mass conversion to compensate the morphological error simultaneously during the concentration inversion. (Baseline) , no correction; when (porous agglomeration), , automatically magnify and convert; when (Near the ball) , automatic shrinkage and conversion.

[0163] In one embodiment of the present invention, the mass concentration of the dust cloud is obtained by inverting the laser PM data in combination with the effective density, the activity amplification factor, and the shape correction factor, including:

[0164] Collect ambient temperature and humidity;

[0165] Laser scattering is extremely sensitive to the refractive index and water vapor content of the air. Fluctuations in ambient temperature and humidity will directly affect the intensity of the scattering signal and the change in density of the particles after absorbing water.

[0166] Ambient temperature and humidity are used as input to ensure that the subsequent concentration inversion model can compensate for the coupled effects of meteorological conditions on the original scattering readings and the composition and morphology correction results.

[0167] The ambient temperature and humidity are normalized with the laser PM data, effective density, activity amplification factor, and shape correction factor and then spliced ​​into a fusion feature vector;

[0168] Each input feature is mapped to the same numerical interval, eliminating dimensional differences. The six normalized scalars are then arranged in a fixed order to form a fused feature vector. This fused feature vector simultaneously carries scatter readings, composition compensation, topography compensation, and meteorological conditions information, serving as the unified input for multimodal fusion and ensuring that subsequent models can collaboratively model the impact of each dimension.

[0169] The fused feature vector is input into the pre-trained mass concentration inversion model to obtain the mass concentration of the dust cloud.

[0170] It should be noted that the sample data and sample labels used in the pre-trained mass concentration inversion model are obtained in the following process:

[0171] 1. Simultaneously deploy at several representative construction sites:

[0172] Laser PM sensor, used to collect raw scattered concentration;

[0173] Cameras and near-infrared imaging equipment to extract fractal dimensions and dust potential index in real time;

[0174] XRF spectroscopy module, used to regularly extract the calcium ratio;

[0175] Environmental instrument for simultaneous recording of ambient temperature and relative humidity.

[0176] The above six signals are normalized and concatenated in a fixed order to form a fused feature vector as the training data of the model.

[0177] 2. During each sampling period, the inhalable particulate matter in the air is collected simultaneously by gravimetric filter sampling and accurately weighed in the laboratory to obtain the actual mass concentration of that period.

[0178] 3. A pre-trained mass concentration inversion model is obtained based on the training data and the true mass concentration.

[0179] The mass concentration of dust cloud refers to the mass concentration of dust cloud particles at the construction site.

[0180] In one embodiment of the present invention, the mass concentration of the dust cloud is used in combination with the fractal dimension and effective density inversion to obtain the surface area concentration of the dust cloud. The surface area concentration is obtained by calculating the power law model as follows:

[0181]

[0182] in, represents the surface area concentration of the dust cloud, represents the concentration weight, Indicates the mass concentration of the dust cloud.

[0183] In fractal theory, the ratio of mass to surface area follows a nonlinear power-law relationship with structural complexity and density; the fractal dimension and effective density must first be combined into an exponent: . Used to describe the combined amplification or contraction effect of structural complexity and density on surface area.

[0184] The surface area and mass are not linearly proportional under the combined effects of fractal and effective density, but should be magnified or reduced according to a power law relationship. A power mapping is used to reflect the nonlinear magnitude of the mass → surface area conversion.

[0185] To accommodate different calibration conditions and unit systems, a calibration coefficient is introduced to adjust the overall output magnitude. The power law mapping result is linearly scaled using the concentration weight to obtain the surface area concentration.

[0186] It should be noted that according to the three-dimensional fractal theory, the particles have a fractal structure and their surface area and quality The relationship satisfies: , effective density factor After joining, the index is adjusted to , accurately reflecting the coupling of surface area and mass of porous high-calcium dust particles.

[0187] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A multimodal urban data fusion analysis system based on digital twins, characterized by: include: Data acquisition module, used to acquire construction images, laser PM data, and XRF spectrum data at fixed time intervals; Fractal dimension module, used to perform box counting method on construction images to obtain fractal dimension; Calcium ratio module, used to obtain calcium ratio using XRF spectrum data; Bare soil potential module, used to extract bare soil images from construction images and calculate the dust potential index based on the bare soil images; The activity coefficient module is used to extract the mean value of the dust potential index within the historical time period and determine the activity amplification coefficient based on the dust potential index; Density correction module, used to calculate effective density based on calcium ratio, combined with preset reference density and calcium density; A shape coefficient module, used to determine a shape correction coefficient based on the fractal dimension; The mass concentration module is used to invert the mass concentration of dust clouds based on laser PM data by combining effective density, activity amplification factor, and shape correction factor; The surface concentration module is used to obtain the surface area concentration of dust clouds by using the mass concentration of dust clouds, combined with fractal dimension and effective density inversion.

2. The multimodal urban data fusion analysis system based on digital twin according to claim 1 is characterized in that: The box counting method is performed on the construction images to obtain the fractal dimension, including: Convert all the time-series construction images into grayscale matrices; Selecting the mean of a preset number of time-series grayscale matrices to construct a reference grayscale matrix; Calculate the absolute difference matrix between each selected grayscale matrix and the reference grayscale matrix; Compare the i-th element in the absolute difference matrix with the preset difference threshold. If the i-th element is greater than the preset difference threshold, mark the i-th element as 1, otherwise mark it as 0 to generate a foreground mask matrix. Perform the 5-pixel kernel opening operation and the 9-pixel kernel closing operation on the foreground mask matrix to form a dust cloud binary matrix, where element 1 represents dust cloud and element 0 represents non-dust cloud; Set scale sequence Pixels; By scale Divide the dust cloud binary matrix and count the number of boxes covered by the dust cloud in the dust cloud binary matrix , forming a sequence ; For sequence Perform least squares regression and obtain the regression slope as the fractal dimension.

3. The multimodal urban data fusion analysis system based on digital twin according to claim 2 is characterized in that: Calcium content is obtained using XRF spectral data, including: The XRF spectrum data was differentially processed using a preset blank background spectrum to obtain a net spectrum; For the characteristic peak of calcium, a fitting window is set; the central energy of the characteristic peak of calcium is is 3.69keV, the fitting window 0.12keV; The area of ​​the characteristic peak of calcium element was extracted by least square fitting using Gaussian linear composite model; Perform the same fitting on the characteristic peaks of all elements in the net spectrum and accumulate them to obtain the sum of the peak areas of all elements; The calcium ratio was obtained by the total element peak area and the area of ​​the normalized calcium characteristic peak.

4. The multimodal urban data fusion analysis system based on digital twin according to claim 3 is characterized in that: The Gaussian linear composite model was used to extract the area of ​​the characteristic peak of calcium by least square fitting, including: In the net spectrum, locate and intercept the energy range of the characteristic peak of calcium , and get the calcium peak energy region spectrum ; Establish a Gaussian linear composite model as follows: in, represents the amplitude of the Gaussian peak to be fitted, represents the broadening of the Gaussian peak to be fitted, represents the slope of the linear term to be fitted, The intercept of the linear term to be fitted; The fitted residuals of the Gaussian linear composite model are as follows: The LM and gradient descent hybrid iterative algorithm is used to minimize the fitting residual of the Gaussian linear composite model to obtain the amplitude of the fitted Gaussian peak, the width of the fitted Gaussian peak, the slope of the fitted linear term, and the intercept of the fitted linear term; The calcium peak area is calculated using the integral analytical algorithm of the Gaussian distribution as follows: in, represents the calcium peak area, represents the amplitude of the fitted Gaussian peak, represents the broadening of the fitted Gaussian peak.

5. The multimodal urban data fusion analysis system based on digital twin according to claim 4 is characterized in that: Extract the bare soil image from the camera video stream and calculate the dust potential index based on the bare soil image, including: The grayscale matrix of the construction image is input into the pre-trained semantic segmentation network to form a bare soil binary matrix. The element 1 of the bare soil binary matrix represents bare soil, and the element 0 represents non-bare soil. Count the sum of the elements 1 in the bare soil binary matrix, and take the ratio of the sum to the number of elements in the bare soil binary matrix as the bare soil area ratio; Obtain the near-infrared channel component of the construction image and convert it into a near-infrared channel grayscale matrix; Based on element 1 of the bare soil binary matrix, the elements with the same coordinates in the near-infrared channel grayscale matrix are extracted and the element mean is calculated to obtain the average near-infrared grayscale of the bare soil; Determine the soil moisture content range in the construction area and calculate the soil dryness as follows: ； in, Indicates the dryness of the soil. Indicates the upper limit of the soil moisture content range, Indicates the lower limit of the soil moisture content range, represents the average near-infrared grayscale of bare soil, Indicates normalization processing; The product of soil dryness and the proportion of bare soil area is used as the dust potential index.

6. The multimodal urban data fusion analysis system based on digital twin according to claim 5 is characterized in that: The mean value of the dust potential index within the historical time period is extracted, and the activity amplification factor is determined in combination with the dust potential index. The activity amplification factor is the ratio of the dust potential index to the mean value of the dust potential index.

7. The multimodal urban data fusion analysis system based on digital twin according to claim 6 is characterized in that: Based on the calcium ratio, the effective density is calculated by combining the preset reference density and the calcium density. The effective density is obtained by linear difference as follows: in, represents the effective density, Indicates the preset reference density, represents the calcium element density, Indicates the calcium ratio.

8. The multimodal urban data fusion analysis system based on digital twin according to claim 7 is characterized in that: Determine the shape correction coefficient based on the fractal dimension, including: Read the preset fractal slope coefficient of dust ; Saturation clipping is performed based on the fractal dimension as follows: If the fractal dimension , , then the fractal dimension is maintained as ; Otherwise, the fractal dimension The value is or 3; According to the fractal dimension Calculate the shape correction factor as follows: in, Represents the shape correction factor.

9. The multimodal urban data fusion analysis system based on digital twin according to claim 8 is characterized in that: Based on the laser PM data, the mass concentration of the dust cloud is obtained by combining the effective density, activity amplification factor and shape correction factor, including: Collect ambient temperature and humidity; The ambient temperature and humidity are normalized with the laser PM data, effective density, activity amplification factor, and shape correction factor and then spliced ​​into a fusion feature vector; The fused feature vector is input into the pre-trained mass concentration inversion model to obtain the mass concentration of the dust cloud.

10. The multimodal urban data fusion analysis system based on digital twin according to claim 9 is characterized in that: The mass concentration of the dust cloud is used, combined with the fractal dimension and effective density inversion, to obtain the surface area concentration of the dust cloud. The surface area concentration is calculated using a power law model as follows: in, represents the surface area concentration of the dust cloud, represents the concentration weight, Indicates the mass concentration of the dust cloud.

Images