Methane emission remote sensing intelligent detection method and system based on sparse spectrum reconstruction

Abstract

The invention provides a methane emission remote sensing intelligent detection method and system based on sparse spectrum reconstruction. The method comprises the following steps: performing first-order Taylor expansion on a Beer-Lambert law; introducing a spatial continuity constraint and a physical diffusion constraint term based on a wind field, and deducing an optimization objective function for inversion of atmospheric methane increment; simulating remote sensing observation hyperspectral radiation brightness values under different methane increment concentrations, and calculating to obtain a unit absorption characteristic spectrum; carrying out dimension reduction processing and clustering operation on the remote sensing hyperspectral image; performing matrix decomposition and sparse reconstruction on the spectrum value according to a clustering result; the hyperspectral image obtained after matrix sparse decomposition reconstruction is equivalent to a background spectral image, and atmospheric methane increment is obtained through calculation; generating an atmospheric methane plume sample based on a large eddy simulation program; constructing a training sample set of methane emission rate machine learning estimation; constructing a machine learning model for estimating the atmospheric methane emission rate; and based on an atmospheric methane increment inversion result, completing methane point source emission rate estimation of a real scene.

Description

Technical Field

[0001] The present invention relates to the field of methane gas emission monitoring, and in particular to a methane emission remote sensing intelligent detection method and system based on sparse spectrum reconstruction. Background Art

[0002] Methane in the atmosphere is an important greenhouse gas with a global warming potential far greater than that of carbon dioxide. Accurate monitoring of its abnormal emissions is of great significance for environmental and climate management. Remote sensing technology has been widely used in recent years to monitor atmospheric methane concentrations due to its advantages of large range, all-weather and non-contact. Matched filter technology is a method that compares the observed spectrum with the known characteristic spectrum of methane to detect and quantify its enhancement in the atmosphere. It is widely used in satellite and aerial remote sensing for monitoring gas emission sources and concentration inversion.

[0003] The matched filter algorithm separates the methane signal from the background spectrum by matching the methane absorption characteristics with the "observation-background" difference spectrum and maximizing the signal-to-noise ratio using a Gaussian model. Existing matched filter techniques face many challenges in methane incremental inversion. First, in traditional algorithms, the background spectrum is usually approximated using the column average spectrum of the image; however, this average spectrum method is difficult to effectively handle surface covers or atmospheric interference with similar methane absorption characteristics, resulting in a large number of misjudgments in complex surface types or low-emission scenarios. Secondly, satellite remote sensing has a lower signal-to-noise ratio than airborne measurements. Due to the decrease in detection sensitivity, the inverted methane plume is prone to problems such as discontinuity and blurred boundaries in space. Finally, the computational efficiency of existing algorithms is low, especially when multiple iterations are required to update the background spectrum, which is particularly time-consuming when processing large-scale data. These existing technical problems significantly affect the accuracy of plume emission rate estimation based on incremental inversion results.

[0004] The traditional integrated mass enhancement algorithm aims to use the instantaneous atmospheric methane concentration distribution inverted by remote sensing, combined with the spatial geometry of the plume and wind speed data, to estimate the average emission rate under the assumption that the emission source intensity remains unchanged. In this process, the atmospheric methane increment inverted by the above steps provides the total mass information of methane emissions, and the geometry of the plume diffusion and the observed wind speed provide the basis for the estimation of emission time information. However, the existing methods still have certain limitations in the accuracy and robustness of emission rate estimation: first, the complex surface and atmospheric conditions make the spatial distribution mode of methane plumes random and diverse, and there is a high uncertainty in quantifying the plume diffusion time by equivalent diffusion length and equivalent wind speed; second, the errors in input data such as methane increment and simulated wind speed can directly and linearly affect the accuracy of emission rate estimation. Therefore, this method is prone to obvious emission rate underestimation under low wind speed and low emission conditions.

[0005] In response to the above problems, there is an urgent need for a technical solution that can simultaneously solve the following problems: (1) False alarm and misestimation of incremental concentration in complex surface areas and low emission scenarios; (2) The decrease in detection sensitivity caused by the signal-to-noise ratio of remote sensing hyperspectral observation data; (3) The problem of underestimation of emission rates under low wind speed and low emission conditions; (4) The inefficiency of traditional iterative algorithms when processing large-scale data Patent document CN116559902A (application number: 202310451358.5) discloses a method and system for rapid remote sensing identification and flux estimation of abnormal near-surface methane emissions, including: using a radiation transfer model to calculate the unit absorption characteristic spectrum of near-surface methane and other trace gases; screening the best channel for near-surface methane increment inversion; constructing a spectrally similar pixel set and a reference background spectrum for the pixels to be inverted; using a matched filtering method to perform optimized iterative inversion of near-surface methane increments; using a morphological operation method to identify methane point source emission outlets and methane point source emission plumes; and calculating the methane point source emission flux in combination with the wind field information of the methane point source emission outlets. Summary of the invention

[0006] In view of the defects in the prior art, the object of the present invention is to provide a remote sensing intelligent detection method and system for methane emissions based on sparse spectral reconstruction.

[0007] According to the present invention, a methane emission remote sensing intelligent detection method based on sparse spectrum reconstruction is provided, comprising: Step S1: Perform a first-order Taylor expansion on the Beer-Lambert law and introduce regularization terms of spatial continuity and wind field diffusion constraints to obtain the objective function of atmospheric methane increment inversion; Step S2: simulating remote sensing observation hyperspectral radiation brightness data under different methane increment concentrations, and calculating the unit absorption characteristic spectrum of the methane increment; Step S3: performing dimensionality reduction and clustering operations on the remote sensing hyperspectral image, and performing matrix sparse decomposition and reconstruction on the spectral values ​​according to the clustering results to obtain a reconstructed hyperspectral image; Step S4: The reconstructed hyperspectral image is equivalent to a background spectrum image, and the objective function of atmospheric methane increment inversion is solved based on the background spectrum image and the unit absorption characteristic spectrum of the methane increment, so as to calculate the atmospheric methane increment inversion result; Step S5: Generate atmospheric methane incremental plume samples and construct a training sample set for machine learning estimation of methane emission rate; Step S6: construct a machine learning model for estimating atmospheric methane emission rate, and use the training sample set to train the machine learning model for estimating atmospheric methane emission rate to obtain a trained machine learning model for estimating atmospheric methane emission rate; use the trained machine learning model for estimating atmospheric methane emission rate to complete the estimation of methane point source emission rate in real scenarios based on the atmospheric methane increment inversion results.

[0008] Preferably, the step S1 comprises: Step S1.1: Perform a first-order Taylor expansion of the Beer-Lambert law:

[0009] in, Observe radiance data for remote sensing hyperspectral sensors, is the background spectrum without methane incremental absorption, is the equivalent optical thickness of methane gas; Step S1.2: Define the observed spectrum to consist of the background spectrum and the incremental spectrum:

[0010] in, is the increase in methane concentration relative to the background concentration, is the unit absorption characteristic spectrum; Step S1.3: Define the optimization objective function for solving the incremental methane concentration:

[0011] in, Indicates the remote sensing observation sample number, is the error covariance matrix of the observed spectrum, is the spatial continuity regularization term, is the regularization term of wind farm physical diffusion; Among them, the spatial continuity regularization term adopts:

[0012] in, is the Laplace filter kernel, is the weight parameter of the spatial continuity constraint; Indicates that spatial filtering operation is performed on the methane increment concentration; The physical diffusion regularization term is:

[0013] in, represents the wind vector field, It's the wind farm The weight in direction, It's the wind farm The weight in direction, represents the methane concentration increment field The partial derivative of is the weight parameter of the spatial continuity constraint.

[0014] Preferably, step S2 comprises: setting simulation conditions of the atmospheric radiation transmission mode based on the auxiliary observation conditions of the hyperspectral remote sensing data; changing the initial background concentration of methane to simulate and obtain hyperspectral radiation brightness data under different incremental concentrations; calculating the change in radiation brightness of each band corresponding to the change in unit methane increment to obtain the unit absorption characteristic spectrum of the methane increment; Among them, the auxiliary observation conditions based on hyperspectral remote sensing data include: the central wavelength of the satellite sensor, the full-peak half-width, the zenith angle and azimuth angle of the sun and the satellite, the atmospheric state and the surface elevation.

[0015] Preferably, the step S3 comprises: Step S3.1: Hyperspectral radiance image Perform principal component dimensionality reduction processing, including data standardization, calculation of covariance matrix, eigendecomposition of covariance matrix, and selection of the front The largest eigenvalues ​​and eigenvectors form the dimension reduction image ; Step S3.2: Use clustering algorithm to reduce the dimension of the image Perform ground object clustering to obtain the category label of each pixel ; Step S3.3: Spectral sample set composed of Perform matrix sparse decomposition:

[0016] in, yes The left singular matrix of contains the spatial structure information; is a diagonal matrix containing singular values; yes The right singular matrix of contains the spectral band information; Step S3.4: By selecting The largest singular values ​​are used to approximate the original matrix and achieve sparse processing:

[0017] in, , and It retains the previous The left singular matrix, diagonal matrix, and right singular matrix of singular values; Step S3.5: Reconstruct the spectral matrix of all categories Merge to obtain a complete reconstructed hyperspectral image .

[0018] Preferably, step S4 comprises: Step S4.1: Solving for methane increment The analytical solution form is:

[0019] Where R represents the reconstructed background spectrum, The unit absorption characteristic spectrum representing the methane increment; Step S4.2: Calculate the error covariance matrix of the initial observed spectrum : ; Where N represents the number of hyperspectral bands; Step S4.3: Correct the methane unit absorption characteristic spectrum:

[0020] in, is the dot product operator; Step S4.4: Calculate the initial methane increment :

[0021] Step S4.5: Update the error covariance matrix :

[0022] Step S4.6: Calculate spatial continuity constraints :

[0023]

[0024] Step S4.7: Calculate the physical diffusion constraint :

[0025] Step S4.8: Calculate the methane increment after introducing the constraint term :

[0026] Step S4.9: According to the objective function of atmospheric methane increment inversion, traverse the pixels in the hyperspectral image to obtain the methane increment inversion result.

[0027] Preferably, step S5 comprises: Step S5.1: Using large eddy simulation software, adopting the Smagorinsky SGS model, setting multi-dimensional hierarchical simulation parameters, and generating atmospheric methane incremental plume samples; The multi-dimensional hierarchical simulation parameters include: emission rate, wind speed, turbulence intensity, diffusion intensity, temperature, time step and space step; Step S5.2: based on the generated atmospheric methane incremental plume samples, extract plume morphological features and construct a training sample set for machine learning estimation of methane emission rate; Among them, the training sample set for machine learning estimation of methane emission rate includes: emission rate, wind speed, incremental concentration distribution characteristics and plume morphology characteristics of atmospheric methane incremental plume samples.

[0028] Preferably, step S6 comprises: Step S6.1: construct a machine learning model for estimating atmospheric methane emission rate;

[0029] in, represents the emission rate data from the atmospheric methane increment plume data, represents the wind speed in the simulated methane emission, , , and are the cumulative sum, mean, standard deviation and peak value of methane increment concentration, Indicates the plume morphological characteristics; Step S6.2: Based on the training sample set of the constructed methane emission rate machine learning estimation, model parameter training is performed using a machine learning technique with regression fitting capability including a random forest; Step S6.3: extract the real emission plume from the atmospheric methane increment inversion result, and use the trained machine learning model for estimating the atmospheric methane emission rate to obtain the emission rate estimation result corresponding to the real emission plume.

[0030] According to the present invention, a methane emission remote sensing intelligent detection system based on sparse spectrum reconstruction is provided, comprising: Module M1: Perform a first-order Taylor expansion on the Beer-Lambert law and introduce regularization terms of spatial continuity and wind field diffusion constraints to obtain the objective function of atmospheric methane increment inversion; Module M2: simulates remote sensing observation hyperspectral radiation brightness data under different methane increment concentrations, and calculates the unit absorption characteristic spectrum of methane increment; Module M3: Perform dimensionality reduction and clustering operations on remote sensing hyperspectral images, perform matrix sparse decomposition and reconstruction on spectral values ​​according to clustering results, and obtain reconstructed hyperspectral images; Module M4: The reconstructed hyperspectral image is equivalent to the background spectrum image, and the objective function of atmospheric methane increment inversion is solved based on the background spectrum image and the unit absorption characteristic spectrum of the methane increment, and the atmospheric methane increment inversion result is calculated; Module M5: Generate atmospheric methane incremental plume samples and construct a training sample set for machine learning estimation of methane emission rate; Module M6: Construct a machine learning model for estimating atmospheric methane emission rate, and use the training sample set to train the machine learning model for estimating atmospheric methane emission rate to obtain the trained machine learning model for estimating atmospheric methane emission rate; use the trained machine learning model for estimating atmospheric methane emission rate to complete the estimation of methane point source emission rate in real scenarios based on the atmospheric methane increment inversion results.

[0031] Preferably, the module M1 comprises: Module M1.1: Perform a first-order Taylor expansion of the Beer-Lambert law:

[0032] in, Observe radiance data for remote sensing hyperspectral sensors, is the background spectrum without methane incremental absorption, is the equivalent optical thickness of methane gas; Module M1.2: Define that the observed spectrum consists of the background spectrum and the incremental spectrum:

[0033] in, is the increase in methane concentration relative to the background concentration, is the unit absorption characteristic spectrum; Module M1.3: Define the optimization objective function for solving the incremental methane concentration:

[0034] in, Indicates the remote sensing observation sample number, is the error covariance matrix of the observed spectrum, is the spatial continuity regularization term, is the regularization term of wind farm physical diffusion; Among them, the spatial continuity regularization term adopts:

[0035] in, is the Laplace filter kernel, is the weight parameter of the spatial continuity constraint; Indicates that spatial filtering operation is performed on the methane increment concentration; The physical diffusion regularization term is:

[0036] in, represents the wind vector field, It's the wind farm The weight in direction, It's the wind farm The weight in direction, represents the methane concentration increment field The partial derivative of is the weight parameter of the spatial continuity constraint; The module M2 includes: setting the simulation conditions of the atmospheric radiation transmission mode based on the auxiliary observation conditions of the hyperspectral remote sensing data; changing the initial background concentration of methane to simulate and obtain the hyperspectral radiation brightness data under different incremental concentrations; calculating the change in the radiation brightness of each band corresponding to the change in the unit methane increment to obtain the unit absorption characteristic spectrum of the methane increment; The auxiliary observation conditions based on hyperspectral remote sensing data include: satellite sensor central wavelength, full peak half maximum width, zenith angle and azimuth angle of the sun and satellite, atmospheric state and surface elevation; The module M3 comprises: Module M3.1: Hyperspectral radiance images Perform principal component dimensionality reduction processing, including data standardization, calculation of covariance matrix, eigendecomposition of covariance matrix, and selection of the front The largest eigenvalues ​​and eigenvectors form the dimension reduction image ; Module M3.2: Use clustering algorithms to reduce the dimension of the image Perform ground object clustering to obtain the category label of each pixel ; Module M3.3: Spectral sample set composed of Perform matrix sparse decomposition:

[0037] in, yes The left singular matrix of contains the spatial structure information; is a diagonal matrix containing singular values; yes The right singular matrix of contains the spectral band information; Module M3.4: By selecting The largest singular values ​​are used to approximate the original matrix and achieve sparse processing:

[0038] in, , and It retains the previous The left singular matrix, diagonal matrix, and right singular matrix of singular values; Module M3.5: Reconstruct the spectral matrix of all classes Merge to obtain a complete reconstructed hyperspectral image .

[0039] Preferably, the module M4 comprises: Module M4.1: Solving for methane increments The analytical solution form is:

[0040] Where R represents the reconstructed background spectrum, The unit absorption characteristic spectrum representing the methane increment; Module M4.2: Calculate the error covariance matrix of the initial observed spectrum : ; Where N represents the number of hyperspectral bands; Module M4.3: Correction of the methane unit absorption characteristic spectrum:

[0041] in, is the dot product operator; Module M4.4: Calculate initial methane increment :

[0042] Module M4.5: Update the error covariance matrix :

[0043] Module M4.6: Computing spatial continuity constraints :

[0044]

[0045] Module M4.7: Computing Physical Diffusion Constraints :

[0046] Module M4.8: Calculation of methane increment after introducing constraints :

[0047] Module M4.9: According to the objective function of atmospheric methane increment inversion, traverse the pixels in the hyperspectral image to obtain the methane increment inversion result; The module M5 comprises: Module M5.1: Using large eddy simulation software, adopting the Smagorinsky SGS model, setting multi-dimensional hierarchical simulation parameters, and generating atmospheric methane incremental plume samples; The multi-dimensional hierarchical simulation parameters include: emission rate, wind speed, turbulence intensity, diffusion intensity, temperature, time step and space step; Module M5.2: Based on the generated atmospheric methane incremental plume samples, extract the plume morphological characteristics and construct a training sample set for machine learning estimation of methane emission rate; The training sample set for machine learning estimation of methane emission rate includes: emission rate, wind speed, incremental concentration distribution characteristics and plume morphology characteristics of atmospheric methane incremental plume samples; The module M6 comprises: Module M6.1: Construct a machine learning model for estimating atmospheric methane emission rates;

[0048] in, represents the emission rate data from the atmospheric methane increment plume data, represents the wind speed in the simulated methane emission, , , and are the cumulative sum, mean, standard deviation and peak value of methane increment concentration, Indicates the plume morphological characteristics; Module M6.2: Based on the training sample set of machine learning estimation of methane emission rate, model parameter training is performed using machine learning techniques with regression fitting capabilities including random forests; Module M6.3: Extract the real emission plume from the atmospheric methane increment inversion results. The real emission plume uses the trained machine learning model for estimating atmospheric methane emission rate to obtain the emission rate estimation result corresponding to the real emission plume.

[0049] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention solves the problem of inaccurate background spectrum estimation under complex surfaces by performing matrix decomposition and sparse reconstruction on observed spectral values; 2. The present invention overcomes the false alarm and misestimation problems of traditional methods in complex surface areas and low emission scenarios by constructing a more accurate background spectrum estimate; 3. The present invention optimizes the spatial structure and boundary clarity of the inversion plume and improves the signal-to-noise ratio of the inversion result by introducing regularization terms of spatial continuity and wind field diffusion constraints into the objective function of methane increment inversion; 4. The present invention reduces the sensitivity of methane increment inversion error and model wind speed error to the emission rate estimation result by constructing an atmospheric methane emission rate estimation model based on machine learning; 5. The present invention overcomes the problem of underestimation of emission rate in low wind speed and low emission scenarios by constructing an atmospheric methane emission rate estimation model based on machine learning; 6. The present invention significantly improves the computational efficiency of the algorithm by eliminating the iterative optimization process in the traditional matched filtering algorithm, and is suitable for rapid processing of large-scale data. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 This is a flow chart of the intelligent remote sensing detection method for methane emissions based on sparse spectral reconstruction.

[0051] Figure 2 Schematic diagram of the inversion effect of the methane emission remote sensing intelligent detection method based on sparse spectral reconstruction.

[0052] Figure 3 Schematic diagram of the emission rate estimation accuracy of the methane emission remote sensing intelligent detection method based on sparse spectral reconstruction.

[0053] Figure 4 A comparison of intelligent remote sensing detection methods for methane emissions based on sparse spectral reconstruction. DETAILED DESCRIPTION

[0054] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0055] Example 1 According to the present invention, a methane emission remote sensing intelligent detection method based on sparse spectrum reconstruction is provided, comprising: Step S1: According to the extinction process of light in the medium described by the Beer-Lambert law, a first-order Taylor expansion is performed, and a regularization term is introduced according to the spatial continuity constraint and the physical diffusion constraint based on the wind field, and the optimization objective function of inverting the atmospheric methane increment is derived; Step S2: using the existing atmospheric radiation transmission model, simulating the satellite remote sensing observation hyperspectral radiation brightness data under different methane increment concentrations, calculating the change in radiation brightness with the change in increment concentration, and obtaining the unit absorption characteristic spectrum of the methane increment; Step S3: Dimensionality reduction processing is performed on the hyperspectral remote sensing image using principal component analysis, objects are clustered based on the spectrum dimension reduction results, and the spectrum values ​​of each type of objects are matrix decomposed and sparsely reconstructed according to the clustering results to obtain the reconstructed spectrum of the remote sensing hyperspectral image; Step S4: According to the spatial sparsity principle of methane increment, the hyperspectral image reconstructed by matrix sparse decomposition is equivalent to the background spectral image, the optimization objective function of step S1 is solved, and the inversion result of atmospheric methane increment is calculated; Step S5: Using the existing large eddy simulation program, by setting key parameters such as emission rate, a large-scale multi-dimensional hierarchical atmospheric methane incremental plume sample is generated, and based on its emission rate, incremental concentration distribution, and spatial morphological characteristics, a training sample set for machine learning estimation of methane emission rate is constructed; Step S6: Construct a machine learning model for estimating atmospheric methane emission rate, use the training sample set obtained in step S5 to train the machine learning model parameters, apply the trained model to the atmospheric methane increment inversion result obtained in step S4, and complete the methane point source emission rate estimation in the real scene.

[0056] Specifically, the step S1 adopts: Step S1.1: Apply the Beer-Lambert law to the atmospheric radiation transfer process. Assuming that the radiation absorption caused by trace gas methane is small, perform a first-order Taylor expansion on the law:

[0057] in Observe radiance data for remote sensing hyperspectral sensors, is the background spectrum without methane incremental absorption, is the equivalent optical thickness of methane gas; Step S1.2: According to the first-order Taylor expansion form of the Beer-Lambert law, it is defined that the observed spectrum can be composed of the background spectrum and the incremental spectrum:

[0058] in is the increase in methane concentration relative to the background concentration, Unit absorption characteristic spectrum Step S1.3: Observe the spectrum by minimizing With simulated spectrum The difference between , define the optimization objective function for solving the incremental methane concentration :

[0059] in Indicates the remote sensing observation sample number, is the error covariance matrix of the observed spectrum, is the spatial continuity regularization term, is the regularization term of wind farm physical diffusion; The incremental spectrum adopts: methane unit absorption characteristic spectrum With incremental concentration The product of is approximately expressed as; The spatial continuity regularization term is:

[0060] in is the Laplace filter kernel, is the weight parameter of the spatial continuity constraint, Indicates that spatial filtering operation is performed on the methane increment concentration; The physical diffusion regularization term is:

[0061] in represents the wind vector field, It's the wind farm The weight in direction, It's the wind farm The weight in direction, represents the methane concentration increment field The partial derivative of is the weight parameter of the spatial continuity constraint.

[0062] Specifically, the step S2 adopts: Based on the auxiliary observation conditions of hyperspectral remote sensing data, the simulation conditions of the atmospheric radiation transmission model are set; the initial background concentration of methane is changed to simulate the hyperspectral radiation brightness data under different incremental concentrations; based on the simulation results, the change in radiation brightness of each band corresponding to the change in unit methane increment is calculated to obtain the unit absorption characteristic spectrum of the methane increment.

[0063] The auxiliary observation conditions include, but are not limited to: the central wavelength of the satellite sensor, the full-peak half-width, the zenith angle and azimuth angle of the sun and the satellite, the atmospheric state and the surface elevation, etc.

[0064] The radiation transmission modes include but are not limited to: MODTRAN, SCIATRAN, etc.

[0065] Specifically, the step S3 adopts: Step S3.1: Hyperspectral radiance image Perform principal component dimensionality reduction processing, including data standardization, calculation of covariance matrix, eigendecomposition of covariance matrix, and selection of the front The largest eigenvalues ​​and eigenvectors form the dimension reduction image ; Step S3.2: Use clustering algorithm to reduce the dimension of the image Perform ground object clustering to obtain the category label of each pixel ; Step S3.3: Spectral sample set composed of Perform matrix sparse decomposition:

[0066] in yes The left singular matrix of contains the spatial structure information, is a diagonal matrix containing singular values, yes The right singular matrix of contains the spectral band information; Step S3.4: By selecting The largest singular values ​​are used to approximate the original matrix and achieve sparse processing:

[0067] in , and It retains the previous The left singular matrix, diagonal matrix, and right singular matrix of singular values; Step S3.5: Reconstruct the spectral matrix of all categories Merge and restore the spatial structure of the original image to obtain a complete reconstructed hyperspectral image ; The hyperspectral image adopts: a hyperspectral image for methane increment inversion, including a two-dimensional spatial dimension and a spectral dimension, wherein the spectral dimension covers the range from visible light to short-wave infrared bands; The clustering algorithms include but are not limited to: k-means clustering algorithm and Gaussian mixture model clustering algorithm.

[0068] Specifically, the step S4 adopts: Step S4.1: Optimize the objective function according to step S1 , solve for the methane increment The analytical solution form is:

[0069] in, and They are the original observed hyperspectral image and the reconstructed hyperspectral image, and They are the spatial continuity constraint and the physical diffusion constraint based on the wind field.

[0070] Step S4.2: Based on the original observed hyperspectral image and reconstructing hyperspectral images , calculate the error covariance matrix of the initial observed spectrum :

[0071] Step S4.3: Correct the characteristic absorption spectrum of methane unit, where is the dot product operator:

[0072] Step S4.4: Calculate the initial methane increment :

[0073] Step S4.5: Initialize methane increment Update the error covariance matrix :

[0074] Step S4.6: Initialize methane increment Compute spatial continuity constraints :

[0075]

[0076] Step S4.7: Initialize methane increment Physical diffusion constraints for wind field calculations :

[0077] Step S4.8: Calculate the methane increment after introducing the constraint term :

[0078] Step S4.9: Traverse the pixels in the hyperspectral image to obtain the methane increment inversion result.

[0079] Specifically, the step S5 adopts: Step S5.1: Use the existing large eddy simulation software, adopt the Smagorinsky SGS model, set multi-dimensional hierarchical simulation parameters, and realize the generation of large-scale methane incremental plume samples; The multi-dimensional hierarchical simulation parameters are: Emission rate: 0~30 t / h; wind speed: 0~30m / s; turbulence intensity: 5~25%; diffusion intensity: 0~10 m 2 / s; temperature: 10~30 °C; time step: 1~5 steps; space step: 30 m Step S5.2: Based on the atmospheric methane incremental plume samples generated in the above steps, the plume morphological characteristics are extracted to construct a training sample set for machine learning estimation of methane emission rate.

[0080] The plume morphological characteristics include: total plume area, total plume circumference, expansion radius (XY direction), plume skewness (XY direction), plume kurtosis (XY direction), expansion radius aspect ratio, centroid and peak deviation (XY direction), incremental plume greater than threshold Covered area, plume coverage, average gradient (XY direction), average gradient, deviation between centroid and geometric center (XY direction), plume density, average distance between centroid, normalized shape index, eccentricity.

[0081] The training sample set for machine learning estimation of methane emission rate is composed of: emission rate, wind speed, incremental concentration distribution characteristics, and plume morphology characteristics extracted from atmospheric methane incremental plume samples.

[0082] Specifically, the step S6 adopts: Step S6.1: construct a machine learning estimation model for atmospheric methane emissions; The machine learning estimation model uses:

[0083] in, represents the emission rate data from the atmospheric methane increment plume data, represents the wind speed in the simulated methane emission, , , and are the cumulative sum, mean, standard deviation and peak value of methane increment concentration, represents the plume morphological characteristics mentioned in step S5.

[0084] Step S6.2: Based on the constructed methane emission rate training sample set, model parameter training is performed using machine learning techniques with regression fitting capabilities such as random forests; Step S6.3: extract the real emission plume from the atmospheric methane increment inversion result of step S4, apply the trained model to the real emission plume, and obtain the emission rate estimation result corresponding to the real emission plume.

[0085] The present invention also provides a methane emission remote sensing intelligent detection system based on sparse spectral reconstruction. The methane emission remote sensing intelligent detection system based on sparse spectral reconstruction can be realized by executing the process steps of the methane emission remote sensing intelligent detection method based on sparse spectral reconstruction, that is, those skilled in the art can understand the methane emission remote sensing intelligent detection method based on sparse spectral reconstruction as a preferred implementation of the methane emission remote sensing intelligent detection system based on sparse spectral reconstruction.

[0086] Example 2 Embodiment 2 is a preferred embodiment of Embodiment 1 According to a methane emission remote sensing intelligent detection method based on sparse spectrum reconstruction provided by the present invention, Figures 1 to 4 As shown, including: Step 1: For methane gas in the atmosphere with radiation absorption characteristics, a first-order Taylor expansion is performed based on the extinction process of light in the medium described by the Beer-Lambert law. Regularization terms are introduced based on spatial continuity constraints and physical diffusion constraints based on wind fields to derive the optimization objective function for inverting the atmospheric methane increment.

[0087] Step 1.1: Apply the Beer-Lambert law to the atmospheric radiation transfer process, and assume that the radiation absorption caused by trace gas methane is small, and perform a first-order Taylor expansion of the law:

[0088] in Observe the spectrum for remote sensing sensors with hyperspectral imaging capabilities, is the incident spectrum without methane absorption or the background spectrum without methane incremental absorption, is the optical density of methane gas.

[0089] Step 1.2: According to the first-order Taylor expansion of the Beer-Lambert law, assuming that the observed spectrum The background spectrum The methane incremental spectral absorption Composition:

[0090] in, is the increase in atmospheric methane concentration relative to the background concentration, is the unit absorption characteristic spectrum, background spectrum and methane optical depth The product of can be approximately expressed as the unit absorption characteristic spectrum of methane With incremental concentration The product of .

[0091] Step 1.3: Observe the spectrum by minimizing With simulated spectrum The difference between , define the optimization objective function for:

[0092] in Represents the serial number in the entire satellite remote sensing observation sample set, is the error covariance matrix of the observed spectrum Step 1.4: To ensure the spatial continuity of the inversion results of methane incremental concentration, a regularization term based on two-dimensional spatial gradient is further added to the optimization objective function, and the optimization objective function is updated as follows:

[0093]

[0094] in is the Laplace filter kernel, is the weight parameter of the spatial continuity constraint, Indicates that spatial filtering operation is performed on the methane increment concentration.

[0095] Step 1.5: To ensure that the inversion results of methane incremental concentration are consistent with the actual physical environment, a regularization term based on wind field diffusion is further added to the optimization objective function, and the optimization objective function is updated as follows:

[0096]

[0097] in represents the wind vector field, It's the wind farm The weight in direction, It's the wind farm The weight in direction, represents the methane concentration increment field The partial derivative of is the weight parameter of the spatial continuity constraint.

[0098] Step 2: Based on the existing open source atmospheric radiation transfer model, simulate the satellite remote sensing observation hyperspectral radiation brightness data under different methane increment concentrations, calculate the change in radiation brightness with the change in incremental concentration, and obtain the unit absorption characteristic spectrum of methane increment.

[0099] Step 2.1: Based on the auxiliary observation conditions of satellite hyperspectral remote sensing data (including but not limited to the central wavelength of the satellite sensor, the full-peak half-width, the zenith angle and azimuth angle of the sun and the satellite, the atmospheric state and the surface elevation, etc.), set the simulation conditions of the atmospheric radiation transfer model (including but not limited to MODTRAN, SCIATRAN mode).

[0100] Step 2.2: In the atmospheric state simulation conditions, change the initial background concentration of methane to simulate the satellite remote sensing observation hyperspectral radiation brightness data under different incremental concentrations.

[0101] Step 2.3: Based on the simulation results, calculate the change in radiation brightness of each band corresponding to a unit methane increment, and obtain the unit absorption characteristic spectrum of the methane increment .

[0102] Step 3: For the remote sensing hyperspectral images of methane increment to be inverted, principal component analysis is used to reduce the dimension of the remote sensing images from visible light to shortwave infrared bands, and the objects are clustered based on the spectral dimension reduction results. According to the clustering results, the spectral values ​​of each type of objects are matrix decomposed and sparsely reconstructed to obtain the reconstructed background spectrum of each pixel to be inverted in the remote sensing hyperspectral image.

[0103] Step 3.1: Prepare remote sensing hyperspectral images for methane increment inversion ,in and represents the spatial dimension of the image, Represents the spectral dimension of the image, which covers spectral information from visible light to short-wave infrared.

[0104] Step 3.2: Hyperspectral image Perform principal component analysis and dimensionality reduction, including data standardization, calculation of covariance matrix, eigenvalue decomposition of covariance matrix, obtain eigenvalues ​​and eigenvectors, and select the former The largest eigenvalue and the corresponding eigenvector constitute the hyperspectral image Dimensionality reduction image ,in represents the spectral dimension after spectral dimension reduction, The number is much smaller than .

[0105] Step 3.3: Set the number of cluster categories to , use clustering algorithms (including but not limited to k-means clustering algorithm and Gaussian mixture model clustering algorithm) to cluster the reduced-dimensional images Perform ground object clustering to obtain the category label of each pixel .

[0106] Step 3.4: For each type of feature , the original spectrum sample set composed of Perform matrix sparse decomposition:

[0107] in yes The left singular matrix contains the spatial structure information of the spectral data. is a diagonal matrix containing singular values ​​representing the energy distribution of the spectral data, yes The right singular matrix of contains the spectral band information.

[0108] Step 3.5: Select the front The largest singular values ​​are used to approximate the original matrix and achieve sparse processing to obtain Reconstructed spectral data of the pixel:

[0109] in, , and It retains the previous The left singular matrix, diagonal matrix, and right singular matrix of singular values.

[0110] Among them, the previous The largest singular value indicates that the singular value with variance explanation rate greater than 1% is selected.

[0111] Step 3.6: Reconstruct the spectral matrix of all categories Merge and restore the spatial structure of the original image to obtain a complete reconstructed hyperspectral image .

[0112] Step 4: According to the spatial sparsity principle of methane increment, the hyperspectral image reconstructed by matrix sparse decomposition is equivalent to the background spectral image, the optimization objective function of step 1 is solved, and the inversion result of atmospheric methane increment is calculated.

[0113] Step 4.1: Optimize the objective function according to step 1 , solve for the methane increment The analytical solution form is:

[0114] in and They are the original observed hyperspectral image and the reconstructed hyperspectral image, and They are the spatial continuity constraint and the physical diffusion constraint based on the wind field.

[0115] Step 4.2: Based on the original observed hyperspectral image and reconstructing hyperspectral images , calculate the error covariance matrix of the initial observed spectrum :

[0116] Step 4.3: Correct the methane unit absorption characteristic spectrum, where is the dot product operator:

[0117] Step 4.4: Calculate the initial methane increment based on the above variables :

[0118] Step 4.5: Initialize methane increment according to Update the error covariance matrix :

[0119] Step 4.6: Initialize methane increment according to Compute spatial continuity constraints :

[0120]

[0121] Step 4.7: Initialize methane increment according to Calculate physical diffusion constraints with wind field information :

[0122] Step 4.8: Calculate the methane increment after introducing the constraint term :

[0123] Step 4.9: Traverse each pixel in the hyperspectral image to obtain the atmospheric methane increment inversion result of the same size as the image space Step 5: Generate large-scale, multi-dimensional and hierarchical atmospheric methane incremental plume samples by setting key parameters of large eddy simulation such as emission rate, atmospheric conditions, and wind speed. Extract the emission rate, incremental concentration distribution, and spatial morphological characteristics of the methane plume samples to construct a training sample set for machine learning estimation of methane emission rate.

[0124] Step 5.1: Use the existing open source fluid dynamics simulation software with large eddy simulation function, adopt the Smagorinsky SGS model, set multi-dimensional hierarchical simulation parameters, and realize the generation of large-scale methane incremental plume samples.

[0125] The multi-dimensional hierarchical simulation parameters include: emission rate: 0~30 t / h, wind speed: 0~30 m / s, turbulence intensity: 5~25%, diffusion intensity: 0~10 m 2 / s, temperature: 10-30 °C, time step: 1-5 steps, space step: 30 m; Step 5.2: Based on the atmospheric methane incremental plume generated in the above steps, extract the plume morphological characteristics and construct a training sample set for machine learning estimation of methane emission rate.

[0126] The plume morphological characteristics include: Total plume area: ,in, Indicates the actual geographical distance represented by the image pixel, Represents the plume pixel coordinates.

[0127] Total circumference of plume: ,in, is the number of vertical or horizontal adjacent pixels of the plume boundary pixel, is the number of pairs of diagonally adjacent pixels.

[0128] Extension radius (XY direction): , , , ,in, is the coordinate of the methane plume The incremental concentration, represents the cumulative incremental concentration of the complete plume, represents the centroid of the methane plume.

[0129] Plume deflection (XY direction): , .

[0130] Plume Kurtosis (XY direction): , .

[0131] Expanded Radius Aspect Ratio: .

[0132] Centroid and peak deviation (XY direction): , ,in, Represents the coordinates of the peak point of methane plume concentration.

[0133] Incremental plume greater than threshold Area covered: .

[0134] Plume coverage: ,in, Represents the area of ​​the minimum enclosing rectangle.

[0135] Average gradient (XY direction): , ,in, It means to average the results of all pixels in the plume.

[0136] Average gradient: .

[0137] Deviation between the center of mass and the geometric center (XY direction): , ,in, Represents the coordinates of the geometric center of the methane plume concentration.

[0138] Plume density: .

[0139] Average distance to centroid: .

[0140] Normalized shape index: .

[0141] Eccentricity: , , ,in, and is the covariance matrix The characteristic value of .

[0142] Step 5.3: Extract the emission rate, wind speed, incremental concentration distribution characteristics, and plume morphology characteristics of the atmospheric methane incremental plume samples to form a training sample set for machine learning estimation of methane emission rate.

[0143] Step 6: Construct a machine learning model for estimating atmospheric methane emission rate. Use the training sample set obtained in step 5 to train the machine learning model parameters. Apply the trained model to the atmospheric methane increment inversion results obtained in step 4 to obtain the corresponding emission rate.

[0144] Step 6.1: Build a machine learning estimation model for atmospheric methane emissions:

[0145] in, represents the emission rate data from the atmospheric methane increment plume data, represents the wind speed in the simulated methane emission, , , and are the cumulative sum, mean, standard deviation and peak value of methane increment concentration, Represents the plume morphological characteristics mentioned in step 5.

[0146] Step 6.2: Based on the constructed methane emission rate training sample set, use machine learning techniques with regression fitting capabilities such as random forests to train model parameters.

[0147] Step 6.3: Using 200 ppm·m as the threshold for distinguishing ambient noise from methane emissions, extract the real emission plume from the atmospheric methane increment inversion result in step 4, apply the trained model to the real emission plume, and obtain the corresponding emission rate estimation result.

[0148] like Figure 4 As shown, the atmospheric methane monitoring effects of the traditional method and the method of the present invention are compared: Among them, the results obtained by the traditional method have more spatial noise, while the results obtained by the method of the present invention can more clearly identify the methane emission plume; the deviation between the results obtained by the traditional method and the true value is larger, while the deviation between the results obtained by the method of the present invention and the true value is closer to 0; there are more false alarm increments in the results obtained by the traditional method, while the results obtained by the method of the present invention have fewer false alarm increments.

[0149] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0150] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A remote sensing intelligent detection method for methane emissions based on sparse spectral reconstruction, characterized in that: include Step S1: Perform a first-order Taylor expansion on the Beer-Lambert law and introduce regularization terms of spatial continuity and wind field diffusion constraints to obtain the objective function of atmospheric methane increment inversion; Step S2: simulating remote sensing observation hyperspectral radiation brightness data under different methane increment concentrations, and calculating the unit absorption characteristic spectrum of the methane increment; Step S3: performing dimensionality reduction and clustering operations on the remote sensing hyperspectral image, and performing matrix sparse decomposition and reconstruction on the spectral values ​​according to the clustering results to obtain a reconstructed hyperspectral image; Step S4: The reconstructed hyperspectral image is equivalent to a background spectrum image, and the objective function of atmospheric methane increment inversion is solved based on the background spectrum image and the unit absorption characteristic spectrum of the methane increment, so as to calculate the atmospheric methane increment inversion result; Step S5: Generate atmospheric methane incremental plume samples and construct a training sample set for machine learning estimation of methane emission rate; Step S6: construct a machine learning model for estimating atmospheric methane emission rate, and use the training sample set to train the machine learning model for estimating atmospheric methane emission rate to obtain a trained machine learning model for estimating atmospheric methane emission rate; use the trained machine learning model for estimating atmospheric methane emission rate to complete the estimation of methane point source emission rate in real scenarios based on the atmospheric methane increment inversion results.

2. The method for remote sensing intelligent detection of methane emissions based on sparse spectrum reconstruction according to claim 1 is characterized in that: The step S1 comprises: Step S1.1: Perform a first-order Taylor expansion of the Beer-Lambert law: in, Observe radiance data for remote sensing hyperspectral sensors, is the background spectrum without methane incremental absorption, is the equivalent optical thickness of methane gas; Step S1.2: Define the observed spectrum to consist of the background spectrum and the incremental spectrum: in, is the increase in methane concentration relative to the background concentration, is the unit absorption characteristic spectrum; Step S1.3: Define the optimization objective function for solving the incremental methane concentration: in, Indicates the remote sensing observation sample number, is the error covariance matrix of the observed spectrum, is the spatial continuity regularization term, is the regularization term of wind farm physical diffusion; Among them, the spatial continuity regularization term adopts: in, is the Laplace filter kernel, is the weight parameter of the spatial continuity constraint; Indicates that spatial filtering operation is performed on the methane increment concentration; The physical diffusion regularization term is: in, represents the wind vector field, It's the wind farm The weight in direction, It's the wind farm The weight in direction, represents the methane concentration increment field The partial derivative of is the weight parameter of the spatial continuity constraint.

3. The method for remote sensing intelligent detection of methane emissions based on sparse spectrum reconstruction according to claim 1 is characterized in that: The step S2 comprises: setting the simulation conditions of the atmospheric radiation transmission mode based on the auxiliary observation conditions of the hyperspectral remote sensing data; changing the initial background concentration of methane to simulate and obtain the hyperspectral radiation brightness data under different incremental concentrations; calculating the change in the radiation brightness of each band corresponding to the change in the unit methane increment to obtain the unit absorption characteristic spectrum of the methane increment; Among them, the auxiliary observation conditions based on hyperspectral remote sensing data include: the central wavelength of the satellite sensor, the full-peak half-width, the zenith angle and azimuth angle of the sun and the satellite, the atmospheric state and the surface elevation.

4. The method for remote sensing intelligent detection of methane emissions based on sparse spectrum reconstruction according to claim 1 is characterized in that: The step S3 comprises: Step S3.1: Hyperspectral radiance image Perform principal component dimensionality reduction processing, including data standardization, calculation of covariance matrix, eigendecomposition of covariance matrix, and selection of the front The largest eigenvalues ​​and eigenvectors form the dimension reduction image ; Step S3.2: Use clustering algorithm to reduce the dimension of the image Perform ground object clustering to obtain the category label of each pixel ; Step S3.3: Spectral sample set composed of Perform matrix sparse decomposition: in, yes The left singular matrix of contains the spatial structure information; is a diagonal matrix containing singular values; yes The right singular matrix of contains the spectral band information; Step S3.4: By selecting The largest singular values ​​are used to approximate the original matrix and achieve sparse processing: in, , and It retains the previous The left singular matrix, diagonal matrix, and right singular matrix of singular values; Step S3.5: Reconstruct the spectral matrix of all categories Merge to obtain a complete reconstructed hyperspectral image .

5. The method for remote sensing intelligent detection of methane emissions based on sparse spectrum reconstruction according to claim 1 is characterized in that: The step S4 comprises: Step S4.1: Solving for methane increment The analytical solution form is: Where R represents the reconstructed background spectrum, The unit absorption characteristic spectrum representing the methane increment; Step S4.2: Calculate the error covariance matrix of the initial observed spectrum : ； Where N represents the number of hyperspectral bands; Step S4.3: Correct the methane unit absorption characteristic spectrum: in, is the dot product operator; Step S4.4: Calculate the initial methane increment : Step S4.5: Update the error covariance matrix : Step S4.6: Calculate spatial continuity constraints : Step S4.7: Calculate physical diffusion constraints : Step S4.8: Calculate the methane increment after introducing the constraint term : Step S4.9: According to the objective function of atmospheric methane increment inversion, traverse the pixels in the hyperspectral image to obtain the methane increment inversion result.

6. The method for remote sensing intelligent detection of methane emissions based on sparse spectrum reconstruction according to claim 1 is characterized in that: The step S5 comprises: Step S5.1: Using large eddy simulation software, adopting the Smagorinsky SGS model, setting multi-dimensional hierarchical simulation parameters, and generating atmospheric methane incremental plume samples; The multi-dimensional hierarchical simulation parameters include: emission rate, wind speed, turbulence intensity, diffusion intensity, temperature, time step and space step; Step S5.2: based on the generated atmospheric methane incremental plume samples, extract plume morphological features and construct a training sample set for machine learning estimation of methane emission rate; Among them, the training sample set for machine learning estimation of methane emission rate includes: emission rate, wind speed, incremental concentration distribution characteristics and plume morphology characteristics of atmospheric methane incremental plume samples.

7. The method for remote sensing intelligent detection of methane emissions based on sparse spectrum reconstruction according to claim 1 is characterized in that: The step S6 comprises: Step S6.1: construct a machine learning model for estimating atmospheric methane emission rate; in, represents the emission rate data from the atmospheric methane increment plume data, represents the wind speed in the simulated methane emission, , , and are the cumulative sum, mean, standard deviation and peak value of methane increment concentration, Indicates the plume morphological characteristics; Step S6.2: Based on the training sample set of the constructed methane emission rate machine learning estimation, model parameter training is performed using a machine learning technique with regression fitting capability including a random forest; Step S6.3: extract the real emission plume from the atmospheric methane increment inversion result, and use the trained machine learning model for estimating the atmospheric methane emission rate to obtain the emission rate estimation result corresponding to the real emission plume.

8. A methane emission remote sensing intelligent detection system based on sparse spectrum reconstruction, characterized in that: include Module M1: Perform a first-order Taylor expansion on the Beer-Lambert law and introduce regularization terms of spatial continuity and wind field diffusion constraints to obtain the objective function of atmospheric methane increment inversion; Module M2: simulates remote sensing observation hyperspectral radiation brightness data under different methane increment concentrations, and calculates the unit absorption characteristic spectrum of methane increment; Module M3: Perform dimensionality reduction and clustering operations on remote sensing hyperspectral images, perform matrix sparse decomposition and reconstruction on spectral values ​​according to clustering results, and obtain reconstructed hyperspectral images; Module M4: The reconstructed hyperspectral image is equivalent to the background spectrum image, and the objective function of atmospheric methane increment inversion is solved based on the background spectrum image and the unit absorption characteristic spectrum of the methane increment, and the atmospheric methane increment inversion result is calculated; Module M5: Generate atmospheric methane incremental plume samples and construct a training sample set for machine learning estimation of methane emission rate; Module M6: Construct a machine learning model for estimating atmospheric methane emission rate, and use the training sample set to train the machine learning model for estimating atmospheric methane emission rate to obtain the trained machine learning model for estimating atmospheric methane emission rate; use the trained machine learning model for estimating atmospheric methane emission rate to complete the estimation of methane point source emission rate in real scenarios based on the atmospheric methane increment inversion results.

9. The methane emission remote sensing intelligent detection system based on sparse spectrum reconstruction according to claim 8 is characterized in that: The module M1 comprises: Module M1.1: Perform a first-order Taylor expansion of the Beer-Lambert law: in, Observe radiance data for remote sensing hyperspectral sensors, is the background spectrum without methane incremental absorption, is the equivalent optical thickness of methane gas; Module M1.2: Define that the observed spectrum consists of the background spectrum and the incremental spectrum: in, is the increase in methane concentration relative to the background concentration, is the unit absorption characteristic spectrum; Module M1.3: Define the optimization objective function for solving the incremental methane concentration: in, Indicates the remote sensing observation sample number, is the error covariance matrix of the observed spectrum, is the spatial continuity regularization term, is the regularization term of wind farm physical diffusion; Among them, the spatial continuity regularization term adopts: in, is the Laplace filter kernel, is the weight parameter of the spatial continuity constraint; Indicates that spatial filtering operation is performed on the methane increment concentration; The physical diffusion regularization term is: in, represents the wind vector field, It's the wind farm The weight in direction, It's the wind farm The weight in direction, represents the methane concentration increment field The partial derivative of is the weight parameter of the spatial continuity constraint; The module M2 includes: setting the simulation conditions of the atmospheric radiation transmission mode based on the auxiliary observation conditions of the hyperspectral remote sensing data; changing the initial background concentration of methane to simulate and obtain the hyperspectral radiation brightness data under different incremental concentrations; calculating the change in the radiation brightness of each band corresponding to the change in the unit methane increment to obtain the unit absorption characteristic spectrum of the methane increment; The auxiliary observation conditions based on hyperspectral remote sensing data include: satellite sensor central wavelength, full peak half maximum width, zenith angle and azimuth angle of the sun and satellite, atmospheric state and surface elevation; The module M3 comprises: Module M3.1: Hyperspectral radiance images Perform principal component dimensionality reduction processing, including data standardization, calculation of covariance matrix, eigendecomposition of covariance matrix, and selection of the front The largest eigenvalues ​​and eigenvectors form the dimension reduction image ; Module M3.2: Use clustering algorithms to reduce the dimension of the image Perform ground object clustering to obtain the category label of each pixel ; Module M3.3: Spectral sample set composed of Perform matrix sparse decomposition: in, yes The left singular matrix of contains the spatial structure information; is a diagonal matrix containing singular values; yes The right singular matrix of contains the spectral band information; Module M3.4: By selecting The largest singular values ​​are used to approximate the original matrix and achieve sparse processing: in, , and It retains the previous The left singular matrix, diagonal matrix, and right singular matrix of singular values; Module M3.5: Reconstruct the spectral matrix of all classes Merge to obtain a complete reconstructed hyperspectral image .

10. The methane emission remote sensing intelligent detection system based on sparse spectrum reconstruction according to claim 8 is characterized in that: The module M4 comprises: Module M4.1: Solving for methane increment The analytical solution form is: Where R represents the reconstructed background spectrum, The unit absorption characteristic spectrum representing the methane increment; Module M4.2: Calculate the error covariance matrix of the initial observed spectrum : ； Where N represents the number of hyperspectral bands; Module M4.3: Correction of the methane unit absorption characteristic spectrum: in, is the dot product operator; Module M4.4: Calculate initial methane increment : Module M4.5: Update the error covariance matrix : Module M4.6: Computing spatial continuity constraints : Module M4.7: Computing Physical Diffusion Constraints : Module M4.8: Calculation of methane increment after introducing constraints : Module M4.9: According to the objective function of atmospheric methane increment inversion, traverse the pixels in the hyperspectral image to obtain the methane increment inversion result; The module M5 comprises: Module M5.1: Using large eddy simulation software, adopting the Smagorinsky SGS model, setting multi-dimensional hierarchical simulation parameters, and generating atmospheric methane incremental plume samples; The multi-dimensional hierarchical simulation parameters include: emission rate, wind speed, turbulence intensity, diffusion intensity, temperature, time step and space step; Module M5.2: Based on the generated atmospheric methane incremental plume samples, extract the plume morphological characteristics and construct a training sample set for machine learning estimation of methane emission rate; The training sample set for machine learning estimation of methane emission rate includes: emission rate, wind speed, incremental concentration distribution characteristics and plume morphology characteristics of atmospheric methane incremental plume samples; The module M6 comprises: Module M6.1: Construct a machine learning model for estimating atmospheric methane emission rates; in, represents the emission rate data from the atmospheric methane increment plume data, represents the wind speed in the simulated methane emission, , , and are the cumulative sum, mean, standard deviation and peak value of methane increment concentration, Indicates the plume morphological characteristics; Module M6.2: Based on the training sample set of machine learning estimation of methane emission rate, model parameter training is performed using machine learning techniques with regression fitting capabilities including random forests; Module M6.3: Extract the real emission plume from the atmospheric methane increment inversion results. The real emission plume uses the trained machine learning model for estimating atmospheric methane emission rate to obtain the emission rate estimation result corresponding to the real emission plume.

Images