Remote sensing detection method of methane emissions based on ground-based hyperspectral imaging

Abstract

The present invention provides a methane emission remote sensing detection method based on ground-based hyperspectral imaging. First, auxiliary ground-based equipment such as a hyperspectral camera and a portable greenhouse gas analyzer are used to collect data. For the ground-based hyperspectral images, radiation calibration, signal denoising, and background denoising are performed for pre-processing to generate a ground-based hyperspectral image data set. Then, two methane emission detection methods are algorithmically implemented to obtain methane concentration quantification results and methane enhancement maps. The corrected and optimized methane enhancement images are subjected to pixel threshold selection to achieve level quantification, and finally scene application is achieved. The present invention addresses the typical characteristics of methane emission sources that are non-continuous and non-organized, and overcomes the problems of low signal-to-noise ratio of traditional point instruments, insufficient spatial coverage and spatiotemporal resolution of existing methane monitoring satellite data, and the inability of small methane emission sources to directly perform methane plume detection based on satellite data.

Description

Technical Field

[0001] The present invention belongs to the technical field of atmospheric environment remote sensing application, and in particular relates to a methane emission remote sensing detection method based on ground-based hyperspectral imaging. Background Art

[0002] Using remote sensing means to accurately evaluate and clarify the characteristics of methane emissions in the coal mining industry is essential for controlling and utilizing methane emissions generated by coal activities. The first step to understanding methane emissions is to identify the source of methane emissions. Methane emissions mainly take two forms, namely extreme point source emission events and area source emissions in key areas. Satellite remote sensing is the best means of conducting large-scale, high-revisit observations. With the growing demand for monitoring tasks for extreme emission events, remote sensing monitoring of abnormal methane emissions has gradually developed, and more satellites have begun to focus on point source detection in key areas. Methane anomaly detection technology based on satellite hyperspectral data inversion has become more mature. At present, in my country, remote sensing satellites represented by Sentinel-5P / TROPOMI, GHGSat-D / WAF-P, and GF-5 / AHSI have been successfully applied to the detection and quantification of methane emissions in the coal industry at regional and point source scales. In the future, the combination of point source satellites and regional satellites will be the technical development trend of using satellites to monitor methane.

[0003] However, the detection of extreme point source emissions has extremely strict requirements on the timeliness and spatial resolution of point source satellite data. The limited temporal and spatial resolution is an important reason for the uncertainty of the inversion results and accuracy; the lack of ability to accurately identify and locate such emission sources is a key reason for the low accuracy of the emission inventory obtained. How to reduce the uncertainty of this method requires further research and exploration. In addition, there are currently few studies on areas with complex surface types and similar high methane emissions, which are considered to be difficult areas for methane anomaly detection. The surface types of coal-rich areas in my country are usually very complex, and the standard matched filtering method is no longer applicable. This technology has large errors and uncertainties in detecting methane emissions caused by major coal mines in my country. Therefore, ground-based monitoring methods play an irreplaceable role in the monitoring and assessment of non-continuous and non-organized methane emissions. Summary of the invention

[0004] Ground-based monitoring methods play an irreplaceable role in the monitoring and assessment of non-continuous and non-organized methane emissions. In order to make up for the problems of insufficient coverage and spatiotemporal resolution of existing satellite data, as well as the adaptation limitations of methane anomaly detection methods to complex surface types, this paper proposes two optimization algorithms for anthropogenic methane emission detection based on ground-based hyperspectral remote sensing (i.e., SUMGHI methane detection algorithm and MLSIE methane detection algorithm), which can be used as an important means to test and improve the current methane emission detection technology, aiming to provide certain technical and data references for my country to carry out large-scale and rapid monitoring and verification of anthropogenic methane emissions.

[0005] Ground-based monitoring methods play an irreplaceable role in the monitoring and assessment of non-continuous and non-organized methane emissions. In order to make up for the problems of insufficient coverage and spatiotemporal resolution of existing satellite data, as well as the adaptation limitations of methane anomaly detection methods to complex surface types, this paper proposes two optimization algorithms for anthropogenic methane emission detection based on ground-based hyperspectral remote sensing (i.e., SUMGHI methane detection algorithm and MLSIE methane detection algorithm), which can be used as an important means to test and improve the current methane emission detection technology, aiming to provide certain technical and data references for my country to carry out large-scale and rapid monitoring and verification of anthropogenic methane emissions.

[0006] The present invention aims at the typical characteristics of methane emission sources being non-continuous and non-organized, overcomes the problems of low signal-to-noise ratio of traditional point instruments, insufficient spatial coverage and spatiotemporal resolution of existing methane monitoring satellite data, and inability to directly detect methane plumes from small methane emission sources based on satellite data. It provides a new methane emission detection method that utilizes ground-based remote sensing observation means and is suitable for man-made sources such as coal mines and coal fire areas with complex terrain.

[0007] To achieve the above objectives, a methane emission remote sensing detection method based on ground-based hyperspectral imaging includes the following steps: construction of ground-based methane dataset, SUMGHI methane detection and MLSIE methane detection;

[0008] The specific technical solutions are:

[0009] S1. Acquisition and construction of preliminary ground-based methane dataset

[0010] First, the instruments involved in acquiring the ground-based data set include: short-wave infrared hyperspectral camera, portable greenhouse gas analyzer, eddy covariance flux tower, GPS receiver and other ground-based auxiliary equipment; the ground-based observation sites mainly involve coal emission sources such as coal mines and coal fire areas, strong emission sources such as garbage transfer stations, gas stations, fertilizer plants, farmland platycodon, natural wetlands, as well as background control areas such as urban atmosphere and mountainous areas; finally, after the ground-based hyperspectral camera completes spectral acquisition, the image preprocessing process includes three core steps: radiation calibration, signal denoising and background denoising.

[0011] Step 1: Radiometric Calibration

[0012] This step performs system radiation correction on the original DN value recorded in the ground-based image data to obtain the radiation brightness value or apparent reflectivity;

[0013] Step 2: Signal denoising

[0014] Use forward / inverse PCA transformation or MNF minimum noise separation technology to remove residual noise; use filtering algorithm to smooth and denoise the spectral curve to highlight the spectral characteristics of the ground objects.

[0015] Step 3: Background Denoising

[0016] If the density function of the random variable X is:

[0017]

[0018] Then X is said to be subject to the parameters μ and σ 2 A normally distributed random variable, denoted by X~N(μ, σ 2 ).

[0019] Perform X~N(μ,σ) on the entire image 2 ) Normal fitting training, find μ0; remove the background mean, get the difference array x i , construct a ground-based hyperspectral image dataset.

[0020] in,

[0021] Further, step 1: radiation calibration, the specific method is:

[0022] The formula for converting the original DN value to radiant brightness is as follows:

[0023] L λ =Gain*DN+Offset

[0024] Where, L λ is the equivalent radiant brightness at the entrance pupil of the sensor equal load channel, the standard unit is W / (m 2 ·sr·μm), Gain and Offset are calibration coefficient gain offsets, and the standard unit is W / (m 2 ·sr·μm).

[0025] Radiance converted to apparent reflectance ρ λ The formula is as follows:

[0026]

[0027] Where, L λSame as above, d is the distance between the sun and the earth, which is a celestial unit, ESUN λ is the solar irradiance, θ is the solar altitude angle, which is complementary to the solar celestial angle. If the solar celestial angle is known, sinθ can be replaced by cosθ.

[0028] Furthermore, in step 2: signal denoising, the filtering algorithm is:

[0029] 9-point weighted moving average method:

[0030] The formula is as follows:

[0031] R i =0.04R i-4 +0.08R i-3 +0.12R i-2 +0.16R i-1 +0.20R i +0.16R i+1 +0.12R i+2 +0.08R i+3 +0.04R i+4

[0032] Among them, R i Refers to the original spectral reflectance / radiance value of the i-th band.

[0033] Or, Ensemble Empirical Mode Decomposition (EEMD):

[0034] The EEMD algorithm adds a Gaussian white noise sequence to the original signal, performs multiple EMD processes on the signal, and obtains the average value of each order of intrinsic mode function (IMF) as the final result.

[0035] The original data sequence X(t) of the EEMD algorithm is represented by the IMF component plus a mean or trend term:

[0036]

[0037] Among them, C i The n IMF components decomposed contain different frequency components of the signal from high frequency to low frequency. n The (t) component is essentially the trend component of the original signal, which represents the overall characteristics of the original signal.

[0038] S2. Methane detection based on spectral unmixing model of ground-based hyperspectral imagery (SUMGHI)

[0039] S2.1, performing spectral unmixing of the derivative spectral ratio of methane and anthropogenic emission sources on the image data to be detected;

[0040] S2.2. Determine the characteristic band range based on the methane spectrum ratio derivative curve and calculate the spectral absorption characteristic parameters;

[0041] S2.3. Perform correlation analysis on the measured methane concentration data and the above characteristic parameters to select the band range with high correlation;

[0042] S2.4, performing correlation analysis on the characteristic parameters of spectral absorption in each band to determine the independent variables of the fitting function;

[0043] S2.5. The coefficients of the spectral unmixing model are solved by multivariate fitting regression, and the corresponding data are substituted to finally calculate the quantitative estimate of the methane concentration.

[0044] Furthermore, in S2.1, the ratio derivative spectral unmixing algorithm establishes the following model:

[0045]

[0046] Where m is the spectral channel, r(λ m ) is in λ m The reflectivity at the wavelength, n is the number of end member components, C n is the abundance of the nth component in the mixture, ξ(λ m ) is the residual error value of the mth spectral channel.

[0047] When the pixel contains only two components, it is simplified to:

[0048] r(λ m )=C1×r1(λ m )+C2×r2(λ m )+ξ(λ m )

[0049] Divide both sides of the above equation by the spectrum of the first component to obtain the following equation:

[0050]

[0051] Taking the derivative on both sides of the equation, we get the following equation:

[0052]

[0053] In S2.1, the background pseudo-source area is calculated based on the absorption characteristics of methane itself. The reliability of the selected ROI is verified by measuring the spectral similarity characteristics (spectral angle SA and spectral information divergence SID, etc.). The specific method is as follows:

[0054] Assume that x and y represent the target spectrum and the test spectrum respectively, and their lengths are both n1.

[0055] Spectral Angle Mapping (SAM) is an algorithm based on the overall similarity of spectral curves. Its principle is to project the target spectrum and the test spectrum into space and calculate the angle between them. The smaller the value, the higher the similarity. It is used to calculate the similarity between two arrays. The formula is as follows:

[0056]

[0057] Among them, SAM (x, y) is the spectral angle between two spectra, and are the moduli of the two spectral vectors. During classification, the spectral angle between the unknown data and the known data is calculated, and the category of the unknown data is classified into the category corresponding to the minimum spectral angle.

[0058] Spectral information divergence (SID) is a spectrum classification method based on information theory to measure the difference between two spectra. The spectral vector is used as a random variable, and the similarity of two random vectors is analyzed based on probability statistics theory. That is, the smaller the value, the more similar the two sets of spectra are. The formula is as follows:

[0059] SID(x,y)=D(x||y)+D(y||x)

[0060] in, and Represent the probabilities of the two spectra, I i (x) = -lga i with I i (y) = -lgb i are the self-information values ​​of the two spectra obtained according to information theory. Similarly,

[0061] The spectral information divergence and spectral angle matching algorithm (SID_SA) takes into account the advantages of both SAM and SID, and can effectively calculate the similarity between spectra from the perspective of spectral shape and reflected energy difference. The calculation formula is as follows:

[0062] SID_SA(x,y)=SID(x,y)sin[SAM(x,y)]

[0063] The smaller the SID_SA value, the higher the spectral similarity, and vice versa, the lower the similarity between spectra.

[0064] In S2.2, the spectral absorption characteristic parameters (such as trough ν, wave depth D, shoulder width ω, area A, slope k and spectral absorption index SAI, etc.) are extracted from the hyperspectral methane spectrum ratio derivative curve, as described below:

[0065] Trough (ν): minimum amplitude in the spectral absorption valley;

[0066] Absorption depth (D): The absorption capacity of the trough, usually expressed in grayscale (0 to 255);

[0067] Shoulder width (ω): the width of the trough;

[0068] Area (A): the area enclosed by the trough and the shoulder width;

[0069] Slope (k): The degree of inclination of the shoulder width line relative to the x-axis;

[0070] Spectral Absorption Index (SAI): The ratio of the reflectance intensity of the non-absorbing baseline at the wavelength position of the spectral band to that at the valley bottom.

[0071] S2.5. In the above scheme, the ground-based methane emission unmixing model is tested and the root mean square error RMSE and the goodness of fit index R are calculated. new , the formula is as follows:

[0072]

[0073] Among them, y i is the actual methane content corresponding to the i-th group of verification spectrum data, is the predicted value of methane content in the i-th group of data obtained by the regression model, is the true methane content mean of the validation group spectrum, and n is the number of samples of the validation group data. The smaller the RMSE, the better the R new The closer it is to 1, the better the prediction effect of the model.

[0074] S3. Methane detection based on least squares image enhancement (MLSIE) using ground-based hyperspectral image correction

[0075] S3.1. Use the SCIATRAN radiation transfer model and the HITRAN spectral database to select the band range of the gas methane to be detected, and determine the wavenumber range of the internal points (methane interval) and external points (transparent interval) of the image to be detected;

[0076] S3.2, perform mean processing on external points;

[0077] S3.3, for each pixel, calculate the residual of the external point, the internal point and the above mean, and perform ratio processing (two-norm ratio) on the two;

[0078] S3.4. Add geographical location and other element information to the enhanced spectral pixel values ​​and draw a methane enhancement map.

[0079] Furthermore, in S3.1, spectral points near the band around 1653.7 nm are selected as the pre-selected range, and the corresponding Hyspex SWIR-384 ground-based image band is B130.

[0080] In S3.1, determine the spectral range of the high absorption wavelength of methane in the image to be detected, that is, the internal point (methane interval) and the spectral range of the low absorption wavelength, that is, the external point (transparent interval). The selected internal point (methane interval) wave number range is (5950~6050)cm -1 , the wave number range of the external point (transparent interval) is (6050~6014)cm -1 and (5991~5950)cm -1 .

[0081] In S3.2, the core content of the modified least squares image enhancement algorithm is the ratio operation of the L2-norm (residual sum of squares) of external points and internal points.

[0082] In Euclidean space, the vector The natural length of , that is, the distance from this point to the origin is:

[0083]

[0084] The residual is expressed as: δ = Σ(f(x i )-y i ) 2 . Norm is a function with the concept of "length", the maximum absolute value of the residual, that is, the maximum residual distance among all data points. L2-norm: Euclid norm, that is, the square root of the sum of the squares of the absolute values ​​of the vector elements.

[0085]

[0086] +∞ norm: the maximum of the absolute values ​​of all vector elements.

[0087]

[0088] Methane internal point y inn and methane internal point y out The L2-norm of is calculated as follows:

[0089]

[0090] Among them, q(y) is the zero-order polynomial fitting result of the internal points, and p(y) is the mean value of the external points.

[0091] The ratio of the two norms above is calculated to obtain the new spectral enhanced pixel value, which is recorded as result:

[0092] BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1It is the overall flow chart of the present invention;

[0094] Figure 2 for Figure 1 Flowchart of step 1 “dataset construction”;

[0095] Figure 3 for Figure 1 Flowchart of step 2 “algorithm implementation” in Figure 1. DETAILED DESCRIPTION

[0096] The present invention will be further described below in conjunction with the accompanying drawings.

[0097] like Figure 1 The main framework process of the anthropogenic methane emission detection method based on ground-based hyperspectral remote sensing is shown, which includes the following steps:

[0098] (1) In the early stage, ground-based methane data collection is carried out, and radiation calibration, background denoising and other preprocessing are performed on the ground-based hyperspectral images to generate a ground-based hyperspectral image dataset.

[0099] (2) The two methane emission detection methods are algorithmically implemented to obtain the methane concentration quantification results and methane enhancement diagrams.

[0100] (3) The pixel threshold of the corrected and optimized methane-enhanced image is selected to achieve level quantization, and finally realize scene application and standardization.

[0101] like Figure 2 The flowchart of the ground-based methane data set construction of the present invention is shown as a detailed description of the "data set construction" in step 1 of the overall flow chart, including the following steps:

[0102] (1) Ground-based hyperspectral camera completes spectral acquisition

[0103] (2) Radiation calibration (sensor calibration)

[0104] This step can use the HyspexRAD module to convert the calibrateddata_cor.hyspex into the calibrated radiancepixelvalues_cor_rad.hyspex, in units of W / (m 2 ·sr·nm).

[0105] (3) Signal denoising

[0106] In this step, the forward / inverse PCA transformation or MNF minimum noise separation technology can be used to remove the residual noise. Then, filtering algorithms such as bilateral filtering, weighted moving average method (9-point weighted average method), and ensemble empirical mode decomposition (EEMD) can be used to smooth and denoise the spectral curve to remove a small amount of noise contained in the signal and highlight the spectral characteristics of the ground object.

[0107] (4) Background denoising

[0108] Perform X~N(μ,σ) on the entire image 2 ) Normal fitting training, find μ0; remove the background mean, get the difference array x i , construct a ground-based hyperspectral image dataset.

[0109] like Figure 3 Shown is the implementation flow chart of the SUMGHI methane detection algorithm and the MLSIE methane detection algorithm, which serves as a detailed explanation of the "algorithm implementation" in step 2 of the overall flow chart.

[0110] The specific process of "SUMGHI" is as follows:

[0111] (1) Visually interpret the image to be detected, classify the background pseudo-source areas (such as mountain shadows with zero methane content, pseudo-coal fire areas such as grass, etc., the emission source here is explained using the coal fire area as an example) and establish ROIs. After spectral analysis, the spectral mean of each ROI is obtained. Then we get the coal fire spectrum background value

[0112] (2) According to the spectral absorption characteristics of methane itself, calculate the The reliability of the selected ROI is verified by measuring the spectral similarity characteristics (spectral angle SA and spectral information divergence SID, etc.);

[0113] (3) The coal fire spectrum background value obtained in step (1) As a divisor, the measured mixed methane spectrum image data is subjected to ratio derivative processing (first-order differential) to obtain the methane-coal fire spectrum ratio derivative unmixing result;

[0114] (4) In combination with the absorption characteristics of methane itself, the corresponding characteristic band range (I, II, III, IV) is selected on the ratio derivative curve of the methane spectrum obtained in step (3) and the spectral absorption characteristic parameters (such as trough ν, wave depth D, shoulder width ω, area A, slope k and spectral absorption index SAI, etc.) are extracted;

[0115] (5) performing correlation analysis on the spectral absorption characteristic parameters obtained in step (4) and the measured data of methane concentration, and selecting the band range (I, II) with higher overall correlation;

[0116] (6) Perform correlation analysis on the spectral absorption characteristic parameters in each band in step (5) to determine the correlation coefficient k i The parameter corresponding to the max value is used as the independent variable of the function to be fitted;

[0117] (7) Using the parameters determined in step (6) as independent variables and the methane content as the dependent variable, a regression equation is established through multivariate fitting regression (univariate, bivariate, quaternary) to solve the spectral unmixing model coefficients;

[0118] (8) Calculate the R of each spectral unmixing model in step (7) 2 and P value, and determine the group with the best fitting effect and strong representativeness as the methane emission unmixing model of ground-based coal fire hyperspectral;

[0119] (9) Select the remaining hyperspectral image sets as the validation group, substitute the corresponding data to calculate the quantitative estimation value of methane concentration, perform model test on the methane emission unmixing model in step (8), and calculate the root mean square error RMSE and the goodness of fit index R new .

[0120] The specific process of "MLSIE" is as follows:

[0121] (1) Using the SCIATRAN radiation transfer model to import the HITRAN spectral database, the spectral characteristics of common gases in coal spontaneous combustion air (or mine air, etc.) in the spectral band of 0.95-2.5μm covered by the short-wave infrared hyperspectral camera and the aerosol sensitivity analysis were simulated to select the specific band range of methane gas to be detected;

[0122] (2) According to step (1), determine the wave number range of the spectral segment with high methane absorption wavelength, i.e., the internal point (methane interval) and the spectral segment with low absorption wavelength, i.e., the external point (transparent interval), and obtain the spectral arrays of the internal point and the external point, which are recorded as y inn and out ;

[0123] (3) Calculate the internal point y in step (2) inn The zero-order polynomial fitting result of is denoted as q(y) and the external point y out The spectral mean of is denoted as p(y);

[0124] (4) For each pixel, calculate the residual between the external point and p(y) in step (3), denoted as z, and the residual between the internal point and q(y), denoted as e;

[0125] (5) According to the L2-norm calculation formula, the L2-norm of the external point and the internal point is calculated, and the ratio of the two is processed to fit the new spectral pixel value, which is recorded as result;

[0126] (6) adding geographical location and other element information to the spectral pixel values ​​obtained after the enhancement processing in step (5) to draw a methane enhancement map;

[0127] (7) Combined with the methane concentration quantification results obtained by the SUMGHI methane detection algorithm described in Algorithm 1, the emission source analysis and verification of the methane enhancement image in step (6) are performed, and the pixel threshold of the corrected and optimized methane enhancement image is selected to achieve level quantification, and finally achieve scene application and standardization.

[0128] The above is only a preferred embodiment of the present invention. It should be pointed out here that for researchers in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A remote sensing detection method for methane emissions based on ground-based hyperspectral imaging, characterized in that: The following steps are involved: Construction of ground-based methane dataset, SUMGHI methane detection and MLSIE methane detection; Combined with the methane concentration quantification results obtained by the SUMGHI methane detection algorithm, the emission source analysis and verification of the methane enhancement map are carried out, and the pixel threshold of the corrected and optimized methane enhancement image is selected to achieve level quantification, and finally achieve scene application and specification; Specific: S1. Acquisition and construction of preliminary ground-based methane dataset First, the instruments involved in acquiring the ground-based data set include: short-wave infrared hyperspectral camera, portable greenhouse gas analyzer, eddy covariance flux tower, GPS receiver and ground-based auxiliary equipment; The ground-based observation sites include coal emission sources in coal mines and coal fire areas, garbage transfer stations, gas stations, fertilizer plants, farmland platycodon, strong emission sources in natural wetlands, and urban atmosphere and mountain background control areas; Finally, after the ground-based hyperspectral camera completes spectrum acquisition, the image is preprocessed; S2. Methane detection based on the spectral unmixing model SUMGHI of ground-based hyperspectral images, including: S2.1, performing spectral unmixing of the derivative spectral ratio of methane and anthropogenic emission sources on the image data to be detected; S2.

2. Determine the characteristic band range based on the methane spectrum ratio derivative curve and calculate the spectral absorption characteristic parameters; S2.

3. Perform correlation analysis on the measured methane concentration data and the above characteristic parameters to select the band range with high correlation; S2.4, performing correlation analysis on the characteristic parameters of spectral absorption in each band to determine the independent variables of the fitting function; S2.5, solve the coefficients of the spectral unmixing model through multivariate fitting regression, substitute the corresponding data and finally calculate the quantitative estimation value of methane concentration; S3. Methane detection based on least squares image enhancement MLSIE with ground-based hyperspectral image correction, including: S3.

1. Use the SCIATRAN radiation transfer model and the HITRAN spectral database to select the band range of the gas methane to be detected and determine the wave number range of the internal and external points of the image to be detected; S3.2, perform mean processing on external points; S3.3, for each pixel, calculate the residual of the external point, the internal point and the above mean, and perform ratio processing on the two, that is, the two-norm ratio; S3.

4. Add geographic location element information to the enhanced spectral pixel values ​​and draw a methane enhancement map.

2. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 1 is characterized in that: The image preprocessing process in S1 includes: Step 1: Radiometric Calibration The original DN value recorded in the ground-based image data is subjected to system radiation correction to obtain the radiation brightness value or apparent reflectivity; Step 2: Signal denoising Use forward / inverse PCA transformation or MNF minimum noise separation technology to remove residual noise; use filtering algorithm to smooth and denoise the spectral curve to highlight the spectral characteristics of the ground object; Step 3: Background Denoising If the density function of a random variable x is: Then x is said to be subject to the parameters μ and σ 2 A normally distributed random variable, denoted by x~N(μ,σ 2 ); Perform x~N(μ,σ) on the entire image 2 ) Normal fitting training, find μ0; remove the background mean, get the difference array x i , construct a ground-based hyperspectral image dataset.

3. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 2 is characterized in that: Step 1: Radiation calibration, the specific method is: The formula for converting the original DN value to radiant brightness is as follows: L λ =Gain*DN+Offset Where, L λ is the equivalent radiant brightness at the entrance pupil of the sensor equal load channel, the standard unit is W / (m 2 ·sr·μm), Gain and offset are the calibration coefficient gain and offset respectively, and the standard unit is W / (m 2 ·sr·μm); Radiance converted to apparent reflectance ρ λ The formula is as follows: Where, L λ Same as above, d is the distance between the sun and the earth, which is a celestial unit, ESUN λ is the solar irradiance, θ is the solar altitude angle, which is complementary to the solar zenith angle. If the solar zenith angle is known, sinθ is replaced by cosθ.

4. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 2 is characterized in that: In the signal denoising of step 2, the filtering algorithm is: 9-point weighted moving average method: The formula is as follows: R i =0.04R i-4 +0.08R i-3 +0.12R i-2 +0.16R i-1 +0.20R i +0.16R i+1 +0.12R i+2 +0.08R i+3 +0.04R i+4 Among them, R i It refers to the original spectral reflectance / radiance value of the i-th band; Or, the ensemble empirical mode decomposition method EEMD: The EEMD algorithm adds a Gaussian white noise sequence to the original signal, performs multiple EMD processes on the signal, and obtains the average value of each order of intrinsic mode function IMF as the final result; The original data sequence X(t) of the EEMD algorithm is represented by the IMF component plus a mean or trend term: Among them, C i The n IMF components decomposed contain the different frequency components of the signal from high frequency to low frequency; r n The (t) component is essentially the trend component of the original signal, which represents the overall characteristics of the original signal.

5. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 1, characterized in that: In S2.1, the ratio derivative spectral unmixing algorithm establishes the following model: Where m is the spectral channel, r(λ m ) is in λ m The reflectivity at the wavelength, n is the number of end member components, C n is the abundance of the nth component in the mixture, ξ(λ m ) is the residual error value of the mth spectral channel; When the pixel contains only two components, it is simplified to: r(λ m )=C1×r1(λ m )+C2×r2(λ m )+ξ(λ m ) Divide both sides of the above equation by the spectrum of the first component to obtain the following equation: Taking the derivative on both sides of the equation, we get the following equation: In S2.1, the background pseudo-source area is calculated based on the absorption characteristics of methane itself. The spectral similarity measurement features, namely the spectral angle SA and the spectral information divergence SID, are used to verify the reliability of the selected ROI. The specific method is as follows: Assume that x and y represent the target spectrum and the test spectrum respectively, and both have a length of n1; Spectral Angle Mapping SAM is an algorithm based on the overall similarity of spectral curves. The target spectrum and the test spectrum are projected into space and the angle between them is calculated. The smaller the value, the higher the similarity. It is used to calculate the similarity between two arrays. The formula is as follows: Among them, SAM (x, y) is the spectral angle between two spectra, and are the moduli of the two spectral vectors respectively; during classification, the spectral angle between the unknown data and the known data is calculated, and the category of the unknown data is classified into the category corresponding to the minimum spectral angle; Spectral information divergence (SID) is a spectrum classification method based on information theory to measure the difference between two spectra. The spectral vector is taken as a random variable and the similarity of two random vectors is analyzed based on probability statistics theory. That is, the smaller the value, the more similar the two sets of spectra are. The formula is as follows: SID(x,y)=D(x||y)+D(y||x) in, and Represent the probabilities of the two spectra, I i (x) = -lga i with I i (y) = -lgb i are the self-information values ​​of the two spectra obtained according to information theory; similarly, Calculate the similarity between spectra using the following formula: SID_SA(x,y)=SID(x,y)sin[SAM(x,y)] The smaller the SID_SA value, the higher the spectral similarity, and vice versa, the lower the similarity between spectra.

6. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 1, characterized in that: In S2.2, the spectral absorption characteristic parameters are extracted from the hyperspectral methane spectrum ratio derivative curve, including trough v, wave depth D, shoulder width ω , area A, slope k and spectral absorbance index SAI: Valley v: minimum amplitude in the spectral absorption valley; Absorption depth D: the absorption capacity of the trough, expressed in grayscale; shoulder width ω : The width of the trough; Area A: the area enclosed by the trough and the shoulder width; Slope k: the inclination of the shoulder width line relative to the x-axis; Spectral Absorption Index SAI: The ratio of the reflection intensity of the non-absorption baseline at the wavelength position of the spectral band to that at the valley bottom.

7. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 1, characterized in that: In S2.5, the ground-based methane emission unmixing model is tested and the root mean square error RMSE and the goodness of fit index R are calculated. new , the formula is as follows: Among them, y i is the actual methane content corresponding to the i-th group of verification spectrum data, is the predicted value of methane content in the i-th group of data obtained by the regression model, is the true mean methane content of the validation group spectrum, and n is the number of samples of the validation group data; the smaller the RMSE, the higher the R new The closer it is to 1, the better the prediction effect of the model.

8. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 1, characterized in that: In S3.1, spectral points near the band around 1653.7 nm are selected as the pre-selected range.

9. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 1, characterized in that: In S3.1, determine the spectral range of the image to be detected with high methane absorption wavelength, i.e., the internal point, i.e., the methane interval, and the spectral range of the low absorption wavelength, i.e., the wavenumber range of the external point; the selected internal point wavenumber range is 5950~6050cm-1, and the external point wavenumber range is 6050~6014cm-1 and 5991~5950 c m- 1 .

10. The method for remote sensing detection of methane emissions based on ground-based hyperspectral imaging according to claim 1, characterized in that: In S3.3, the modified least squares image enhancement algorithm has the core content of the ratio operation of the L2-norm of the external point and the internal point, i.e., the residual sum of squares; In Euclidean space, the vector The natural length of , that is, the distance from this point to the origin is: The residual is expressed as: δ = ∑(f(x i )-y i ) 2 ; Norm norm is a function with the concept of "length", the maximum absolute value of the residual, that is, the maximum residual distance among all data points; L2-norm: Euclid norm, that is, the square root of the sum of the squares of the absolute values ​​of the vector elements; +∞ norm: the maximum absolute value of all vector elements; Methane internal point y inn and methane internal point y out The L2-norm of is calculated as follows: Among them, q(y) is the zero-order polynomial fitting result of the internal points, and p(y) is the mean value of the external points; The ratio of the two norms above is calculated to obtain the new spectral enhanced pixel value, which is recorded as result:

Images