Hyperspectral remote sensing image methane point source detection method based on superpixel matched filtering

Abstract

The invention provides a hyperspectral remote sensing image methane point source detection method based on superpixel matched filtering. The hyperspectral remote sensing image methane point source detection method comprises the following steps: acquiring an original hyperspectral remote sensing image and a standard methane spectrum absorptivity; obtaining a true color image according to the original hyperspectral remote sensing image, and obtaining a background spectrum according to the true color image; constructing a methane target spectrum according to the standard methane spectrum absorptivity and the background spectrum; filtering the methane target spectrum and the original hyperspectral image by adopting a matched filtering algorithm to obtain a methane point source filtering result; and obtaining a final methane point source position according to a methane point source filtering result. The methane point source detection complexity can be reduced, and the detection precision can be improved.

Description

Technical Field

[0001] The present invention relates to the field of atmospheric technology, and more specifically, to a method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering. Background Art

[0002] As one of the main greenhouse gases, accurate monitoring of methane has important environmental and social significance. Currently, methane detection technology focuses on detection sensors, traditional estimation methods and deep learning methods.

[0003] Short-wave infrared sensors are more sensitive to changes in near-ground methane concentrations than thermal infrared sensors and are now widely used in multi-scale methane monitoring. Among them, the data volume and temporal resolution advantages of multispectral data make it applicable to long-term point source monitoring, but its accuracy and detection limit for methane point sources are far behind those of hyperspectral data, and it is often used in combination with hyperspectral data. Hyperspectral remote sensing images have high potential in methane point source detection and concentration inversion due to their fine and nearly continuous spectral resolution. Traditional estimation methods mainly include physical algorithms, CO2 proxy methods, and matched filtering methods. The physical algorithm is based on the radiation transfer equation and is greatly affected by weather and surface illumination. The CO2 proxy method is a type of proxy algorithm. Since both CO2 and methane have absorption characteristics at 1.6 μm, CO2 can be used to perform concentration inversion on the common absorption band, and CO2 can be used as a reference to eliminate the optical path changes caused by scattering and correct the methane concentration. The matched filtering method directly obtains methane enhancement rather than the methane concentration of the entire area. Compared with the above methods, it has higher computational efficiency and is suitable for point source detection tasks. Deep learning has been widely used in the environmental field, including solar energy forecasting, wind energy forecasting, etc., but its application in methane emission detection is relatively lacking. Current methane detection deep networks mostly use CNN and other infrastructure.

[0004] In summary, the development trend of detection sensors in methane monitoring shows high spatial, high spectral, high temporal resolution and multi-source remote sensing combined research, which provides a relatively complete data source for multi-scale methane emission monitoring. Traditional methods such as physical algorithms, CO2 proxy methods and matched filtering methods are still widely used. The effects of each method in specific applications are different and need to be flexibly selected according to data and goals. Methane detection deep networks have certain potential in methane emission detection tasks, but their current application scope is relatively small. Summary of the invention

[0005] In order to overcome the above-mentioned defect that the current methane point source detection accuracy is not high, the present invention provides a methane point source detection method based on hyperspectral remote sensing images based on superpixel matched filtering.

[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0007] The present invention proposes a method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering, which comprises the following steps:

[0008] Obtain original hyperspectral remote sensing images and standard methane spectral absorbance;

[0009] A true color image is obtained based on the original hyperspectral remote sensing image, and a background spectrum is obtained based on the true color image;

[0010] Construct a methane target spectrum according to the standard methane spectrum absorbance and background spectrum;

[0011] The matched filter algorithm is used to filter the methane target spectrum and the original hyperspectral image to obtain the methane point source filtering result;

[0012] The final methane point source location is obtained based on the methane point source filtering results.

[0013] Preferably, acquiring a hyperspectral remote sensing image, and obtaining a true color image based on the hyperspectral remote sensing image includes synthesizing the hyperspectral remote sensing image through RGB bands to obtain a true color image.

[0014] Preferably, obtaining the background spectrum according to the true color image includes segmenting the background blocks of the true color image using a superpixel segmentation algorithm, and performing superpixel aggregation on the segmented blocks to obtain the background spectrum.

[0015] Preferably, obtaining the standard methane spectrum absorbance includes:

[0016] Based on the MODTRAN platform, the standard atmospheric temperature curve was selected as the median atmospheric model for simulation, and the absorption characteristics of methane in the short-wave infrared range were obtained as the standard methane spectral absorbance.

[0017] Preferably, using a matched filtering algorithm to filter the methane target spectrum and the original hyperspectral image to obtain a methane point source filtering result includes:

[0018]

[0019] Among them, P is the distribution probability of methane point source, that is, the methane point source detection filtering result, x p is the methane enhancement value of the methane point source, and the probabilistic processing result is the methane point source detection filtering result of the test data.

[0020] Preferably, obtaining the maximum possible methane point source location according to the methane point source filtering result includes:

[0021] The methane point source filtering results are subjected to Gaussian filtering and eight-field feature estimation in turn to obtain the most likely methane point source location.

[0022] Preferably, the step of sequentially performing Gaussian filtering and eight-domain feature estimation on the methane point source filtering result includes:

[0023] The methane point source filtering result is centered on each pixel, and the average grayscale value of all pixels in the 3×3 area is taken as the grayscale value of the center point. A two-dimensional Gaussian function is used as the weight distribution method, and the Gaussian kernel generated by the Gaussian function is used to perform a convolution operation on the image. Specifically:

[0024]

[0025] Among them, x and y represent the two-dimensional plane coordinate position of the pixel, σ is the standard deviation, and the separation characteristics of the Gaussian function are used for optimization:

[0026]

[0027] Preferably, the step of sequentially performing Gaussian filtering and eight-domain feature estimation on the methane point source filtering result further comprises:

[0028] After Gaussian filtering, the eight-domain feature estimation is performed on the image after Gaussian filtering based on the eight-connected neighborhood labeling algorithm. The statistics of the pixel values ​​in the eight-connected neighborhood can be expressed as:

[0029] N8(P)={(x+i,y+j)|i,j∈{-1,0,1}}\(x,y)

[0030] For any pixel in the image, its eight-connected neighborhood N8(P) is defined as the set of eight pixels including P(X,Y) and its surrounding pixels:

[0031] The specific algorithm steps are as follows:

[0032] 1) Region marking: traverse each pixel in the image and use a marking function f(P) to record the region to which each pixel belongs;

[0033] 2) Isolated point detection: For each pixel point P(X,Y), check its eight-connected neighborhood N8(P). If P belongs to a specific category and there are no other pixels belonging to the same category in its eight-connected neighborhood, then P is considered an isolated point.

[0034] 3) Isolated point removal: The pixel value of the detected isolated point P is set to the background value of the point or the average value of the neighboring pixels. Here, the isolated point is set to the background value. The mathematical expression is as follows:

[0035] Let I(X,Y) represent the pixel value of the image at the coordinate (X,Y), and B(X,Y) represent the background pixel value. For each pixel point P(X,Y), the statistics of the pixel values ​​in its eight-connected neighborhood can be expressed as:

[0036]

[0037] If P is an isolated point, then S(P) will not contain pixel values ​​of the same category as P. In this case, the pixel value of P can be updated as:

[0038]

[0039] The present invention also proposes a hyperspectral remote sensing image methane point source detection system based on superpixel matched filtering, which is used to implement the above method, including:

[0040] A data acquisition module, used to obtain original hyperspectral remote sensing images and standard methane spectrum absorbance;

[0041] A true color image and background spectrum acquisition module is used to obtain a true color image based on an original hyperspectral remote sensing image and to obtain a background spectrum based on the true color image;

[0042] A target spectrum building module, used for building a methane target spectrum according to a standard methane spectrum absorbance and a background spectrum;

[0043] A filtering module is used to filter the methane target spectrum and the original hyperspectral image using a matched filtering algorithm to obtain a methane point source filtering result;

[0044] The result acquisition module is used to obtain the final methane point source location according to the methane point source filtering result.

[0045] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0046] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0047] The present invention proposes a method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering. By introducing superpixel segmentation technology, the initial hyperspectral remote sensing image is divided into multiple sub-regions, thereby simplifying the texture features of the image, reducing the computational complexity and improving the analysis efficiency. Secondly, the image is smoothed by Gaussian filtering, and combined with the statistics of the eight-connected regional features, the background interference is effectively suppressed and isolated points are removed. The present invention can reduce the complexity of methane point source detection and improve the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the process of the method described in Example 1;

[0049] Figure 2 is a system block diagram of the method described in Example 2;

[0050] Figure 3 The original hyperspectral image of the method described in Example 2;

[0051] Figure 4 The matched filtering result of the method described in Example 2;

[0052] Figure 5 The Gaussian filtering result of the method described in Example 2;

[0053] Figure 6 The final methane point source position of the method described in Example 2;

[0054] Figure 7 This is an enlarged view of the final methane point source position of the method described in Example 2;

[0055] Figure 8 This is a schematic diagram of the structure of the system described in Example 3. DETAILED DESCRIPTION

[0056] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0057] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0058] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0059] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0060] Example 1

[0061] This embodiment provides a method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering. Figure 1 As shown, the following steps are included:

[0062] Obtain original hyperspectral remote sensing images and standard methane spectral absorbance;

[0063] A true color image is obtained based on the original hyperspectral remote sensing image, and a background spectrum is obtained based on the true color image;

[0064] Construct a methane target spectrum according to the standard methane spectrum absorbance and background spectrum;

[0065] The matched filter algorithm is used to filter the methane target spectrum and the original hyperspectral image to obtain the methane point source filtering result;

[0066] The final methane point source location is obtained based on the methane point source filtering results.

[0067] In the specific implementation process, the original hyperspectral remote sensing image and the standard methane spectrum absorbance are first obtained, and then the background spectrum is obtained based on the original hyperspectral remote sensing image. Then, the methane target spectrum is constructed based on the standard methane spectrum absorbance and the background spectrum. The matched filtering algorithm is used to filter the methane target spectrum and the original hyperspectral image to obtain the methane point source filtering result. Finally, the final methane point source position is obtained based on the methane point source filtering result.

[0068] Example 2

[0069] This embodiment provides a method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering, including the following steps:

[0070] Obtain original hyperspectral remote sensing images and standard methane spectral absorbance;

[0071] A true color image is obtained based on the original hyperspectral remote sensing image, and a background spectrum is obtained based on the true color image;

[0072] Construct a methane target spectrum according to the standard methane spectrum absorbance and background spectrum;

[0073] The matched filter algorithm is used to filter the methane target spectrum and the original hyperspectral image to obtain the methane point source filtering result;

[0074] The final methane point source location is obtained based on the methane point source filtering results.

[0075] like Figure 2 As shown in FIG. 1 , the method described in this embodiment consists of four steps: using GF-5 02AHSI data as the original hyperspectral image, the hyperspectral image is first divided into superpixels using a superpixel segmentation and aggregation strategy. Secondly, the standard methane spectrum absorbance simulated by MODTRAN is fused with the background spectrum. Then, the matched filter algorithm is used to search for methane emission points in each superpixel block. Finally, Gaussian filtering and feature statistical analysis are used to obtain the final methane emission point source location.

[0076] The specific contents are as follows:

[0077] (1) MODTRAN methane point source spectrum simulation

[0078] AHSI is one of the main payloads of GF5-02 satellite, which can obtain spectral information of ground objects with a width of 60km, a spatial resolution of 30m and a spectral range of 400~2500nm (spectral resolution: 5-10nm), totaling 330 spectral channels.

[0079] This embodiment is based on the MODTRAN platform (http: / / modtran.spectral.com / ) and selects the 1976 U.S. standard atmospheric temperature curve as an effective median atmospheric model for simulation. The transmission characteristics and absorption characteristics of methane in the atmospheric greenhouse gas in the short-wave infrared range are obtained by the atmospheric radiation transmission simulation in MODTRAN.

[0080] Among them, the standard methane spectrum absorbance and transmittance can be converted by the following formula:

[0081] n(β A )=-lg[n(β T )] (1)

[0082] In the formula, β A and β T represents the absorption and transmission coefficients.

[0083] Methane exhibits a strong absorption feature near 2300nm, and the 2100-2450nm window of GF-5 02AHSI is more sensitive to XCH4 changes, so this embodiment uses the spectral characteristics of this window band to detect the location of methane point sources.

[0084] (2) Superpixel segmentation and aggregation

[0085] Superpixel segmentation refers to dividing an initial image into multiple sub-regions (a collection of pixels). In order to simplify the texture of the image and facilitate analysis, superpixel segmentation algorithms are widely used in the field of computer vision. In this embodiment, the entropy rate segmentation (ERS) algorithm is used to aggregate pixels of the true color image of the hyperspectral image to generate a two-dimensional superpixel pattern. Specifically, the image is mapped to an undirected graph as shown below:

[0086] G=(V,E) (2)

[0087] Where the vertex set V represents the pixels in the initial image, and the edge set E consists of the similarities between pixel pairs defined by the weight function. Next, by selecting a subset of edges The graph is divided into several connected subgraphs, so that the resulting graph G′=(V,O) is composed of smaller connected subgraphs. Therefore, a clustering objective function based on graph theory is proposed, and the entropy rate term H(·) and the balance term B(·) are introduced to optimize superpixel segmentation:

[0088]

[0089] Where H(O) tends to form uniform and compact clusters, while B(O) is used to force clusters to have similar sizes. The parameter α is introduced to balance the contribution of the equilibrium term and the entropy rate term. In addition, the entropy rate H(·) of a random walk on G′=(V,O) can be expressed as a set of functions:

[0090] H(O)=-∑ i u i ∑ j p i,j (O)logp i,j (O) (4)

[0091] In the formula, p i,j (O) represents the transition probability of a particle when it randomly walks from vertex i to vertex j. i,j represents the weight of the edge connecting vertices i and j. The sum of the weights of the edges passing through vertex i. The steady-state distribution is given by u i =w i / w T Calculated, among which V is the number of vertices.

[0092] This embodiment uses superpixel segmentation and aggregation algorithms to synthesize the GF-5 observation data through RGB bands to obtain true color images. For true color images, the above superpixel segmentation algorithm is used to segment the background plots, and the segmented plots are superpixel aggregated to obtain the background spectrum. Finally, the methane target spectrum is constructed in combination with the methane spectrum absorbance established by the MODTRAN simulation platform.

[0093] (3) Matched filtering

[0094] The matched filter algorithm has been successfully applied in the inversion of methane point source column concentration of aviation and aerospace hyperspectral sensors. Based on this, this embodiment uses GF5-02-AHSI data and an improved matched filter algorithm to carry out the inversion of methane point source column concentration. The core idea of ​​the matched filter algorithm is to regard each input spectral radiation value as the superposition of the background radiation brightness without methane enhancement and the radiation disturbance caused by the change in methane concentration. This disturbance is modeled as the target spectral radiation, which represents the radiation increment transfer signal of methane absorption per unit area. The target parameter of methane point source inversion is the methane point source column concentration. Relative to the surrounding background X b The pixel-by-pixel enhancement is The specific calculation form is as follows:

[0095]

[0096] Where, L iis the input observed radiation value, μ and C are the mean and covariance of the background radiation, t is the target spectral radiation disturbance, which indicates the disturbance value relative to the background radiation signal caused by the enhanced methane concentration, and T indicates transposition. The target spectral radiation disturbance t is obtained by multiplying the unit methane absorption spectrum s by the background radiation mean μ. The unit methane absorption spectrum s is obtained using the atmospheric radiation transfer simulation of the MODTRAN platform. Since the real surface is a non-Lambertian body, the bidirectional reflection characteristics of the surface are obvious. The empirical Minnaert coefficient k is introduced to describe the bidirectional reflection characteristics of the surface. When k = 1, it means that the surface is a Lambertian body, and the decrease in the k value indicates an increase in the surface anisotropy:

[0097] L H =L T (cosθ s / cosγ s ) k (6)

[0098] Where, L H is the observed radiance under a flat surface, L T is the observed radiance under the mountain surface. s is the cosine of the solar incident zenith angle, cosγ s is the cosine of the solar zenith angle relative to the incident sun.

[0099] The spectral absorption rate adjustment factor obtained by MODTRAN simulation is fused with the background spectrum obtained by superpixel aggregation to obtain the target spectrum, and the target characteristics are extracted from the target spectrum. Subsequently, the spectral matching filter is performed with the original GF5 observation data to obtain the probability of methane point source distribution. The probabilistic formula is as follows:

[0100]

[0101] In the formula, x p is the methane enhancement value of the methane point source, and the probabilistic processing result is the methane point source detection filtering result of the test data.

[0102] (4) Background suppression and isolated point removal

[0103] According to prior knowledge, the location where methane point sources are emitted usually shows high concentration in the center, and the concentration gradually decreases as the distance from the center increases. Therefore, some isolated points in the obtained methane enhancement value image may be caused by problems with the data itself, and such points are not point sources to be extracted. Therefore, this embodiment uses Gaussian filtering and feature statistics of the eight-connected area to complete background suppression and isolated point removal.

[0104] 1) Gaussian filtering algorithm: With each pixel as the center, take the average grayscale value of all pixels in the 3×3 area as the grayscale value of the center point, and adopt the idea of ​​weighted average. The closer the point is, the greater the weight is, and the farther the point is, the smaller the weight is. Based on this, this embodiment uses a two-dimensional Gaussian function as a weight distribution method, and uses the Gaussian kernel (Gaussian template) generated by the Gaussian function to perform a convolution operation on the image, as follows:

[0105]

[0106] In the formula, x and y represent the two-dimensional plane coordinate position of the pixel, and σ is the standard deviation. And the separation characteristics of the Gaussian function are used for optimization:

[0107]

[0108] 2) Removal of eight-connected isolated points: After Gaussian filtering, the noise in the image is effectively suppressed, but some isolated pixels may be generated, which are visually inconsistent with the surrounding environment. In order to remove these isolated points, this embodiment adopts a region labeling algorithm based on eight connectivity. The eight-connected region refers to a connected region consisting of a pixel and its eight adjacent pixels around it (including horizontal, vertical and diagonal directions) in a two-dimensional image. This connectivity definition is used to identify and analyze continuous objects in images. Eight connectivity means that a pixel can be connected to adjacent pixels in the horizontal, vertical or diagonal directions. The mathematical definition of eight connectivity is expressed as follows:

[0109] N8(P)={(x+i,y+j)|i,j∈{-1,0,1}}\(x,y) (10)

[0110] For any pixel in the image, its eight-connected neighborhood N8(P) is defined as the set of eight pixels including P(X,Y) and its surrounding pixels:

[0111] The specific algorithm steps are as follows:

[0112] 1. Region marking: Traverse each pixel in the image and use a marking function f(P) to record the region to which each pixel belongs.

[0113] 2. Isolated point detection: For each pixel point P(X,Y), check its eight-connected neighborhood N8(P). If P belongs to a specific category (for example, a boundary pixel after edge detection) and there are no other pixels belonging to the same category in its eight-connected neighborhood, then P is considered an isolated point.

[0114] 3. Isolated point removal: The pixel value of the detected isolated point P is set to the background value of the point or the average value of the neighboring pixels to achieve the removal of isolated points. Here, the isolated point is set to the background value. Its mathematical expression is as follows:

[0115] Let I(X,Y) represent the pixel value of the image at coordinate (X,Y), and B(X,Y) represent the background pixel value. For each pixel point P(X,Y), the statistics of the pixel values ​​in its eight-connected neighborhood can be expressed as:

[0116]

[0117] If P is an isolated point, then S(P) will not contain pixel values ​​of the same category as P. In this case, the pixel value of P can be updated as:

[0118]

[0119] This embodiment uses a Gaussian filtering algorithm to process the image. Gaussian filtering can effectively reduce random noise in the image while maintaining the edge features of the image and avoiding excessive blurring. Taking into account the actual characteristics of methane distribution, more clusters appear rather than independent distributions. Therefore, based on Gaussian filtering, the image is further subjected to a feature statistical analysis of the eight-connected areas. Through the above steps, this embodiment can effectively remove the eight-connected isolated points in the image, thereby improving the overall quality and consistency of the image. Connectivity is used to remove the obtained scattered methane points, and the connected methane sample points are retained to improve detection accuracy.

[0120] This embodiment uses GF5-02-AHSI data as the original hyperspectral image. Figure 3 As shown in the figure, it is the original hyperspectral remote sensing image, which marks 6 detection areas A1-A6. These areas are detected, such as Figure 4 As shown in, it is the detection probability of methane point source after matched filtering, such as Figure 5 As shown, this is the structure diagram after Gaussian filtering. Figure 4 In terms of Figure 5 Reduces random noise in the image and maintains the edge features of the image, such as Figure 6 Shown is the final methane point source location, Figure 7 for Figure 6 The enlarged image of the corresponding detection area, the bright area in the image is the location of the methane point source.

[0121] This embodiment also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering are implemented.

[0122] Example 3

[0123] This embodiment provides a hyperspectral remote sensing image methane point source detection system based on superpixel matched filtering, which is used to implement the method described in embodiment 1 or 2, such as Figure 8 As shown, including:

[0124] A data acquisition module, used to obtain original hyperspectral remote sensing images and standard methane spectrum absorbance;

[0125] A true color image and background spectrum acquisition module is used to obtain a true color image based on an original hyperspectral remote sensing image and to obtain a background spectrum based on the true color image;

[0126] A target spectrum building module, used for building a methane target spectrum according to a standard methane spectrum absorbance and a background spectrum;

[0127] A filtering module is used to filter the methane target spectrum and the original hyperspectral image using a matched filtering algorithm to obtain a methane point source filtering result;

[0128] The result acquisition module is used to obtain the final methane point source location according to the methane point source filtering result.

[0129] The same or similar reference numerals correspond to the same or similar components;

[0130] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;

[0131] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering, characterized in that: The following steps are involved: Obtain original hyperspectral remote sensing images and standard methane spectral absorbance; A true color image is obtained based on the original hyperspectral remote sensing image, and a background spectrum is obtained based on the true color image; Construct a methane target spectrum according to the standard methane spectrum absorbance and background spectrum; The matched filter algorithm is used to filter the methane target spectrum and the original hyperspectral image to obtain the methane point source filtering result; The final methane point source location is obtained based on the methane point source filtering results.

2. The method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering according to claim 1 is characterized in that: Acquire a hyperspectral remote sensing image, and obtain a true color image based on the hyperspectral remote sensing image, including synthesizing the hyperspectral remote sensing image through RGB bands to obtain a true color image.

3. The method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering according to claim 2 is characterized in that: Obtaining the background spectrum according to the true color image includes segmenting the background blocks of the true color image using a superpixel segmentation algorithm, and performing superpixel aggregation on the segmented blocks to obtain the background spectrum.

4. The method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering according to claim 3 is characterized in that: Obtaining the standard methane spectrum absorbance includes: Based on the MODTRAN platform, the standard atmospheric temperature curve was selected as the median atmospheric model for simulation, and the absorption characteristics of methane in the short-wave infrared range were obtained as the standard methane spectral absorbance.

5. The method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering according to claim 4 is characterized in that: The matched filtering algorithm is used to filter the methane target spectrum and the original hyperspectral image to obtain the methane point source filtering results, including: P=1 / (1+e -xp ) Among them, P is the distribution probability of methane point source, that is, the methane point source detection filtering result, x p is the methane enhancement value of the methane point source, and the probabilistic processing result is the methane point source detection filtering result of the test data.

6. The method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering according to claim 5 is characterized in that: The most likely methane point source locations obtained based on the methane point source filtering results include: The methane point source filtering results are subjected to Gaussian filtering and eight-field feature estimation in turn to obtain the most likely methane point source location.

7. The method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering according to claim 6 is characterized in that: The methane point source filtering result is subjected to Gaussian filtering and eight-field feature estimation in sequence, including: The methane point source filtering result is centered on each pixel, and the average grayscale value of all pixels in the 3×3 area is taken as the grayscale value of the center point. A two-dimensional Gaussian function is used as the weight distribution method, and the Gaussian kernel generated by the Gaussian function is used to perform a convolution operation on the image. Specifically: Among them, x and y represent the two-dimensional plane coordinate position of the pixel, σ is the standard deviation, and the separation characteristics of the Gaussian function are used for optimization:

8. The method for detecting methane point sources in hyperspectral remote sensing images based on superpixel matched filtering according to claim 7 is characterized in that: The step of sequentially performing Gaussian filtering and eight-domain feature estimation on the methane point source filtering results also includes: After Gaussian filtering, the eight-domain feature estimation is performed on the image after Gaussian filtering based on the eight-connected neighborhood labeling algorithm. The statistics of the pixel values ​​in the eight-connected neighborhood can be expressed as: N8(P)={(x+i,y+j)|i,j∈{-1,0,1}}\(x,y) For any pixel in the image, its eight-connected neighborhood N8(P) is defined as the set of eight pixels including P(X,Y) and its surrounding pixels: The specific algorithm steps are as follows: 1) Region marking: traverse each pixel in the image and use a marking function f(P) to record the region to which each pixel belongs; 2) Isolated point detection: For each pixel point P(X,Y), check its eight-connected neighborhood N8(P). If P belongs to a specific category and there are no other pixels belonging to the same category in its eight-connected neighborhood, then P is considered an isolated point. 3) Isolated point removal: The pixel value of the detected isolated point P is set to the background value of the point or the average value of the neighboring pixels. Here, the isolated point is set to the background value. The mathematical expression is as follows: Let I(X,Y) represent the pixel value of the image at the coordinate (X,Y), and B(X,Y) represent the background pixel value. For each pixel point P(X,Y), the statistics of the pixel values ​​in its eight-connected neighborhood can be expressed as: If P is an isolated point, then S(P) will not contain pixel values ​​of the same category as P. In this case, the pixel value of P can be updated as:

9. A hyperspectral remote sensing image methane point source detection system based on superpixel matched filtering, used to implement the method according to any one of claims 1 to 8, characterized in that: include: A data acquisition module, used to obtain original hyperspectral remote sensing images and standard methane spectrum absorbance; A true color image and background spectrum acquisition module is used to obtain a true color image based on an original hyperspectral remote sensing image and to obtain a background spectrum based on the true color image; A target spectrum building module, used for building a methane target spectrum according to a standard methane spectrum absorbance and a background spectrum; A filtering module is used to filter the methane target spectrum and the original hyperspectral image using a matched filtering algorithm to obtain a methane point source filtering result; The result acquisition module is used to obtain the final methane point source location according to the methane point source filtering result.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images