A method, system, device and medium for automatically identifying a point source emission of greenhouse gas

Abstract

This invention discloses an automatic identification method, system, device, and medium for greenhouse gas point source emissions, relating to the field of environmental monitoring technology. The method includes: acquiring a greenhouse gas emission inventory and satellite remote sensing data of the area to be monitored; performing initial screening of emission sources based on the greenhouse gas emission inventory to determine preliminary greenhouse gas sources; retrieving a greenhouse gas concentration distribution map using the satellite remote sensing data, and employing a Mask R-CNN model to perform target detection and pixel segmentation on the greenhouse gas concentration distribution map to determine greenhouse gas plumes; the greenhouse gas plumes include specific locations and shapes; performing quantitative analysis of the greenhouse gas plumes using the IME algorithm, and combining this with the preliminary screening of greenhouse gas sources to determine the final emission sources. This invention can improve the accuracy and real-time performance of greenhouse gas emission source identification.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to an automatic identification method, system, equipment and medium for greenhouse gas point source emissions. Background Technology

[0002] Greenhouse gases, also known as greenhouse effect gases, are the gaseous components in the atmosphere that contribute to the greenhouse effect. The greenhouse effect helps maintain the surface temperature of a planet, preventing excessive heat from escaping through radiation. However, an excessive greenhouse effect can also lead to rising planetary temperatures and global climate anomalies. Generally, greenhouse gases mainly include the following six: carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O), hydrofluorocarbons (HFCs), perfluorocarbons (PFCs), and sulfur hexafluoride (SF6). With increasing global industrialization, the concentration of greenhouse gases on the Earth's surface has gradually increased due to human activities. Among them, carbon dioxide (CO2) and methane (CH4), the two greenhouse gases that contribute the most to the greenhouse effect, have seen particularly rapid increases in recent years.

[0003] Currently, greenhouse gas emission monitoring mainly follows two technical approaches: bottom-up inventory methods and top-down measurement methods.

[0004] Bottom-up inventory approaches rely on statistical data and mathematical models to estimate emission activities from emission sources. For example, based on global emission databases such as Carbon Mapper and Climate Trace, emission fluxes can be calculated using methods such as first-order decay models by statistically analyzing emission activities (e.g., energy use, landfill volume) and emission factors. This approach is convenient and comprehensive when data is complete. However, its accuracy heavily depends on the quality of the statistical data and the reliability of the model parameters. For instance, key parameters such as methane production rates need to be obtained through field experiments, which introduces significant uncertainty. Furthermore, inventory approaches cannot reflect emission dynamics in real time and have limited applicability to regions with scarce data.

[0005] Top-down measurement methods retrieve emissions by directly detecting the concentration of greenhouse gases in the atmosphere, and mainly include two technical approaches: ground-based monitoring and spaceborne remote sensing.

[0006] Ground-based monitoring: This method uses handheld devices or fixed monitoring stations to monitor specific emission sources (such as power plants and landfills). It is characterized by high accuracy and reliability, and can output timely greenhouse gas concentration data. However, its application is limited by the number of monitoring stations and terrain conditions, making it difficult to cover large-scale or concealed emission sources.

[0007] Spaceborne remote sensing: Satellite-based remote sensing technology detects the absorption characteristics of greenhouse gases in the atmospheric reflectance spectrum and inverts their concentration distribution. Spaceborne remote sensing has advantages such as wide coverage and short revisit cycles, making it an important means of greenhouse gas monitoring. For example, hyperspectral satellites can quickly provide regional emission data. However, emission plumes in spectral identification images are complex, making it difficult to identify emission signals from small emission sources or complex areas.

[0008] Hyperspectral remote sensing technology, due to its high spectral density and high spatial resolution, can identify greenhouse gas emission hotspots in small or complex scenes. However, the following challenges remain in the identification of emission plumes: 1. Strong background interference: Hyperspectral data often contains interference signals from high-reflectivity ground features (such as building roofs and solar panels), affecting the accuracy of gas concentration inversion. 2. Complex plume morphology: Emission plumes are usually affected by topography, airflow, and other factors, resulting in complex and irregular morphologies, making it difficult for traditional image processing methods to effectively extract their boundaries and concentration distribution. 3. Large data processing volume: Hyperspectral data has many bands and high spatial resolution, placing high demands on computational resources and algorithm optimization. Although hyperspectral matched filtering methods can extract greenhouse gas concentration information, their adaptability to complex background conditions is weak, especially in identifying irregular plume morphologies and quantifying small emission sources. Summary of the Invention

[0009] The purpose of this invention is to provide an automatic identification method, system, device, and medium for greenhouse gas point source emissions, which can improve the accuracy and real-time performance of greenhouse gas emission source identification.

[0010] To achieve the above objectives, the present invention provides the following solution:

[0011] An automatic identification method for greenhouse gas point source emissions includes:

[0012] Obtain greenhouse gas emission inventories and satellite remote sensing data of the areas to be monitored;

[0013] The emission sources are initially screened based on the greenhouse gas emission inventory to determine the preliminary greenhouse gas sources.

[0014] Greenhouse gas concentration distribution maps are retrieved using satellite remote sensing data, and the Mask R-CNN model is used to perform target detection and pixel segmentation on the greenhouse gas concentration distribution maps to determine greenhouse gas plumes; the greenhouse gas plumes include specific locations and shapes.

[0015] The greenhouse gas plume was quantitatively analyzed using the IME algorithm, and the final emission source was determined in conjunction with the initially screened greenhouse gas sources.

[0016] Optionally, the process of initial screening of emission sources based on the greenhouse gas emission inventory specifically includes:

[0017] The data in the greenhouse gas emission inventory are judged according to the screening criteria. Greenhouse gas sources with large emissions and obvious emission characteristics that meet the screening criteria are identified as greenhouse gas sources in the preliminary screening. The greenhouse gas emission inventory adopts at least the CarbonMapper and Climate Trace global public greenhouse gas emission inventories.

[0018] Optionally, the calculation formula for retrieving the greenhouse gas concentration distribution map using satellite remote sensing data includes:

[0019]

[0020] Where, α i x represents the methane concentration increment, in ppm·m. a Σ is the average radiance value of the entire image, Σ is the covariance matrix of the selected calculation band of the image, k is the unit absorption spectrum of methane, and x is the inverted concentration value.

[0021] Optionally, before performing target detection and pixel segmentation on the greenhouse gas concentration distribution map using the Mask R-CNN model, the method further includes: training the Mask R-CNN model;

[0022] The process of training the Mask R-CNN model specifically includes:

[0023] Acquire training data; the training data includes historical gas concentration distribution maps and corresponding plume labels;

[0024] A pre-trained network is constructed based on a convolutional neural network and a region proposal network; the convolutional neural network is used to extract spatial features from remote sensing images; the region proposal network is used to generate regions of interest.

[0025] The training data is input into the pre-trained network, and training is performed with the goal of minimizing the loss between the network output and the plume label. The network that meets the accuracy evaluation is determined as the final Mask R-CNN model.

[0026] Optionally, the calculation formula for quantitative analysis of the greenhouse gas plume using the IME algorithm includes:

[0027]

[0028] Where: Q is the emission flux, ΔC is the concentration increment, A is the area of ​​the plume region, U is the wind speed, and L is the plume length.

[0029] The present invention also provides an automatic identification system for greenhouse gas point source emissions, comprising:

[0030] The data acquisition unit is used to acquire greenhouse gas emission inventories and satellite remote sensing data of the area to be measured;

[0031] The initial screening unit is used to conduct initial screening of emission sources based on the greenhouse gas emission inventory to determine the preliminary greenhouse gas sources.

[0032] The plume identification unit is used to retrieve greenhouse gas concentration distribution maps from satellite remote sensing data, and to perform target detection and pixel segmentation on the greenhouse gas concentration distribution maps using the Mask R-CNN model to determine greenhouse gas plumes; the greenhouse gas plumes include specific locations and shapes.

[0033] The quantitative analysis unit is used to perform quantitative analysis of the greenhouse gas plume using the IME algorithm, and to determine the final emission source in conjunction with the initially screened greenhouse gas sources.

[0034] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the automatic identification method for greenhouse gas point source emissions described above.

[0035] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the automatic identification method for greenhouse gas point source emissions as described above.

[0036] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0037] This invention discloses an automatic identification method, system, device, and medium for greenhouse gas point source emissions. The method includes acquiring a greenhouse gas emission inventory and satellite remote sensing data of the area to be monitored; performing initial screening of emission sources based on the greenhouse gas emission inventory to determine preliminary greenhouse gas sources; retrieving a greenhouse gas concentration distribution map using the satellite remote sensing data, and using a Mask R-CNN model to perform target detection and pixel segmentation on the greenhouse gas concentration distribution map to determine greenhouse gas plumes; the greenhouse gas plumes include specific locations and shapes; performing quantitative analysis of the greenhouse gas plumes using the IME algorithm, and combining this with the preliminary screening of greenhouse gas sources to determine the final emission sources. This invention can improve the accuracy and real-time performance of greenhouse gas emission source identification. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart illustrating the automatic identification method for greenhouse gas point source emissions according to the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] The purpose of this invention is to provide an automatic identification method, system, device, and medium for greenhouse gas point source emissions, which can improve the accuracy and real-time performance of greenhouse gas emission source identification.

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, this invention provides an automatic identification method for greenhouse gas point source emissions. The aim is to address the challenges of wide distribution, high identification difficulty, and complex emission quantification of greenhouse gas emission sources by combining global greenhouse gas emission inventories, spaceborne hyperspectral imager data, and deep learning technology. By comprehensively utilizing statistical inventories and hyperspectral remote sensing data, this invention provides an automatic identification and quantitative calculation method for greenhouse gas point source emissions based on hyperspectral imagery and deep learning, providing technical support for accurate greenhouse gas monitoring and emission reduction policy formulation. The main steps include:

[0044] Step 100: Obtain a global greenhouse gas emissions inventory through public platforms (such as Carbon Mapper, Climate Trace, etc.) and collect hyperspectral data covering the target area from satellite remote sensing data sources (such as PRISMA, GF-5, etc.). The obtained remote sensing data includes hyperspectral imagery covering the absorption bands of methane (1600–2500 nm) and carbon dioxide (2000–2200 nm), providing data support for subsequent analysis and target detection. This step not only ensures data diversity but also enables the acquisition of key information from multiple global emission sources.

[0045] Step 200: Conduct a preliminary screening of emission sources using the greenhouse gas emission inventory to identify key emission sources for analysis. Focus on point sources with high methane and carbon dioxide emission intensity, such as landfills, coal mines, chemical plants, and oil fields. These sources may exhibit significant plume characteristics and emission intensity exceeding predetermined thresholds. The screening criteria for emission sources include emission intensity, the significance of emission characteristics, and the completeness and temporal coverage of the data.

[0046] Step 300: Use satellite remote sensing data to retrieve the greenhouse gas concentration distribution map, and use the Mask R-CNN model to perform target detection and pixel segmentation on the greenhouse gas concentration distribution map to determine the greenhouse gas plume; the greenhouse gas plume includes specific location and shape.

[0047] Step 400: Quantitatively analyze the greenhouse gas plume using the IME algorithm, and determine the final emission source in conjunction with the initially screened greenhouse gas sources.

[0048] Based on the above technical solution, the following embodiments are provided.

[0049] First aspect: Methods for identifying key emission sources and constructing training datasets based on global greenhouse gas emission inventories.

[0050] Step 1.1 Obtain global greenhouse gas emissions inventory data

[0051] Emission data for facilities or regions are obtained from publicly available global greenhouse gas emission inventories such as Carbon Mapper and Climate Trace. This data provides key information such as the geographical location of emission sources, emission amounts, and emission characteristics, which is used to screen key targets for subsequent hyperspectral identification.

[0052] Step 1.2 Screening key emission sources

[0053] The emission inventory data obtained in step 1.1 was screened, and point sources with high methane and carbon dioxide emission intensity, including facilities such as landfills, coal mines, chemical plants and oil fields, were selected as the analysis targets.

[0054] The screening criteria include: whether the emission intensity exceeds the set threshold; whether the emission characteristics are clear and obvious (e.g., significant plume characteristics); and whether the data integrity and coverage meet the requirements for subsequent analysis.

[0055] Step 1.3 Constructing the training dataset

[0056] Methane and carbon dioxide plume data related to the screening targets were collected from the Carbon Mapper platform and used as training data for the deep learning model.

[0057] Data cleaning: Remove missing values, outliers, and noisy data to ensure data quality.

[0058] Data annotation: The spatial distribution and concentration values ​​of the plume are annotated to provide high-quality input for model training.

[0059] Data augmentation: Enhance data through methods such as rotation, scaling, and noise addition to improve the model's adaptability in different scenarios.

[0060] Preferably, step 1.3 includes:

[0061] Step 1.3.1 Data Grouping

[0062] The plume data are classified and grouped according to the type of emission source (such as landfill, coal mine, oil field) to facilitate targeted optimization during model training.

[0063] The second aspect: a method for generating greenhouse gas concentration increment maps based on hyperspectral imager data.

[0064] Step 2.1 Acquire hyperspectral data and perform preprocessing

[0065] 1. Hyperspectral data acquisition

[0066] Use a spaceborne hyperspectral imager (such as PRISMA, GF-5, or EMIT) to acquire hyperspectral data covering the target area. Select image data in the methane absorption band (e.g., 1600–2500 nm) and the carbon dioxide absorption band (e.g., 2000–2200 nm).

[0067] 2. Radiation calibration

[0068] The digital values ​​(DN) of hyperspectral images are converted to radiance using the following formula:

[0069] L = Gain·DN + Bias

[0070] Where LLL is the radiance, with units of W / (m²). 2 (·sr·μm), where Gain and Bias are calibration coefficients for the hyperspectral instrument.

[0071] 3. Terrain Correction

[0072] For mountainous areas and complex terrain, the effect of solar incidence angle on radiance is corrected using the following formula:

[0073]

[0074] Among them, L H For the corrected brightness of a flat surface, L T For the actual brightness under complex terrain, cosθs and cosγ s These are the cosine values ​​of the solar incidence angle and the observation angle, respectively.

[0075] 4. Removal of interfering data

[0076] Cloud masking technology is used to remove areas covered by clouds; high reflectivity features (such as building roofs and solar panels) are identified and their values ​​are replaced with background brightness.

[0077] Step 2.2 Matched Filter Inversion Concentration Increment

[0078] 1. Generate a unit absorption spectrum

[0079] Using a radiative transfer model combined with the HITRAN molecular absorption library, the unit absorption spectrum of greenhouse gases (such as methane) is simulated for concentration increment calculations.

[0080] 2. Concentration increment inversion

[0081] The greenhouse gas concentration increment is retrieved using a matched filter algorithm, with the following formula:

[0082]

[0083] Where, α i x represents the methane concentration increment (in ppm·m). a Σ is the average radiance value of the entire image, Σ is the covariance matrix of the selected calculation band of the image, k is the unit absorption spectrum of methane, and x is the inverted concentration value.

[0084] Step 2.3 Optimize the incremental map and generate results

[0085] 1. Noise Removal

[0086] The inversion results may contain isolated noise points or false detection areas, which do not represent actual greenhouse gas emissions. Therefore, image processing techniques, such as morphological operations and filtering denoising, are needed to remove noise and ensure the accuracy of the results.

[0087] 2. Incremental map generation

[0088] The denoised concentration increment data is further processed to generate a high-resolution greenhouse gas concentration increment map. This map accurately reflects the changes in greenhouse gas concentrations in each region, providing basic data for subsequent automatic plume identification and quantitative emission calculation.

[0089] The third aspect: Deep learning-based methods for automatic plume identification and emission quantification.

[0090] Step 3.1 Model Training

[0091] Data preparation

[0092] Using the hyperspectral concentration increment map generated in the second aspect mentioned above, and the plume distribution data extracted from Carbon Mapper, a training set for the Mask R-CNN model is constructed.

[0093] Data normalization: Normalize the concentration increment data to ensure the consistency of the model input features.

[0094] Data annotation: Accurately label the greenhouse gas plume regions in the training data, including their spatial location and morphology. This annotation information will be used for supervised learning, helping the model learn how to identify plumes. Model training.

[0095] The Mask R-CNN deep learning model was used for training.

[0096] Using convolutional neural networks (CNNs), spatial features in hyperspectral images are extracted by processing image data through convolutional and pooling layers.

[0097] Regions of interest (RoIs), which are areas where greenhouse gas plumes may exist, are generated through the Region Proposal Network (RPN).

[0098] Within each RoI, classification and regression prediction are performed to output the specific location and morphology of the plume.

[0099] Model Validation

[0100] Model validation was performed on independent test sets to evaluate its accuracy and robustness in automatic plume identification. Precision, recall, and F1 score were used to evaluate the model's performance in different scenarios, ensuring its universality and efficiency.

[0101] Step 3.2 Model Application

[0102] 1. Plumage Recognition

[0103] Using a trained Mask R-CNN model, hyperspectral concentration increment maps are analyzed to automatically identify the location and shape of greenhouse gas plumes. The model output generates plume distribution maps, displaying the concentration distribution of greenhouse gases in various regions.

[0104] 2. Data Post-processing

[0105] Post-processing of the identification results: Isolated noise points are removed, and continuous plume regions are preserved. The plume boundaries are smoothed to improve the stability of the identification results.

[0106] Step 3.3 Quantifying Emissions

[0107] Flux calculation

[0108] Using plume distribution maps and wind speed information, combined with the IME algorithm, greenhouse gas emission fluxes are calculated. The calculation formula is as follows:

[0109]

[0110] Where: Q: emission flux (unit: kg / h); ΔC: concentration increment (unit: ppm); A: plume area (unit: m2); U: wind speed (unit: m / s); L: plume length (unit: m).

[0111] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0112] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for automatic identification of greenhouse gas point source emissions, characterized in that, include: Obtain greenhouse gas emission inventories and satellite remote sensing data of the areas to be monitored; The emission sources are initially screened based on the greenhouse gas emission inventory to determine the preliminary greenhouse gas sources. Greenhouse gas concentration distribution maps are retrieved using satellite remote sensing data, and the Mask R-CNN model is used to perform target detection and pixel segmentation on the greenhouse gas concentration distribution maps to determine greenhouse gas plumes; the greenhouse gas plumes include specific locations and shapes. The greenhouse gas plume was quantitatively analyzed using the IME algorithm, and the final emission source was determined in conjunction with the initially screened greenhouse gas sources. The calculation formula for retrieving greenhouse gas concentration distribution maps using satellite remote sensing data. include: ， in, This represents the increase in methane concentration, expressed in ppm·m. The average radiance value of the entire image. Calculate the covariance matrix of the selected bands for the image. This is the unit absorption spectrum of methane. The concentration value is the result of the inversion. Before performing target detection and pixel segmentation on the greenhouse gas concentration distribution map using the Mask R-CNN model, the method further includes: training the Mask R-CNN model; The process of training the Mask R-CNN model specifically includes: Acquire training data; the training data includes historical gas concentration distribution maps and corresponding plume labels; A pre-trained network is constructed based on a convolutional neural network and a region proposal network; the convolutional neural network is used to extract spatial features from remote sensing images; the region proposal network is used to generate regions of interest. The training data is input into the pre-trained network, and training is performed with the goal of minimizing the loss between the network output and the plume label. The network that meets the accuracy evaluation is determined as the final Mask R-CNN model. The calculation formula for quantitative analysis of the greenhouse gas plume using the IME algorithm includes: ， in: For emission flux, For the concentration increment, The area of ​​the plume region. Where L is the wind speed and L is the plume length.

2. The automatic identification method for greenhouse gas point source emissions according to claim 1, characterized in that, The process of initial screening of emission sources based on the greenhouse gas emission inventory specifically includes: The data in the greenhouse gas emission inventory are judged according to the screening criteria. Greenhouse gas sources with large emissions and obvious emission characteristics that meet the screening criteria are identified as greenhouse gas sources in the preliminary screening. The greenhouse gas emission inventory adopts at least the Carbon Mapper and Climate Trace global public greenhouse gas emission inventories.

3. An automatic greenhouse gas point source emission identification system, employing the automatic greenhouse gas point source emission identification method as described in any one of claims 1-2, characterized in that, include: The data acquisition unit is used to acquire greenhouse gas emission inventories and satellite remote sensing data of the area to be measured; The initial screening unit is used to conduct initial screening of emission sources based on the greenhouse gas emission inventory to determine the preliminary greenhouse gas sources. The plume identification unit is used to retrieve greenhouse gas concentration distribution maps from satellite remote sensing data, and to perform target detection and pixel segmentation on the greenhouse gas concentration distribution maps using the Mask R-CNN model to determine greenhouse gas plumes; the greenhouse gas plumes include specific locations and shapes. The quantitative analysis unit is used to perform quantitative analysis of the greenhouse gas plume using the IME algorithm, and to determine the final emission source in conjunction with the initially screened greenhouse gas sources.

4. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the automatic identification method for greenhouse gas point source emissions according to any one of claims 1-2.

5. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the automatic identification method for greenhouse gas point source emissions as described in any one of claims 1-2.

Images