A satellite observation sulfur dioxide vertical column total amount strip correction method, system and device

Abstract

The application discloses a satellite observation sulfur dioxide vertical column total amount strip correction method, system and equipment, relates to the satellite remote sensing data processing technical field, and the method comprises the following steps: based on any solar spectrum data in the solar spectrum data set of the target area in the preset time length collected by the satellite, determining SO2 column total amount original data-strip correction data pairing sample, and then forming a training sample set; a neural network model is constructed by fusing multi-scale feature extraction, spatial attention mechanism and adaptive high-value protection weight; the neural network model is trained by using the training sample set, and a sulfur dioxide vertical column total amount strip correction model is obtained; the SO2 column total amount original data corresponding to the new solar spectrum data is input into the sulfur dioxide vertical column total amount strip correction model, so that the corresponding SO2 column total amount strip correction data is obtained. The application can realize SO2 strip correction considering noise removal accuracy and signal fidelity.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of satellite remote sensing data processing, in particular to a satellite observation sulfur dioxide vertical column total amount strip correction method, system and equipment. BACKGROUND

[0002] As one of the main atmospheric pollutants, the global distribution and concentration change of sulfur dioxide has an important influence on atmospheric environmental quality, climate system and ecological environment. Satellite remote sensing technology has become the core means of global SO2 concentration monitoring due to its wide coverage and short observation period. Two-dimensional satellite spectrometers (such as the EMI of Gaofen-5 satellite and TROPOMI instrument of Sentinel-5) can obtain high spatial resolution SO2 vertical distribution observation data. However, due to the pixel performance difference of two-dimensional CCD detector, the interference of complex space environment (such as cosmic rays, temperature fluctuations) on the optical system of the instrument, there is obvious strip noise in the observed SO2 vertical column total amount data. The strip noise shows systematic deviation along the detector column direction, which seriously interferes with the accurate inversion of SO2 concentration, reduces the accuracy of distinguishing high concentration areas and background areas, and even affects the reliability of global SO2 source emission estimation.

[0003] The existing strip removal methods mainly include traditional algorithms based on statistical filtering (such as median filtering and wavelet denoising) and correction methods based on physical models. The traditional statistical filtering method is easy to cause smoothing distortion of the real SO2 concentration signal, especially in high concentration pollution areas, which will cause peak attenuation; the physical model correction method relies on accurate modeling of instrument response function and environmental parameters, but the instrument performance drift in complex space environment is difficult to model accurately, which limits the strip removal effect. Therefore, the existing methods cannot balance the noise removal accuracy and signal fidelity. SUMMARY

[0004] The purpose of the present application is to provide a satellite observation sulfur dioxide vertical column total amount strip correction method, system and equipment, which can realize SO2 strip correction with consideration of noise removal accuracy and signal fidelity.

[0005] To achieve the above purpose, the present application provides the following solutions.

[0006] In a first aspect, the present application provides a satellite observation sulfur dioxide vertical column total amount strip correction method, comprising:

[0007] acquiring a set of solar spectrum data of a target area collected by a satellite within a preset time length;

[0008] determine SO2 column total original data-band correction data paired samples based on any solar spectrum data in the solar spectrum data set; a plurality of SO2 column total original data-band correction data paired samples constitute a training sample set;

[0009] construct a neural network model that integrates multi-scale feature extraction, spatial attention mechanism and adaptive high-value protection weight;

[0010] train the neural network model using the training sample set to obtain a sulfur dioxide vertical column total band correction model;

[0011] input the SO2 column total original data corresponding to the new solar spectrum data into the sulfur dioxide vertical column total band correction model to obtain the corresponding SO2 column total band correction data.

[0012] In a second aspect, the present application provides a satellite observation sulfur dioxide vertical column total band correction system, comprising:

[0013] The data acquisition module is configured to acquire a solar spectrum data set of a target area within a preset time length collected by a satellite;

[0014] The sample generation module is configured to determine SO2 column total original data-band correction data paired samples based on any solar spectrum data in the solar spectrum data set; a plurality of SO2 column total original data-band correction data paired samples constitute a training sample set;

[0015] The model construction module is configured to construct a neural network model that integrates multi-scale feature extraction, spatial attention mechanism and adaptive high-value protection weight;

[0016] The model training module is configured to train the neural network model using the training sample set to obtain a sulfur dioxide vertical column total band correction model;

[0017] The band correction module is configured to input the SO2 column total original data corresponding to the new solar spectrum data into the sulfur dioxide vertical column total band correction model to obtain the corresponding SO2 column total band correction data.

[0018] In a third aspect, the present application provides a computer device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the satellite observation sulfur dioxide vertical column total band correction method.

[0019] According to the specific embodiments provided in the application, the application discloses the following technical effects: based on any one of the solar spectrum data in the solar spectrum data set of the target region collected by the satellite within the preset time length, the application determines the SO2 column total original data-band correction data pairing sample, and thus realizes data pairing. Then, a neural network model that fuses multi-scale feature extraction, spatial attention mechanism and adaptive high-value protection weight is constructed, and the neural network model is trained. Through the technical scheme of "data pairing training + neural network learning", compared with the traditional statistical filtering and physical model method, the strip removal precision is higher and the real signal of the high concentration region can be effectively retained, the peak value decay is avoided, and the neural network can be directly used for global SO2 source emission inventory compilation, atmospheric pollution cross-border transmission simulation and other researches. Moreover, the multi-scale feature extraction fused in the neural network is designed for the strip noise characteristics of the SO2 observation data, and the processing of fusing multi-scale feature extraction, spatial attention mechanism and adaptive high-value protection weight can realize the balance between accurate noise removal and signal fidelity.

[0020] The application effectively solves the strip noise problem caused by the performance difference of the instrument and the interference of the space environment, significantly improves the accuracy and data availability of the SO2 vertical column total inversion, and provides reliable data support for atmospheric pollution monitoring, climate change research and the like. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0022] Figure 1 It is a flowchart of the satellite observation sulfur dioxide vertical column total strip correction method in an embodiment of the application.

[0023] Figure 2 It is an example schematic diagram of the training data set.

[0024] Figure 3 It is a comparison schematic diagram of the strip correction of the traditional method and the strip correction of the deep learning method.

[0025] Figure 4 It is a structural schematic diagram of a computer device provided in an embodiment of the application. DETAILED DESCRIPTION

[0026] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of the present application.

[0027] In view of the problem that the vertical distribution of sulfur dioxide observed by the two-dimensional satellite spectrometer in the prior art has obvious bands and is difficult to remove, the present application provides a satellite observed SO2 vertical column amount band correction method using a neural network, which realizes accurate removal of band noise and efficient correction of SO2 vertical column concentration, while ensuring signal fidelity in high concentration areas, and finally outputs regular gridded high-quality SO2 column amount products.

[0028] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the present application will be described in further detail below with reference to the accompanying drawings and specific embodiments.

[0029] In one exemplary embodiment, as shown in Figure 1 A satellite observed sulfur dioxide vertical column amount band correction method is provided, which is executed by a computer device, specifically, can be executed by a terminal or a server, or can be executed by a terminal and a server together, and in the embodiments of the present application, the method includes the following steps 101 to 105.

[0030] Step 101, acquiring a set of solar spectrum data of a target area collected by a satellite within a preset time length. Since there are high concentration areas of SO2 in the global SO2 concentration distribution, the data acquisition process before the training set is constructed needs to cover clean SO2 areas and polluted SO2 areas, etc., to cover different SO2 concentration scenarios, so as to ensure the generality of the model trained in the subsequent steps for band correction of clean areas and polluted areas. Therefore, the target area at least includes clean SO2 areas and polluted SO2 areas.

[0031] In one specific application example, step 101 includes the following steps.

[0032] (1) Download a set of global solar spectrum data observed by a satellite within a preset time length.

[0033] (2) Determine the acquisition frequency of the training sample set; for example, set the data observed by the satellite for 12 months a year, 3 days a month (a total of 36 days), to ensure that the SO2 concentration distribution characteristics in different seasons are covered.

[0034] (3) Extracting a plurality of solar spectrum data from the global solar spectrum data set according to the collection frequency; the solar spectrum data includes two-dimensional spectrum data corresponding to a clean SO2 region and two-dimensional spectrum data corresponding to a polluted SO2 region in the same day.

[0035] The clean SO2 region can be an open sea area without volcanic activity, the region has no artificial SO2 emission, and the SO2 concentration during the non-volcanic activity period is close to 0. The polluted SO2 region can be selected from a high SO2 concentration region. Correspondingly, the solar spectrum data can include 1 clean region sample and 2 polluted region (volcanic activity area / industrial intensive area) samples in the same day, to ensure the balance of samples of the two types of scenes.

[0036] (4) The plurality of solar spectrum data constitutes a solar spectrum data set.

[0037] Step 102, based on any solar spectrum data in the solar spectrum data set, determining SO2 column total original data-band correction data paired samples; a plurality of SO2 column total original data-band correction data paired samples constitute a training sample set.

[0038] In one specific application example, step 102 includes: for the two-dimensional spectrum data corresponding to the clean SO2 region and the polluted SO2 region in the same day, the following processing is performed:

[0039] (21) Inversion is performed to determine the SO2 column total original matrix; specifically, based on the satellite observed two-dimensional spectrum data, an adaptive iteration technique is used, a plurality of parameters including SO2 column total are simultaneously iteratively fitted, combined with the atmospheric radiation transfer model and the atmospheric extinction law, to make the observed and simulated atmospheric albedo data reach the optimal fitting, and obtain the SO2 column total original matrix.

[0040] In actual application, this processing can also be completed after step (1) in step 101. That is, after determining the SO2 column total original matrix, data collection and sample data construction are performed.

[0041] (22) Based on a predetermined condition, dividing the SO2 column total original matrix into valid pixels and invalid pixels; wherein the predetermined condition is that the pixel with cloud cover less than 30%, SO2 column total greater than -0.5 DU (1 DU=2.69e16 mole / cm 2 ) and less than 0.5 DU is a valid pixel, and vice versa.

[0042] (23) Based on all the valid pixels, the SO2 band value is calculated by the formula In short, the median is calculated as the SO2 band value of the corresponding column.

[0043] (24) Using the formula SO2 striping correction was performed to obtain the corrected SO2 values; for invalid pixels, the corrected SO2 value was set to 0; where, For the first SO2 stripe values ​​of the column, For the first A list of all rows of cells in a column, where MEDIAN is the mean function; This represents the corrected SO2 value in column i.

[0044] (25) Extract multiple two-dimensional matrices from the two-dimensional spectral data by random cropping.

[0045] (26) The original SO2 column total amount matrix and the corrected SO2 value corresponding to any of the two-dimensional matrices constitute a paired sample of original SO2 column total amount data and strip correction data.

[0046] After the above correction process is performed on both clean SO2 areas and areas with SO2 pollution, 200 and 100 two-dimensional matrices of 450×450 pixels are randomly selected respectively (corresponding to the single observation field of the TROPOMI instrument) to form 10,800 pairs of "raw data-strip correction data" samples, achieving full coverage in both time and space dimensions.

[0047] To address the striping phenomenon caused by differences in 2D CCD performance and the space environment, the original distribution matrix of SO2 column totals was manually destriped to construct a paired training set of "original data - striped correction data". The matrix of the original SO2 column totals served as the input features for the neural network model, while the destriped data matrix after strip correction served as the model training labels.

[0048] Step 103: Construct a neural network model that integrates multi-scale feature extraction, spatial attention mechanism, and adaptive high-value protection weights. The neural network model includes a multi-scale feature extraction module, a feature fusion and scale recovery module, an attention and adaptive weights module, and a noise prediction and signal recovery module.

[0049] (i) The multi-scale feature extraction module is used to capture the strip noise and SO2 concentration signal features at different scales in the original data of total SO2 column.

[0050] Before inputting the raw SO2 column total data into the multi-scale feature extraction module, data preprocessing is generally required. Specifically, the input features and labels are subjected to min-max normalization, with a normalization range of [0,1], and the formula is as follows: . These are the original input features. for the normalized value, , respectively the minimum and maximum value in the sample set, ensuring the consistency of the input data distribution, improving the model convergence speed.

[0051] The multi-scale feature extraction module comprises a convolution reinforcement unit, an original scale branch, a 2-fold downsampling branch and a 4-fold downsampling branch.

[0052] The convolution reinforcement unit is configured to perform initial feature extraction and original scale feature reinforcement on the SO2 column total quantity original data to obtain a convolution-reinforced feature map; specifically, the input data (input features after normalization processing) is subjected to initial feature extraction through a 7x7 convolution layer, and then four 7x7 convolution blocks are connected in series to strengthen the spatial correlation capture of the original scale under-belt noise to obtain the convolution-reinforced feature map.

[0053] The original scale branch is configured to perform convolution processing on the convolution-reinforced feature map to obtain an original scale feature map; specifically, the convolution-reinforced feature map is output as the original scale feature map through a 3x3 convolution layer.

[0054] The 2-fold downsampling branch is configured to perform convolution and downsampling processing on the convolution-reinforced feature map to obtain a 2-fold downsampling feature map; specifically, the convolution-reinforced feature map is subjected to downsampling through a 3x3 stride 2 convolution to focus on the mesoscale under-belt noise distribution rule to obtain the 2-fold downsampling feature map.

[0055] The 4-fold downsampling branch is configured to perform twice convolution and downsampling processing on the convolution-reinforced feature map to obtain a 4-fold downsampling feature map; specifically, the convolution-reinforced feature map is subjected to downsampling through a 3x3 stride 2 convolution, and then subjected to downsampling again through a 3x3 stride 2 convolution to capture the global scale under-belt noise trend to obtain the 4-fold downsampling feature map.

[0056] The feature fusion and scale restoration module is configured to align and fuse the under-belt noise and SO2 concentration signal features of different scales in a manner combining deconvolution and interpolation to obtain a comprehensive feature map; specifically, the feature fusion and scale restoration module comprises a 4-fold upsampling branch, a 2-fold upsampling branch and a fusion unit.

[0057] The 4-fold upsampling branch is configured to perform deconvolution upsampling processing on the 4-fold downsampling feature map to obtain a 4-fold restored original scale feature map; specifically, the 4-fold restored original scale feature map is obtained through 4x4 stride 2 deconvolution upsampling and then 4x4 stride 2 deconvolution restoration to the original scale.

[0058] The 2x upsampling branch is configured to perform deconvolution upsampling processing on the 2x down-sampling feature map to obtain a 2x restored original scale feature map; specifically, 4x4 stride 2 deconvolution is used to directly restore to the original scale.

[0059] The fusion unit is configured to fuse the original scale feature map, the 2x restored original scale feature map, and the 4x restored original scale feature map by element addition, integrate multi-dimensional noise and signal features, and obtain a comprehensive feature map.

[0060] The attention and adaptive weight module is configured to dynamically learn the spatial attention weight distribution of the comprehensive feature map, assign a coefficient to each pixel position, and adjust the noise removal intensity of different SO2 concentration regions to obtain an attention-enhanced feature map. The attention and adaptive weight module includes a spatial attention mechanism unit and an adaptive weight prediction unit.

[0061] A dedicated spatial attention mechanism unit is designed to enhance the focusing ability of the model on the noise region in view of the specificity of the column distribution of SO2 strip noise. The spatial attention mechanism unit includes multiple convolution layers to compress the channel features of the comprehensive feature map to obtain a single-channel attention map, and normalize it to the [0, 1] interval through an activation function to obtain a spatial attention weight matrix. Specifically, the spatial attention mechanism unit is composed of 3 convolution layers, which compress the 64-channel feature to 32-channel, 16-channel, and finally output a 1-channel attention map, which is normalized to the [0, 1] interval through the Sigmoid activation function.

[0062] An adaptive weight prediction unit is designed to dynamically adjust the noise removal intensity of different concentration regions in view of the signal fidelity requirement of the high concentration region of SO2 (such as volcanic and industrial areas). The core is to dynamically adjust the noise removal intensity of different concentration regions. The adaptive weight prediction unit is configured to:

[0063] Based on the data statistical characteristics (such as standard deviation, mean, etc.) of the satellite observed SO2 column total in the training sample set, the SO2 column total matrix is divided into multiple regions, and a preset adaptive strategy is used to set the corresponding differentiated noise removal intensity coefficients for different regions; the adjustment coefficient matrix is obtained by multiplying the spatial attention weight matrix and the noise removal intensity coefficient, realizing the adaptive strategy of "weak denoising" in high value area and "strong denoising" in basic area, ensuring that the high concentration signal is not attenuated; the adjustment coefficient matrix and the comprehensive feature map are weighted and fused to obtain an attention-enhanced feature map.

[0064] Wherein, according to different SO2 concentration (X), the region is divided into extreme high value area, high value area, basic area, low value area and zero value area.

[0065] Extreme high value area: X ≥ mean + 8 times standard deviation (corresponding to strong emission scenarios such as volcanic eruption).

[0066] High value area: mean + 3 times standard deviation ≤ X < mean + 8 times standard deviation (corresponding to industrial intensive areas).

[0067] Basic area: mean - 3 times standard deviation ≤ X < mean + 3 times standard deviation (corresponding to background area).

[0068] Low value area: X < mean - 3 times standard deviation (corresponding to abnormally low noise).

[0069] Zero value area: X = 0 (corresponding to ideal clean area).

[0070] (Four) The noise prediction and signal recovery module is configured to predict a strip noise component based on the attention-enhanced feature map (e.g., by 2 layers of 3x3 convolution to predict a strip noise component), recover a clean signal using a residual learning mechanism, and obtain SO2 column total strip correction data.

[0071] In the noise prediction and signal recovery module, the following function formula is used: ; wherein, is the SO2 column total strip correction data, is the SO2 column total original data, is the strip noise component predicted by convolution, is an adjustment coefficient.

[0072] Compared with directly predicting a clean image, the residual learning focuses on the accurate modeling of noise components, reduces the difficulty of model training, dynamically adjusts the noise removal degree, and balances the strip removal effect and signal fidelity.

[0073] In this application, a strip noise correction network based on multi-scale feature fusion is used, which is specially designed for the strip noise characteristics of SO2 observation data. The core point is to fuse multi-scale feature extraction, spatial attention mechanism and adaptive high value protection weight, to realize the balance between accurate noise removal and signal fidelity.

[0074] Step 104, using the training sample set to train the neural network model to obtain a sulfur dioxide vertical column total strip correction model.

[0075] Based on the model constructed in step 103, model training is performed. If the validation set loss does not decrease (decrease amplitude <1e-5) for 15 consecutive rounds (which can be adjusted as needed), the early stopping mechanism is triggered, and the current optimal model is saved.

[0076] Step 105, input the SO2 column total original data corresponding to the new solar spectrum data into the sulfur dioxide vertical column total strip correction model to obtain the corresponding SO2 column total strip correction data.

[0077] Based on the optimal model trained, the SO2 column total original data observed by the satellite is strip corrected, and the specific steps are as follows: ensure that the input data is a two-dimensional matrix, call the saved 、 parameters in the training stage, and execute the min-max normalization consistent with the training set on the input original data. The data range after normalization is forced to be constrained in [0, 1], and the abnormal values exceeding the range are truncated according to the boundary value. The value less than is 0, and the value greater than is 1. Then, the preprocessed data is input into the sulfur dioxide vertical column total strip correction model, and the clean image after strip correction and the predicted noise component are output. Finally, the normalization result of the clean image is restored to the actual SO2 column total value, and the formula is as follows: . Wherein, is the strip-corrected secondary SO2 vertical column total data, is the strip-corrected normalized secondary SO2 vertical column total data.

[0078] In a specific application example, after step 105, the method further comprises: gridding the SO2 column total strip correction data to obtain a regular gridded tertiary SO2 column total product. Specifically, the region to be gridded is divided into 0.25°x0.25° (about 28 km x 28 km) grids, for each target grid, all secondary product pixels (i.e. SO2 column total strip correction data) within a radius of 18 km (which can be adjusted as needed) centered on the grid are selected, and the weighted average value thereof is calculated as the SO2 column total value after strip correction of the grid, and the weight is the reciprocal of the distance from the pixel to the grid, and finally a regular gridded tertiary SO2 column total product is obtained.

[0079] In summary, the present application proposes a technical scheme of "data pairing training + neural network learning" for the strip noise problem of the SO2 observation data of the two-dimensional satellite spectrometer, and sequentially performs the processing of the three levels of training sample set based on the inversion of SO2 column total of the two-dimensional satellite spectrometer, obtaining SO2 column total distribution, neural network model construction, model training, and subsequent SO2 prediction and grid acquisition of tertiary product. Compared with the traditional statistical filtering and physical model method, the strip removal precision is higher, the inversion precision is reduced to below 0.05 DU, and the real signal in the high concentration area can be effectively preserved to avoid peak attenuation, which can be directly used for global SO2 source emission inventory compilation, atmospheric pollution cross-border transmission simulation and other researches.

[0080] In one specific application, the present application also aims at strip correction of SO2 column total amount in industrial intensive areas in May of the target year, and an application example is given in which Band3 two-dimensional spectral data observed by TROPOMI are selected. The specific process of the application example is as follows:

[0081] (1) Download the Band3 two-dimensional spectral data observed by TROPOMI from December of the historical year to November of the target year, and inverse the clean areas and industrial intensive areas therein, and use the adaptive iteration technology to inverse the original matrix of SO2 column total amount, so as to obtain the monthly average SO2 distribution in May of the target year.

[0082] (2) Select the clean areas on 1-3 days of each month, and screen the pixels with cloud amount less than 30%, SO2 column total amount greater than -0.5 DU and less than 0.5 DU as effective pixels, and vice versa. Select the spectrum with the most SO2 effective pixels in the area, and perform SO2 strip correction. After the strip correction is completed, 100 two-dimensional matrices of 450x450 pixels are extracted in a random clipping manner.

[0083] Similarly, for the above selected 1-3 days, 3 obvious SO2 high concentration areas are selected each month, the strip value is calculated and manual strip correction is performed. After the strip correction is completed, 200 two-dimensional matrices of 450x450 pixels are also extracted in a random clipping manner.

[0084] (3) The original SO2 column total amount matrix is taken as the input feature of the neural network, and the manually corrected despeckled data matrix is taken as the model training label, and the obtained training data set is as shown in Figure 2 , which contains the comparison of clean areas, polluted areas and corresponding statistical method corrected data.

[0085] First, the data is preprocessed, and then the neural network is constructed, and the model training is performed based on the model architecture. When the validation set loss does not decrease for 15 consecutive rounds, it is the current optimal model. Based on the current optimal model trained, the original data of satellite observed SO2 column total amount in May of the target year is subjected to strip correction.

[0086] If there is already a trained model, the model can be directly called, the preprocessed data is input into the current optimal model, and the despeckled clean image and the predicted noise component are output. Finally, the normalized result of the clean image is restored to the actual SO2 column total amount value.

[0087] (4) Select part of the area to divide into 0.25°*0.25° grid, for each target grid, select all secondary product pixels within the range of 18km radius with the grid as the center, calculate the weighted average value as the SO2 column total value of the grid after strip correction, and the weight is the reciprocal of the distance from the pixel to the grid. Finally, the regular gridded tertiary SO2 column total product is obtained. As shown in Figure 3 The strip correction and debanding effect based on the neural network provided in the application is more obvious compared with the original result and the traditional statistical filtering method, and the real signal of the high concentration area can be effectively retained, and the peak value decay is avoided.

[0088] Based on the same inventive concept, the embodiments of the application also provide a system. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme described in the above method, so the specific limitations in one or more system embodiments provided below can refer to the limitations of the method in the above text, and will not be repeated here.

[0089] In an exemplary embodiment, a satellite observation sulfur dioxide vertical column total strip correction system is provided, comprising:

[0090] A data acquisition module is configured to acquire a set of solar spectrum data of a target area collected by a satellite within a preset time length.

[0091] A sample generation module is configured to determine SO2 column total original data-strip correction data paired samples based on any solar spectrum data in the set of solar spectrum data. A plurality of SO2 column total original data-strip correction data paired samples constitute a training sample set.

[0092] A model construction module is configured to construct a neural network model that integrates multi-scale feature extraction, spatial attention mechanism and adaptive high-value protection weight.

[0093] A model training module is configured to train the neural network model using the training sample set to obtain a sulfur dioxide vertical column total strip correction model.

[0094] A strip correction module is configured to input SO2 column total original data corresponding to new solar spectrum data into the sulfur dioxide vertical column total strip correction model to obtain corresponding SO2 column total strip correction data.

[0095] In summary, in view of the obvious banding and difficulty in removing the banding of the vertical column total amount distribution of sulfur dioxide (SO2) observed by a two-dimensional satellite spectrometer, the application realizes banding removal and accurate correction of the SO2 column total amount by constructing a raw data-banded data paired training set, designing a special neural network model, and finally obtaining a regular gridded three-level SO2 column total amount product through gridding processing. Compared with the traditional mathematical method, the application effectively solves the problem of banding noise caused by instrument performance difference and space environment interference, significantly improves the accuracy and data availability of SO2 vertical column total amount inversion, and provides reliable data support for atmospheric pollution monitoring, climate change research and the like.

[0096] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal, and an internal structure diagram thereof can be as shown in Figure 4 The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement the satellite observation sulfur dioxide vertical column total amount band correction method.

[0097] Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the application, and does not constitute a limitation on the computer device to which the scheme of the application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0098] In an exemplary embodiment, a computer device is provided, which includes a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps in the above method embodiments when executing the computer program.

[0099] In an exemplary embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by the processor to implement the steps in the above method embodiments.

[0100] In an example embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0101] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0102] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0103] The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on blockchain, etc., without being limited thereto. The processor involved in the embodiments provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0104] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, it should be understood that the application encompasses all possible combinations of the technical features unless such a combination is not technically possible.

[0105] The principles and implementation manners of the present application are described herein by using specific examples, and the above embodiments are only used to help understand the method of the present application and its core idea; meanwhile, according to the idea of the present application, the specific implementation manners and application scopes will be changed by those skilled in the art. In conclusion, the content of the present specification should not be understood as a limitation of the present application.

Claims

1. A method for correcting the vertical column total sulfur dioxide content stripe from satellite observations, characterized in that, The method includes: Acquire a set of solar spectral data of the target area collected by satellite within a preset time period; Based on any solar spectral data in the solar spectral data set, a paired sample of SO2 total column raw data and strip correction data is determined; multiple paired samples of SO2 total column raw data and strip correction data constitute a training sample set. A neural network model integrating multi-scale feature extraction, spatial attention mechanism, and adaptive high-value protection weights is constructed; the neural network model includes a multi-scale feature extraction module, a feature fusion and scale recovery module, an attention and adaptive weight module, and a noise prediction and signal recovery module; The multi-scale feature extraction module is used to capture the strip noise and SO2 concentration signal features at different scales in the original SO2 column total data; the feature fusion and scale restoration module is used to align and fuse the strip noise and SO2 concentration signal features at different scales using a combination of deconvolution and interpolation to obtain a comprehensive feature map; the attention and adaptive weight module is used to dynamically learn the spatial attention weight distribution of the comprehensive feature map and assign coefficients to each pixel position to adjust the noise removal intensity in different SO2 concentration regions, thereby obtaining an attention-enhanced feature map; the noise prediction and signal recovery module is used to predict the strip noise component through convolution based on the attention-enhanced feature map and recover the clean signal using a residual learning mechanism to obtain SO2 column total strip correction data; The multi-scale feature extraction module includes a convolutional enhancement unit, an original scale branch, a 2x downsampling branch, and a 4x downsampling branch. The convolutional enhancement unit is used to perform initial feature extraction and original scale feature enhancement on the original SO2 column total data to obtain a convolutionally enhanced feature map. The original scale branch is used to perform convolution processing on the convolutionally enhanced feature map to obtain an original scale feature map. The 2x downsampling branch is used to perform convolution and downsampling processing on the convolutionally enhanced feature map to obtain a 2x downsampling feature map. The 4x downsampling branch is used to perform two convolution and downsampling processes on the convolutionally enhanced feature map to obtain a 4x downsampling feature map. The feature fusion and scale restoration module includes a 4x upsampling branch, a 2x upsampling branch, and a fusion unit. The 4x upsampling branch performs deconvolution upsampling on the 4x downsampling feature map to obtain a 4x restored original scale feature map. The 2x upsampling branch performs deconvolution upsampling on the 2x downsampling feature map to obtain a 2x restored original scale feature map. The fusion unit fuses the original scale feature map, the 2x restored original scale feature map, and the 4x restored original scale feature map by element-wise addition to obtain a comprehensive feature map. The noise prediction and signal recovery module uses the following function formula: ;in, For SO2 column total amount strip correction data, This is the original data for the total SO2 column volume; For the strip noise components predicted by convolution, For adjustment coefficients; The neural network model is trained using the training sample set to obtain a vertical column total sulfur dioxide strip correction model. The original data of total SO2 column corresponding to the new solar spectral data is input into the sulfur dioxide vertical column total strip correction model to obtain the corresponding SO2 column total strip correction data.

2. The satellite observation vertical column total sulfur dioxide strip correction method according to claim 1, characterized in that, The target area includes clean SO2 areas and areas containing SO2 pollution. Acquire a set of solar spectral data of the target area collected by satellite within a preset time period, including: Download the set of global solar spectral data observed by satellite within a preset time period; Determine the collection frequency of the training sample set; Based on the acquisition frequency, multiple solar spectral data are extracted from the global solar spectral dataset; the solar spectral data includes two-dimensional spectral data corresponding to clean SO2 areas and two-dimensional spectral data corresponding to areas with SO2 pollution on the same day; The solar spectral data from multiple sources constitute a solar spectral data set.

3. The satellite observation vertical column total sulfur dioxide strip correction method according to claim 2, characterized in that, Based on any solar spectral data from the aforementioned solar spectral data set, determine a paired sample of raw SO2 column total data and strip correction data, including: For the two-dimensional spectral data corresponding to both clean SO2 areas and areas with SO2 contamination on the same day, the following processing was performed: Inversion is performed to determine the original matrix of total SO2 column volume; Based on preset conditions, the original matrix of the total amount of SO2 columns is divided into effective pixels and invalid pixels; Based on all the aforementioned valid pixels, the formula is used. Calculate the SO2 stripe value; Using formula Perform SO2 stripe correction to obtain the corrected SO2 value; Multiple two-dimensional matrices are extracted from the two-dimensional spectral data using a random pruning method; The original SO2 column total amount matrix and the corrected SO2 value corresponding to any of the two-dimensional matrices constitute a paired sample of original SO2 column total amount data and strip correction data. in, For the first SO2 stripe values ​​of the column, For the first A list of all rows of cells in a column, where MEDIAN is the mean function; This represents the corrected SO2 value in column i.

4. The satellite observation vertical column total sulfur dioxide strip correction method according to claim 1, characterized in that, The attention and adaptive weight module includes a spatial attention mechanism unit and an adaptive weight prediction unit; The spatial attention mechanism unit includes multiple convolutional layers to compress the channel features of the comprehensive feature map to obtain a single-channel attention map, and normalizes it to the [0,1] interval through an activation function to obtain the spatial attention weight matrix; The adaptive weight prediction unit is used to: divide the total SO2 column matrix into multiple regions based on the statistical characteristics of the total SO2 column data observed by satellite, and set corresponding noise removal intensity coefficients for different regions using a preset adaptive strategy; obtain an adjustment coefficient matrix by multiplying the spatial attention weight matrix with the noise removal intensity coefficients; and perform weighted fusion of the adjustment coefficient matrix with the comprehensive feature map to obtain the attention-enhanced feature map. The different regions include extremely high value regions, high value regions, basic value regions, low value regions, and zero value regions.

5. The satellite observation vertical column total sulfur dioxide strip correction method according to claim 1, characterized in that, The method further includes: The SO2 column total amount strip correction data is processed by gridding to obtain a regular gridded three-level SO2 column total amount product.

6. A satellite observation vertical column total sulfur dioxide stripe correction system, employing the satellite observation vertical column total sulfur dioxide stripe correction method according to any one of claims 1-5, characterized in that, The system includes: The data acquisition module is used to acquire a set of solar spectral data of the target area collected by the satellite within a preset time period; The sample generation module is used to determine paired samples of SO2 total column raw data and strip correction data based on any solar spectral data in the solar spectral data set; multiple paired samples of SO2 total column raw data and strip correction data constitute a training sample set. The model building module is used to build a neural network model that integrates multi-scale feature extraction, spatial attention mechanism and adaptive high-value protection weights; The model training module is used to train the neural network model using the training sample set to obtain a vertical column total sulfur dioxide strip correction model. The strip correction module is used to input the original data of SO2 column total amount corresponding to the new solar spectral data into the sulfur dioxide vertical column total amount strip correction model to obtain the corresponding SO2 column total amount strip correction data.

7. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the satellite observation method for vertical column total sulfur dioxide strip correction as described in any one of claims 1-5.

Images