Surface reflectance inversion method and device for optical satellite remote sensing data

Abstract

The embodiment of the invention provides a surface reflectance inversion method and device for optical satellite remote sensing data, and the method comprises the steps: obtaining sensitive parameter data corresponding to pixels, the pixels are all pixels in to-be-inverted optical satellite remote sensing data, and the optical satellite remote sensing data are collected by an optical satellite sensor, the sensitive parameter data comprises observation geometric data, atmosphere auxiliary data and elevation data; inputting the sensitive parameter data into an atmospheric correction neural network model to obtain an atmospheric correction coefficient of the pixel, the atmospheric correction neural network model being obtained by pre-training based on an atmospheric correction coefficient lookup table of the optical satellite sensor; and calculating according to the atmospheric correction coefficient and the radiation brightness of each pixel to obtain the surface reflectance of the optical satellite remote sensing data. According to the method, the inversion of the surface reflectance is carried out pixel by pixel based on the neural network model, and the method has the characteristics of high universality, high inversion precision and high calculation speed.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing data processing, and in particular to a method and device for inverting surface reflectivity of optical satellite remote sensing data. Background Art

[0002] Surface reflectivity is a key input parameter for the inversion of many surface geophysical remote sensing parameters (such as leaf area index, chlorophyll, biomass, etc.), and is also the basis for long-term surface change monitoring. However, optical satellite remote sensing data is affected by atmospheric scattering and absorption during imaging, resulting in a large difference between the actual data values ​​obtained by the optical satellite and the true observation values, which affects the application of optical satellite remote sensing data in quantitative inversion and time series change monitoring. Therefore, it is necessary to perform atmospheric correction on optical satellite remote sensing data to eliminate the radiation distortion caused by atmospheric influence and obtain the true surface reflectivity.

[0003] At present, the mainstream surface reflectance inversion methods generally use 6S (Second Simulation of the Satellite Signal in the Solar Spectrum) and MODTRAN (Moderate Resolution Transmission) radiation transfer models to simulate the transmission process of solar radiation in the ground-atmosphere system, thereby eliminating the influence of the atmosphere on remote sensing data. For example, the Landsat5 / 7 atmospheric correction processing model and the Landsat8 atmospheric correction model released by the United States Geological Bureau are based on the 6S radiation transfer model to invert the surface reflectance of Landsat data; the Sentinel-2 surface reflectance inversion module Sen2Cor, ACTOR surface reflectance inversion software, and FLAASH atmospheric correction module in ENVI released by the European Space Agency are developed based on the MODTRAN radiation transfer model. The key to using these two radiation transfer models for atmospheric correction is the acquisition and input of sensitive factors such as geometric parameters, atmospheric parameters (aerosol, water vapor content, ozone, etc.), and altitude parameters.

[0004] In terms of parameter input: ① In terms of geometric parameter input, the usual practice is to use only one fixed value for one scene image. ② In terms of atmospheric parameter input, parameters such as water vapor content and ozone use only the mean or scene center value as input for one scene image; and because aerosol data has a greater impact on the inversion of surface reflectivity, the usual practice is to set the aerosol type to continental aerosol, and at the same time build a lookup table for aerosol optical thickness to consider the change in aerosol optical thickness; ③ In terms of altitude parameter input, it is also usually the case that only the mean value is used as input for one scene image. This processing method of using only one fixed value as input in one scene image during parameter input will result in poor accuracy of the inverted surface reflectivity, which is suitable for satellite remote sensing data with narrow widths such as Landsat and Sentinel-2, but not for satellite remote sensing data with larger widths. This is because for satellite remote sensing data with larger widths, the angle change within one scene is as high as 30° or more, and the large geographical range often leads to large changes in atmospheric conditions and terrain, making the inverted surface reflectivity low in accuracy. However, if the lookup table is constructed by taking into account factors such as geometric parameters, aerosols, water vapor content, ozone, and changes in altitude, the computational time complexity will increase exponentially, the lookup time will be very long, and the efficiency of surface reflectivity inversion from optical satellite remote sensing data will be very low, which will pose great difficulties in practical applications. Summary of the invention

[0005] In view of this, an embodiment of the present disclosure provides at least one method and device for inverting the surface reflectivity of optical satellite remote sensing data.

[0006] Specifically, the embodiments of the present disclosure are implemented through the following technical solutions:

[0007] In a first aspect, a method for inverting surface reflectivity of optical satellite remote sensing data is provided, characterized in that the method comprises:

[0008] Acquire sensitive parameter data corresponding to a pixel, wherein the pixel is each pixel in the optical satellite remote sensing data to be inverted, wherein the optical satellite remote sensing data is acquired by an optical satellite sensor, and the sensitive parameter data includes observation geometry data, atmospheric auxiliary data, and elevation data;

[0009] Inputting the sensitive parameter data into an atmospheric correction neural network model to obtain an atmospheric correction coefficient of the pixel, wherein the atmospheric correction neural network model is pre-trained based on an atmospheric correction coefficient lookup table of the optical satellite sensor;

[0010] The surface reflectivity of the optical satellite remote sensing data is obtained by calculating according to the atmospheric correction coefficient and the radiation brightness of each pixel.

[0011] In combination with any embodiment provided in the present disclosure, the atmospheric correction neural network model is obtained by the following training method:

[0012] Acquire an atmospheric correction coefficient lookup table of the optical satellite sensor, wherein the atmospheric correction coefficient lookup table is constructed according to a mapping relationship between sample sensitive parameter data and sample atmospheric correction coefficients;

[0013] Inputting the sample sensitive parameter data in the atmospheric correction coefficient lookup table into the initial neural network to obtain the initial atmospheric correction coefficient output by the initial neural network;

[0014] determining a loss function based on a difference between the initial atmospheric correction coefficient and the sample atmospheric correction coefficient corresponding to the sample sensitive parameter data;

[0015] The initial neural network is trained according to the loss function until the loss function satisfies a preset stop training condition, thereby obtaining the atmospheric correction neural network model.

[0016] In combination with any embodiment provided in the present disclosure, the step of obtaining the atmospheric correction coefficient lookup table of the optical satellite sensor includes:

[0017] Acquiring the spectral response function of the optical satellite sensor and the sample sensitive parameter data under different sensitive parameter conditions, wherein the parameter conditions include: observation geometry conditions, atmospheric conditions, and elevation conditions;

[0018] For each spectral band in the spectral response function, a sample atmosphere correction coefficient corresponding to the sample sensitive parameter data is calculated according to the sample sensitive parameter data and a radiation transfer model.

[0019] In combination with any of the embodiments provided in the present disclosure, the atmospheric auxiliary data includes: aerosol optical thickness data;

[0020] When obtaining the aerosol optical thickness data corresponding to the pixel, it includes:

[0021] Obtain aerosol optical depth data from multiple data sources;

[0022] The aerosol optical thickness data from the multiple data sources are spatially fused to obtain the aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data.

[0023] In combination with any of the embodiments provided in the present disclosure, the aerosol optical thickness data from the multiple data sources are spatially fused to obtain the aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data, including:

[0024] Determine primary data and secondary data, wherein the primary data is aerosol optical thickness data from a data source with the highest resolution among multiple data sources, and the secondary data is aerosol optical thickness data from data sources other than the primary data;

[0025] Obtaining, according to the secondary data, an estimated value of aerosol optical thickness data of an area not covered by the primary data in the optical satellite remote sensing data;

[0026] The aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data are determined according to the main data and the estimated value.

[0027] In combination with any of the embodiments provided in the present disclosure, the observation geometry data includes: a solar zenith angle and an observation zenith angle;

[0028] The atmospheric auxiliary data include: aerosol type, aerosol optical thickness, atmospheric water vapor content, and ozone content;

[0029] The elevation data is digital elevation model data.

[0030] In combination with any implementation manner provided in the present disclosure, the step of obtaining sensitive parameter data corresponding to a pixel includes:

[0031] Based on the optical satellite remote sensing data, obtaining the observed geometric data corresponding to the pixel;

[0032] Processing the collected initial atmospheric auxiliary data to obtain atmospheric auxiliary data matching the optical satellite remote sensing data;

[0033] Based on the atmospheric auxiliary data matched with the optical satellite remote sensing data, obtaining the atmospheric auxiliary data of each pixel in the optical satellite remote sensing data;

[0034] Processing the collected initial elevation data to obtain elevation data matching the optical satellite remote sensing data;

[0035] Based on the elevation data matched with the optical satellite remote sensing data, the elevation data of each pixel in the optical satellite remote sensing data is obtained.

[0036] In a second aspect, a device for inverting surface reflectivity of optical satellite remote sensing data is provided, the device comprising:

[0037] A data acquisition module is used to acquire sensitive parameter data corresponding to a pixel, wherein the pixel is each pixel in the optical satellite remote sensing data to be inverted, wherein the optical satellite remote sensing data is acquired by an optical satellite sensor, and the sensitive parameter data includes observation geometry data, atmospheric auxiliary data, and elevation data;

[0038] An atmospheric correction module, used for inputting the sensitive parameter data into an atmospheric correction neural network model to obtain an atmospheric correction coefficient of the pixel, wherein the atmospheric correction neural network model is pre-trained based on an atmospheric correction coefficient lookup table of the optical satellite sensor;

[0039] The reflectivity calculation module is used to calculate the surface reflectivity of the optical satellite remote sensing data according to the atmospheric correction coefficient and the radiation brightness of each pixel.

[0040] According to a third aspect, an electronic device is provided, comprising a memory and a processor, wherein the memory is used to store computer instructions executable on the processor, and the processor is used to implement the surface reflectivity inversion method for optical satellite remote sensing data as described in any embodiment of the present disclosure when executing the computer instructions.

[0041] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method for inverting the surface reflectivity of optical satellite remote sensing data as described in any embodiment of the present disclosure is implemented.

[0042] The surface reflectivity inversion method for optical satellite remote sensing data provided by the embodiment of the present disclosure fully considers the changes in the observed geometric data, atmospheric auxiliary data, and elevation data corresponding to each pixel in the optical satellite remote sensing data based on a deep learning method, and calculates the surface reflectivity through the atmospheric correction coefficient of each pixel generated by the atmospheric correction neural network model, wherein the parameter input of a scene image uses the sensitive parameter data of each pixel in the scene, thereby solving the problems of low surface reflectivity inversion accuracy and poor universality caused by fixed value parameter input. The method is not only applicable to optical satellite remote sensing data with a narrow width, but also to the surface reflectivity inversion of optical satellite remote sensing data with a large width. The method uses a neural network model to generate the atmospheric correction coefficient, without the need to use a large-scale lookup table, thereby solving the problem of long search time of the traditional lookup table. The method has the characteristics of strong universality, high inversion accuracy, and fast calculation speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in one or more embodiments of the present disclosure or related technologies, the drawings required for use in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in one or more embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 is a flow chart of a method for inverting surface reflectance of optical satellite remote sensing data shown in at least one embodiment of the present disclosure;

[0045] Figure 2 is a flow chart of a training method of an atmospheric correction neural network model shown in at least one embodiment of the present disclosure;

[0046] Figure 3 It is a flowchart of another method for inverting the surface reflectance of optical satellite remote sensing data according to at least one embodiment of the present disclosure;

[0047] Figure 4 is a block diagram of a surface reflectivity inversion system for optical satellite remote sensing data shown in at least one embodiment of the present disclosure;

[0048] Figure 5 is a block diagram of a device for inverting surface reflectivity of optical satellite remote sensing data shown in at least one embodiment of the present disclosure;

[0049] Figure 6 It is a schematic diagram of the hardware structure of an electronic device shown in at least one embodiment of the present disclosure. DETAILED DESCRIPTION

[0050] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this specification. Instead, they are merely examples of devices and methods consistent with some aspects of this specification as detailed in the appended claims.

[0051] The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit this specification. The singular forms "a", "the" and "the" used in this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0052] It should be understood that although the terms first, second, third, etc. may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this specification, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0053] To facilitate understanding, relevant terms involved in the embodiments of the present disclosure are first introduced below.

[0054] Surface reflectivity: refers to the ratio of the amount of radiation reflected by the ground to the amount of incident radiation, which represents the ground's ability to absorb and reflect solar radiation. The greater the reflectivity, the less solar radiation the ground absorbs; the smaller the reflectivity, the more solar radiation the ground absorbs. When calculating the surface reflectivity of a large area, optical satellite remote sensing data is usually used.

[0055] Optical satellite remote sensing data: Optical satellite remote sensing data is collected by optical satellites. Optical satellite remote sensing data is generated by optical satellite sensors receiving electromagnetic waves from the earth's surface and is composed of multiple pixels.

[0056] Pixel: Pixel is the individual pixel in the optical satellite remote sensing data to be inverted. The pixel of the optical satellite remote sensing data is the smallest unit for the sensor to scan and sample the ground scene when the sensor scans and images.

[0057] Each pixel represents the reflection or radiation information of a specific location on the surface, and the value of each pixel represents the reflection or radiation intensity of the location in a specific band (such as visible light, infrared light, etc.). These values ​​are usually received by satellite sensors and quantized into values ​​(such as grayscale values ​​between 0 and 255) for storage.

[0058] That is, a pixel is a data element that has both spatial and spectral characteristics. The spatial characteristic is the ground area corresponding to the pixel, and the spectral characteristic is that its spectral variable represents the intensity of the spectral response of the pixel in a specific band.

[0059] Atmospheric correction: refers to the conversion of radiation brightness into actual surface reflectivity to eliminate errors caused by atmospheric scattering, absorption, and reflection.

[0060] The purpose of atmospheric correction is to eliminate the influence of factors such as atmosphere and light on the reflection of ground objects and obtain the true surface reflectivity, including eliminating the influence of water vapor, oxygen, carbon dioxide, methane and ozone in the atmosphere on the reflection of ground objects, eliminating the influence of atmospheric molecules and aerosol scattering, etc.

[0061] Atmospheric correction usually uses radiation transfer models to simulate and eliminate the effects of atmospheric scattering and absorption on optical satellite remote sensing data, thereby restoring the true reflectivity of the surface.

[0062] For example, radiation transfer models such as 6S and MODTRAN are used to simulate the transmission process of solar radiation in the earth-atmosphere system, and atmospheric correction is performed on the radiation brightness received by the sensor to eliminate the influence of the atmosphere when inverting the surface reflectivity.

[0063] Radiant brightness: also known as the radiant brightness of the top atmosphere, refers to the radiant brightness received by the satellite's sensor. It is obtained by calibrating the optical satellite remote sensing data. It represents the sum of the radiation energy of a certain area and is a value affected by factors such as atmospheric absorption, scattering, and reflection.

[0064] Radiation transfer model: The input of the radiation transfer model is geometric parameters, atmospheric parameters (aerosols, water vapor content, ozone, etc.), and altitude parameters. The radiation transfer model simulates the transmission process of solar radiation in the earth-atmosphere system based on the above input parameters to calculate various physical parameters between the satellite and the ground.

[0065] The radiation transfer model can calculate the radiation brightness at the satellite entrance pupil from bottom to top, and can also calculate the surface reflectivity of ground objects from top to bottom taking into account various influencing factors.

[0066] The present disclosure does not limit the radiation transfer model used, which may be a 6S radiation transfer model, a MODTRAN radiation transfer model, or an ATCOR (A Spatially-Adaptive Fast Atmospheric Correction) radiation transfer model, etc.

[0067] Sensitive parameter data: parameter data related to atmospheric influences such as atmospheric scattering and absorption on sunlight during the transmission process from the sun to the ground to the sensor, including observation geometry data, atmospheric auxiliary data, and elevation data.

[0068] The observed geometric data may be data of geometric parameters (zenith angle and azimuth angle of the satellite and the sun, etc.). The atmospheric auxiliary data may be data of parameters such as the atmospheric mode, aerosol mode, and optical thickness of the area where the satellite is located when acquiring the image. The elevation data may be the height of the ground target and the height of the sensor itself, which is not limited in this embodiment.

[0069] Atmospheric correction coefficient: a parameter used to correct the atmospheric influence in remote sensing data when inverting the surface reflectance. It can be obtained by inputting geometric parameters, atmospheric parameters (aerosol, water vapor content, ozone, etc.) and altitude parameters related to atmospheric influence (i.e., sensitive parameter data defined in this disclosure) into the radiation transfer model for simulation and calculation.

[0070] The following takes the 6S radiation transfer model as an example to illustrate the process of obtaining the atmospheric correction coefficient based on the radiation transfer model and calculating the surface reflectivity based on the atmospheric correction coefficient and radiation brightness.

[0071] Assuming that the surface is a uniform Lambertian body and the atmospheric level is uniform, the apparent reflectance received by the satellite sensor can be expressed as:

[0072]

[0073] Among them, ρ s is the surface reflectivity; θ s ,θ v are the solar zenith angle and the observation zenith angle; are the solar azimuth and the observation azimuth; ρ a is the path radiation; T g (θ s ,θ v ) is the atmospheric absorption rate; ρ R+A is the reflectivity of Rayleigh and aerosol; T(θ s ) is the atmospheric transmittance from the sun to the ground; T(θ v ) is the atmospheric transmittance from the ground to the sensor; S is the atmospheric lower hemisphere reflectivity.

[0074] ρ a , T g (θ s ,θ v ),ρ R+A 、T(θ s )、T(θ v ) and S can be simulated by the 6S radiation transfer model.

[0075] The apparent reflectivity It can also be expressed based on the top atmosphere radiation brightness as:

[0076]

[0077] Where Le is the radiance of the top layer of the atmosphere, d is the relative distance between the sun and the earth, which is approximately 1, and ESUN is the solar irradiance in the outer layer of the atmosphere.

[0078] Substituting formula (2) into formula (1), the expression of surface reflectivity can be obtained after conversion:

[0079]

[0080] The above x a 、x b 、x c This is the atmospheric correction coefficient used in this disclosure. In this example, it can be used to convert the top atmosphere radiation brightness to obtain the true surface reflectivity. ρ is obtained by simulating the 6S radiation transfer model. a , T g (θ s ,θ v ),ρ R+A 、T(θ s )、T(θ v), S is substituted into the above formula (4) to obtain the atmospheric correction coefficient x a 、x b 、x c .

[0081] After obtaining the atmospheric correction coefficient, the radiation brightness value is substituted into formula (3) to obtain the surface reflectivity.

[0082] It is understandable that the radiance and the apparent reflectance can be converted into each other. In other examples, the atmospheric correction coefficient corresponding to the apparent reflectance can also be constructed based on formula (1) to calculate the actual surface reflectance using the atmospheric correction coefficient and the apparent reflectance, which will not be elaborated here.

[0083] Current technology usually adopts the traditional method of constructing a lookup table to obtain the atmospheric correction coefficient. For example, when using Landsat 8 data to generate surface reflectance, a pixel-by-pixel calculation method based on a lookup table is adopted. For the Landsat-8OLI sensor, the five parameters of atmospheric correction influencing factors, such as solar zenith angle, water vapor, ozone, aerosol, and elevation, are used to input the 6S radiation transfer model to establish a five-dimensional atmospheric correction coefficient lookup table. The atmospheric correction coefficient of the point is calculated pixel by pixel based on the lookup table interpolation, and finally the surface reflectance image is obtained according to the atmospheric correction coefficient.

[0084] However, this traditional method of constructing a lookup table cannot cover the atmospheric correction coefficients under all atmospheric conditions. It is necessary to obtain the atmospheric correction coefficients that do not exist in the lookup table through interpolation calculation, resulting in low efficiency and low accuracy in the inversion of surface reflectivity of optical satellite remote sensing data.

[0085] Based on this, at least one embodiment of the present disclosure provides a method for inverting the surface reflectivity of optical satellite remote sensing data. Instead of using a traditional lookup table, the atmospheric correction coefficient is generated through an atmospheric correction neural network model. The deep learning-based method uses a large amount of sample data to learn the relationship between the input parameters and the atmospheric correction coefficient to obtain the atmospheric correction neural network model. Therefore, the accurate atmospheric correction coefficient can be obtained pixel by pixel by inputting the parameters of the atmospheric correction neural network model, thereby quickly and accurately realizing the surface reflectivity inversion of optical satellite remote sensing data.

[0086] like Figure 1 As shown, Figure 1 The present invention is a flowchart of a method for inverting the surface reflectance of optical satellite remote sensing data according to at least one embodiment of the present invention. The method may include the following steps:

[0087] In step 102, sensitive parameter data corresponding to the pixel is obtained.

[0088] In this step, sensitive parameter data corresponding to each pixel is obtained based on the pixel, and the sensitive parameter data corresponding to the pixel is the sensitive parameter data of the surface area corresponding to the pixel.

[0089] Among them, sensitive parameter data include observation geometry data, atmospheric auxiliary data, and elevation data.

[0090] In one embodiment, the observed geometric data include: solar zenith angle and observation zenith angle; the atmospheric auxiliary data include: aerosol type, aerosol optical thickness, atmospheric water vapor content, ozone content; and the elevation data is digital elevation model data.

[0091] According to the sensitivity analysis of the input parameters of the atmospheric correction process by the inventors, it is determined that the above parameters have a great influence on the inversion accuracy of the surface reflectivity.

[0092] This embodiment does not limit the method of obtaining sensitive parameter data corresponding to pixels. It can be to first obtain sensitive parameter data corresponding to optical satellite remote sensing data, and then process to obtain sensitive parameter data corresponding to multiple pixels respectively, or to calculate the sensitive parameter data corresponding to the pixel one by one.

[0093] In step 104, the sensitive parameter data is input into the atmospheric correction neural network model to obtain the atmospheric correction coefficient of the pixel.

[0094] The atmospheric correction neural network model is pre-trained based on the atmospheric correction coefficient lookup table of the optical satellite sensor.

[0095] The atmospheric correction neural network model proposed in the present disclosure is a pre-trained neural network model based on deep learning construction, whose input is the sensitive parameter data of the pixel and whose output is the atmospheric correction coefficient of the same pixel.

[0096] The atmospheric correction neural network model corresponding to the optical satellite sensor can be a model specially trained for the sensor of the optical satellite, or a universal model can be trained to be applicable to sensors of different types of optical satellites.

[0097] The present disclosure does not limit the specific training process of the atmospheric correction neural network model.

[0098] The present disclosure does not limit the specific network structure of the atmospheric correction neural network model used. For example, an MLP (Multilayer Perceptron) neural network model can be used, an LSTM (Long Short-Term Memory) neural network model can be used, and a fully convolutional neural network model can also be used.

[0099] Exemplarily, the dimension of the input layer of the atmospheric correction neural network model may be the number of parameters of the sensitive parameter data, and the dimension of the output layer may be the number of atmospheric correction coefficients.

[0100] In this step, the sensitive parameter data of each pixel in the optical satellite remote sensing data may be sequentially input into the atmospheric correction neural network model to obtain the atmospheric correction coefficient of the pixel.

[0101] In step 106, the surface reflectance of the optical satellite remote sensing data is obtained by calculation based on the atmospheric correction coefficient and the radiation brightness of each pixel.

[0102] As mentioned above, based on the transmission process of solar radiation in the earth-atmosphere system simulated by the radiation transfer model, the functional relationship between radiation brightness and surface reflectivity can be determined, for example, the above formula (3).

[0103] This step can be to perform radiation calibration on each pixel, calculate the top atmospheric radiance of each pixel in the image, and convert the atmospheric correction coefficient x of each pixel into a 、x b 、x c Substituting into the above formula (3) to obtain the surface reflectance of the pixel, thereby obtaining the surface reflectance of each pixel in the optical satellite remote sensing data, and further obtaining the surface reflectance of the optical satellite remote sensing data.

[0104] It should be noted that in the above steps, since the optical satellite remote sensing data collected by the sensor has multiple spectral bands, for each spectral band corresponding to each pixel, the atmospheric correction coefficient corresponding to each spectral band can be obtained by inputting the sensitive parameter data into the atmospheric correction neural network model corresponding to the optical satellite sensor, and then the radiation brightness of the spectral band of the pixel is corrected by the atmospheric correction coefficient corresponding to each spectral band to obtain the surface reflectivity, that is, the pixels of the optical satellite remote sensing data have corresponding surface reflectivity in different spectral bands.

[0105] The atmospheric correction neural network model used here is pre-trained based on the spectral response function of the sensor.

[0106] The surface reflectivity inversion method of optical satellite remote sensing data provided by the embodiment of the present disclosure fully considers the changes in the observed geometric data, atmospheric auxiliary data, and elevation data corresponding to each pixel in the optical satellite remote sensing data based on the deep learning method, and calculates the surface reflectivity through the atmospheric correction coefficient of each pixel generated by the atmospheric correction neural network model, wherein the parameter input of a scene image uses the sensitive parameter data of each pixel in the scene, which has higher and more reasonable accuracy, and solves the problems of low surface reflectivity inversion accuracy and poor universality caused by fixed value parameter input. The atmospheric correction neural network model is used to quickly and accurately generate the atmospheric correction coefficient of each pixel, and solves the problems of long traditional lookup table lookup time and inaccurate interpolation calculation. The method is not only suitable for optical satellite remote sensing data with narrow width, but also for surface reflectivity inversion of optical satellite remote sensing data with large width, and there is no need to use a large-scale lookup table, and has the characteristics of strong universality, high inversion accuracy and fast calculation speed.

[0107] The following is an explanation of the training process of the atmospheric correction neural network model used in the above embodiment.

[0108] In one embodiment, if Figure 2 As shown, the atmospheric correction neural network model is obtained by the following training method:

[0109] In step 202, an atmospheric correction coefficient lookup table of the optical satellite sensor is obtained.

[0110] The atmospheric correction coefficient lookup table will be used as a training sample for the atmospheric correction neural network model, so that the trained atmospheric correction neural network model can learn the relationship between sensitive parameter data and atmospheric correction coefficients.

[0111] The atmospheric correction coefficient lookup table is constructed based on the mapping relationship between sample sensitive parameter data and sample atmospheric correction coefficients.

[0112] Exemplarily, the process of constructing the atmospheric correction coefficient lookup table is as follows:

[0113] The sample sensitive parameter data is input into the radiation transfer model, and the spectral response function of the sensor of the optical satellite can also be input. Through the radiation transfer model simulation, various physical parameters between the sun, satellite and ground can be obtained, such as ρ mentioned above. a , T g (θ s ,θ v ),ρ R+A 、T(θ s )、T(θ v ), S.

[0114] Then substitute the above physical parameters into formula (4) to calculate the corresponding sample atmosphere correction coefficient. There is a mapping relationship between the sample sensitive parameter data and the sample atmosphere correction coefficient calculated therefrom.

[0115] By performing the above operations on a large number of sample sensitive parameter data, the atmospheric correction coefficient lookup table is finally obtained.

[0116] This embodiment does not limit the method for obtaining the atmospheric correction coefficient lookup table of the optical satellite.

[0117] In order to obtain a rich training set, improve the training effect, and increase the accuracy and efficiency of atmospheric correction, in one example, obtaining the atmospheric correction coefficient lookup table of the optical satellite sensor includes:

[0118] The spectral response function of the optical satellite sensor and the sample sensitive parameter data under different sensitive parameter conditions are obtained. The parameter conditions include: observation geometry conditions, atmospheric conditions, and altitude conditions.

[0119] The spectral response function describes the sensitivity of a sensor to light of different wavelengths, that is, the degree to which the optical satellite's sensor responds to each band when receiving energy reflected or radiated from the surface.

[0120] When constructing an atmospheric correction coefficient lookup table for a certain optical satellite, it is necessary to obtain the spectral response function of the optical satellite sensor. When the radiation transfer model already has the spectral response function of the sensor, it is not necessary to obtain it additionally.

[0121] Taking the 6S radiation transfer model as an example, the spectral response function is a parameter that must be input into the 6S radiation transfer model. The spectral range of each band of the optical satellite sensor to be inverted is generally in the range of 0.25-4um. According to the spectral range of each band of the optical satellite sensor to be inverted, the response function within the specified spectral range of the band can be input, for example, with a step size of 2.5nm.

[0122] In this step, the parameter values ​​under different observation geometry conditions, atmospheric conditions, and elevation conditions can be arranged and combined, and the 6S radiation transfer model can be used to simulate and calculate the atmospheric correction coefficients corresponding to the sensors of the optical satellite under different sensitive parameter conditions, and a lookup table of the atmospheric correction coefficients of the sensors of the optical satellite to be inverted can be constructed.

[0123] In order to balance the accuracy and efficiency of building the lookup table, it is necessary to determine the dimension of the lookup table and the step size of each dimension.

[0124] According to the sensitivity analysis of the atmospheric correction input parameters, the dimensions of the lookup table (i.e., sensitive parameters) are determined as follows: observation geometry data: solar zenith angle and satellite zenith angle; atmospheric auxiliary data: WV (Water Vapor, atmospheric water vapor content), TO (Total Ozone, ozone content), aerosol type and AOD (Aerosol Optical Depth, aerosol optical depth); elevation data: DEM (Digital Elevation Model, digital elevation model data, referred to as elevation).

[0125] After setting the value range and step size of each sample sensitive parameter, the value of each sample sensitive parameter can be determined. For example, the value is as follows:

[0126] The values ​​of solar zenith angle (°) are: 0, 10, 20, 30, 40, 50, 60, 70, 75; the values ​​of satellite zenith angle (°) are: 0, 10, 20, 30, 40, 50, 60, 70, 75; the values ​​of atmospheric water vapor content (g / cm 2 ) is 0, 0.5, 1, 2, 3, 5; the ozone content (cm·atm) is 0, 0.2, 0.4, 0.5, 1; the aerosol types include strongly absorbing, moderately absorbing, weakly absorbing, and dust; the aerosol optical depth is 0.01, 0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.5, 0.6, 0.8, 1, 1.2, 1.4, 1.6, 1.8, 2; the altitude (km) is 0.0, 0.5, 1, 1.5, 2, 3, 4, 6, 8.

[0127] For each spectral band in the spectral response function, the sample sensitive parameter data is input into the radiation transfer model for simulation and calculation to obtain the sample atmospheric correction coefficient corresponding to the sample sensitive parameter data. For example, according to the values ​​of the above-mentioned sensitive parameters, for each spectral band of the sensor of the optical satellite to be inverted, it is necessary to perform 9×9×6×5×4×16×9 simulations of the 6S radiation transfer model to obtain the sample atmospheric correction coefficient mapped thereto.

[0128] The above 6S radiation transfer model is constructed with solar zenith angle, satellite zenith angle, WV, TO, AOD and DEM as independent variables, and the atmospheric correction coefficient x a 、x b 、x c It is a lookup table of the dependent variable, in which there is a mapping relationship between the independent variable and the dependent variable.

[0129] In other examples, after inputting the spectral response functions of the optical satellite sensors of each band to be inverted, other value ranges and step sizes of the sensitive parameters can also be set. When traversing different solar zenith angles, satellite zenith angles, atmospheric water vapor content, ozone content, aerosol optical thickness and elevation values, the 6S radiation transfer model is used to construct the atmospheric correction coefficient x with the solar zenith angle, satellite zenith angle, WV, TO, AOD and DEM as independent variables. a 、x b 、x c A multi-dimensional lookup table of the optical satellite sensor to be inverted for the dependent variable.

[0130] Taking the Gaofen-1 satellite (GF-1) as an example, the GF-1 sensor has four bands. When constructing the lookup table, it is constructed for these four bands respectively. The input to the radiation transfer model is the spectral response function of each band of GF-1. When the atmospheric correction neural model is applied after the training is completed, only the atmospheric auxiliary parameters, geometric parameters, elevation and other sensitive parameter data need to be input (the spectral response function does not need to be input here), and the atmospheric correction coefficients of these four bands of each pixel can be obtained respectively, and then the surface reflectivity of each pixel is calculated with the radiation calibration result of each band of the remote sensing image to be inverted.

[0131] In step 204, the sample sensitive parameter data in the atmospheric correction coefficient lookup table is input into the initial neural network to obtain the initial atmospheric correction coefficient output by the initial neural network.

[0132] The initial neural network is a neural network to be trained based on deep learning.

[0133] The obtained atmospheric correction coefficient lookup table is used as a sample data set, wherein the sample sensitive parameter data in the atmospheric correction coefficient lookup table is the input of the network, and the sample atmospheric correction coefficients corresponding to the sample sensitive parameter data in the atmospheric correction coefficient lookup table are used as training labels.

[0134] Exemplarily, 80% of the sample data may be used as a training set to input into the initial neural network for training and learning, and 20% of the sample data may be used as a test set for verification to obtain the atmospheric correction neural network model corresponding to the optical satellite sensor to be inverted.

[0135] In step 206, a loss function is determined based on the difference between the initial atmospheric correction coefficient and the sample atmospheric correction coefficient.

[0136] In step 208, the initial neural network is trained according to the loss function until the loss function satisfies a preset stop training condition, thereby obtaining an atmospheric correction neural network model corresponding to the optical satellite sensor.

[0137] Exemplarily, the network model of the above-mentioned initial neural network can use an MLP neural network model, which is an artificial neural network with nonlinear function fitting and adaptive learning capabilities, and can use the error back propagation algorithm to autonomously learn the nonlinear relationship between dependent variables and independent variables in a data-driven manner.

[0138] For example, according to the 7 sensitive parameter independent variables and 3 atmospheric correction coefficient dependent variables of the lookup table, the number of neurons in the input layer of the MLP network can be set to 7, the number of neurons in the output layer can be set to 3, the number of neurons in the hidden layer can be set to 15, and a 3-layer network structure can be adopted. The hidden layer can use the softmax activation function, the loss function can use MSE (Mean Square Error), and the gradient optimization algorithm can use the Adam algorithm.

[0139] In a specific implementation, the network parameters in the encoder can be adjusted by back propagation. When the network iteration end condition is reached, the network training is terminated. This embodiment does not limit the preset stop training condition. The condition may be that the iteration reaches a certain number of times, or the loss value is less than a certain threshold.

[0140] In the disclosed embodiment, an atmospheric correction neural network model based on deep learning is used to obtain the atmospheric correction coefficient, which has higher accuracy than the traditional linear interpolation method for searching the atmospheric correction coefficient. Because it is difficult for the traditional lookup table to traverse all parameters under atmospheric conditions, when the corresponding atmospheric correction coefficient does not exist in the lookup table, the atmospheric correction coefficient can only be calculated by linear interpolation, and a huge lookup table needs to be constructed. This solves the shortcomings of long search time in the traditional lookup table and unstable accuracy of atmospheric correction results of optical satellite remote sensing data, and improves the efficiency and accuracy of surface reflectivity inversion of optical satellite remote sensing data.

[0141] like Figure 3 As shown, Figure 3 1 is a flowchart of another method for inverting the surface reflectance of optical satellite remote sensing data according to at least one embodiment of the present disclosure. The method describes in more detail the process of acquiring sensitive parameter data of each pixel, wherein the atmospheric correction neural network model used may be an MLP model trained by the training method of the above embodiment. The method may include the following steps:

[0142] In step 302, based on the optical satellite remote sensing data, the observed geometric data corresponding to the pixel is obtained.

[0143] This embodiment does not limit the method of obtaining the observation geometry data of each pixel in the optical satellite remote sensing data.

[0144] Exemplarily, the observation geometry data may refer to data such as the solar zenith angle, the solar azimuth angle, the observation zenith angle, and the observation azimuth angle.

[0145] Some satellite data provide data file information such as solar zenith angle, solar azimuth angle, observation zenith angle, observation azimuth angle, etc., such as Landsat data and GF-6WFV data. The solar zenith angle, solar azimuth angle, observation zenith angle, and observation azimuth angle values ​​of each pixel in the optical satellite remote sensing data to be inverted can be obtained based on the angle information provided.

[0146] For satellite data that do not provide angle file information, the solar zenith angle, solar azimuth angle, observation zenith angle and observation azimuth angle of the image center point and four corner points can be calculated according to the angle calculation formula, and spatial interpolation can be performed to obtain the solar zenith angle, solar azimuth angle, observation zenith angle and observation azimuth angle values ​​of each pixel in the optical satellite remote sensing data to be inverted.

[0147] In step 304, the collected initial atmospheric auxiliary data is processed to obtain atmospheric auxiliary data matching the optical satellite remote sensing data; based on the atmospheric auxiliary data matching the optical satellite remote sensing data, the atmospheric auxiliary data of each pixel in the optical satellite remote sensing data is obtained.

[0148] This embodiment does not limit the method of obtaining the atmospheric auxiliary data of each pixel in the optical satellite remote sensing data.

[0149] Exemplarily, atmospheric auxiliary data refers to data on atmospheric water vapor content, ozone content, aerosol type and aerosol optical thickness. The collected atmospheric auxiliary data usually cannot be used directly and needs to be processed, such as cropping, reprojection, resampling, etc., so that the obtained atmospheric auxiliary data and the optical satellite remote sensing data to be inverted have consistent geographical range, resolution and projection, so that the pixels in the optical satellite remote sensing data correspond one-to-one with the data at the corresponding position of the pixel in the atmospheric auxiliary data, and then the atmospheric water vapor content, ozone content, and aerosol optical thickness data values ​​of each pixel in the optical satellite remote sensing data to be inverted are obtained.

[0150] The atmospheric water vapor content data can be NCEP (NOAA National Centers for Environmental prediction) reanalysis data or MODIS atmospheric water vapor products. The ozone content data can be TOMS (Total Ozone Mapping-Spectrometer) data or MODIS ozone content products.

[0151] It should be noted that the key to using the radiation transfer model for atmospheric correction lies in the accurate acquisition of atmospheric parameters. The usual practice is to use relevant atmospheric products, such as TOMS data, NCEP reanalysis data and MODIS atmospheric products, or to use relevant inversion algorithms to invert various atmospheric parameters.

[0152] The inventors analyzed the sensitive factors that affect the atmospheric correction and found that aerosols have a great influence on the accuracy of atmospheric correction. In the inversion of aerosol optical thickness, specific bands are often required. For example, the dark target method requires a short-wave infrared band, while domestic high-resolution satellites, SPOT series and other satellites usually lack short-wave infrared bands. Using MODIS aerosol optical thickness products as the main source of aerosol optical thickness parameters is an important way to invert the surface reflectivity of most optical satellite remote sensing data. However, MODIS aerosol optical thickness products often have a large range of invalid values. The usual practice is to use interpolation algorithms, such as spline function interpolation, Kriging interpolation, etc. to fill in. Although this method can solve the problem of invalid values ​​to a certain extent, interpolation in areas with a large number of invalid values ​​may cause data distortion, making the inversion accuracy unstable and uncertain.

[0153] The inventors further discovered that multiple sets of MODIS AOD products with different resolutions have been shared and released internationally, including the 3km and 10km MOD04 series, MOD08 series, MOD09CMA and 1km MCD19A2. If these multi-source AOD products can be spatially fused, the situation of large-scale invalid values ​​can be reduced.

[0154] Based on this, when the atmospheric auxiliary data includes aerosol optical depth data, the following methods can be used to obtain the aerosol optical depth data corresponding to the pixel:

[0155] Obtain aerosol optical depth data from multiple data sources;

[0156] The aerosol optical thickness data from the multiple data sources are spatially fused to obtain the aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data.

[0157] Among them, the aerosol optical depth data from multiple data sources can be from different aerosol optical depth products, different satellite sensor observation data, such as MODIS, VIIRS (Visible infrared Imaging Radiometer) and CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation), and can also be from different ground observation stations.

[0158] Spatial fusion refers to the integration of information from multiple spatial data sources to improve data accuracy, spatial resolution or information richness. The aerosol optical depth data from multiple data sources can be fused by pixel-level fusion, decision-level fusion, model fusion, or deep learning fusion.

[0159] This embodiment does not limit the method of spatial fusion.

[0160] In one embodiment, the spatial fusion may adopt pixel-level fusion to spatially fuse the aerosol optical thickness data from the multiple data sources to obtain the aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data, including:

[0161] Determine primary data and secondary data, wherein the primary data is aerosol optical thickness data from a data source with the highest resolution among multiple data sources, and the secondary data is aerosol optical thickness data from data sources other than the primary data;

[0162] Obtaining, according to the secondary data, an estimated value of aerosol optical thickness data of an area not covered by the primary data in the optical satellite remote sensing data;

[0163] The aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data are determined according to the main data and the estimated value.

[0164] Exemplarily, the universal kriging method is used to spatially fuse multiple MODIS aerosol optical depth products to obtain aerosol optical depth data. In other examples, other spatial fusion algorithms may also be used.

[0165] The aerosol type used can be a global regional and seasonal custom dynamic aerosol type proposed by Levy et al. through long-term dynamic analysis of AERONET station aerosol observation data, which divides aerosols into four types: weakly absorbing, moderately absorbing, strongly absorbing and dust.

[0166] The various MODIS aerosol optical depth products used include the MOD08 AOD product, the MOD09CMA AOD product, and the MCD19A2 AOD product.

[0167] The spatial fusion of multiple MODIS AOD products using pan-kriging can reduce the situation of large-scale invalid values ​​of aerosol products, and has good continuity and consistency overall.

[0168] The pan-kriging method is a geostatistical method for unbiased linear estimation that takes into account the existence of spatial drift. For example, the MCD19A2 with the highest resolution is used as the primary data, and MOD09CMA AOD and MOD08 AOD are used as secondary data. For areas not covered by the primary data but covered by the secondary data, the influence of spatial position and secondary data is considered to estimate the aerosol optical thickness of the areas not covered by the primary data to obtain the estimated value of the aerosol optical thickness data. The estimated values ​​of the primary data and the aerosol optical thickness data are combined to obtain complete and continuous aerosol optical thickness data.

[0169] In step 306, the collected initial elevation data is processed to obtain elevation data matching the optical satellite remote sensing data; based on the elevation data matching the optical satellite remote sensing data, the elevation data of each pixel in the optical satellite remote sensing data is obtained.

[0170] Elevation data, namely DEM data, needs to be cropped, reprojected, and resampled so that the obtained DEM data has the same geographical range, resolution, and projection as the optical satellite remote sensing data to be inverted, so that each pixel in the optical satellite remote sensing data corresponds one-to-one with the data at the corresponding position of the pixel in the DEM data, making it convenient to read the DEM value of each pixel in the optical satellite remote sensing data to be inverted.

[0171] In step 308, radiation calibration is performed on each pixel of the optical satellite remote sensing data to be inverted, and the top atmosphere radiation brightness of the image pixel is calculated.

[0172] Radiometric calibration is the process of converting the digital quantization value of an image into a radiometric brightness value. The radiometric calibration coefficients involved can be obtained from the metadata of the optical satellite remote sensing data to be inverted.

[0173] In step 310, the observed geometric data, atmospheric auxiliary data, and elevation data of each pixel of the optical satellite remote sensing data to be inverted are input into the trained atmospheric correction neural network model to obtain the atmospheric correction coefficient x of the pixel. a 、x b 、x c .

[0174] For example, for each pixel, the solar zenith angle, satellite zenith angle, atmospheric water vapor content, ozone content, aerosol type, aerosol optical depth and elevation are first input into an atmospheric correction neural network model, such as an MLP neural network, to obtain x a 、x b 、x c Value

[0175] In step 312, the surface reflectance of each pixel of the optical satellite remote sensing data to be inverted is calculated by combining the atmospheric correction coefficient of each pixel and the top atmosphere radiance.

[0176] The atmospheric correction coefficient x of the pixel of the optical satellite remote sensing data to be inverted is obtained a 、x b 、x c Substituting the top atmosphere radiance value into formula (3), the surface reflectance of the pixel can be obtained, and then the surface reflectance of the optical satellite remote sensing data of each pixel can be obtained.

[0177] In the above embodiment, the atmospheric correction neural network model of the optical satellite remote sensing sensor to be inverted is obtained by pre-training based on the 6S model and the MLP neural network deep learning method, and a method of spatial fusion of multi-source aerosol optical thickness products based on the universal Kriging method is used to obtain complete and continuous aerosol optical thickness data. The solar zenith angle, satellite zenith angle, atmospheric water vapor content, ozone content, aerosol type and aerosol optical thickness, and DEM data of the optical satellite remote sensing data to be inverted are input pixel by pixel into the trained atmospheric correction neural network model to obtain the atmospheric correction coefficient of the optical satellite remote sensing data to be inverted pixel by pixel, and the surface reflectivity of the optical satellite remote sensing data to be inverted pixel by pixel is calculated by the top atmosphere radiation brightness of each pixel obtained after atmospheric radiation calibration.

[0178] The neural network atmospheric correction coefficient model of the optical satellite remote sensing sensor to be inverted is obtained by pre-training based on the radiation transfer model and deep learning method, which can be used to obtain the atmospheric correction coefficient of the optical satellite remote sensing sensor under any observation geometry, atmospheric condition and elevation condition.

[0179] In terms of accuracy, on the one hand, the sample data set used for training is based on the sensitivity analysis of the atmospheric correction input parameters, the change parameters and the change step size are determined, and the atmospheric correction lookup table is simulated and estimated using the radiation transfer model. The sample data set is reliable and representative. At the same time, the multilayer perceptron neural network model is an artificial neural network with nonlinear function fitting and adaptive learning capabilities. It can use the error back propagation algorithm to autonomously learn the nonlinear relationship between the dependent variable and the independent variable in a data-driven manner, and has higher accuracy than the traditional linear interpolation method. On the other hand, in constructing such a method based on the solar zenith angle, , satellite zenith angle, atmospheric water vapor content, ozone content, aerosol type and aerosol optical thickness, and elevation as dependent variables, and the atmospheric correction coefficient as the independent variable of the atmospheric correction neural network model, it is possible to realize the surface reflectivity inversion of remote sensing images that considers the solar zenith angle, satellite zenith angle, atmospheric water vapor content, ozone content, aerosol type and aerosol optical thickness, and elevation changes pixel by pixel. Compared with the surface reflectivity inversion of a scene data that only considers a fixed solar zenith angle, observation zenith angle, atmospheric water vapor, aerosol type, ozone content and elevation, it has higher and more reasonable accuracy, especially for satellite remote sensing data with a larger width.

[0180] In terms of applicability, as mentioned above, when constructing the neural network atmospheric correction coefficient model, the atmospheric correction sensitive parameters are fully considered: changes in solar zenith angle, observation zenith angle, atmospheric water vapor, aerosol type and aerosol optical thickness, ozone content, and elevation. The surface reflectivity inversion of remote sensing images that consider solar zenith angle, observation zenith angle, atmospheric water vapor, aerosol type, aerosol optical thickness, ozone content, and elevation changes can be realized pixel by pixel. This method is not only applicable to conventional Landsat, Sentinel-2 and other narrow-width optical satellite remote sensing data, but also to GF-6WFV, SDGSAT-1 and other wide-width optical satellite remote sensing data. It is suitable for most optical satellite remote sensing data and has strong universality.

[0181] In terms of efficiency, although it takes a relatively long time to construct the multidimensional atmospheric correction coefficient lookup table of the optical satellite remote sensing sensor to be inverted using the 6S model in the early stage, this step can be completed in advance and the processing can be accelerated by deploying multiple computers and using multi-threading. However, once completed, it is a one-time thing. As long as the remote sensing image of the satellite sensor is used for surface reflectivity inversion, the atmospheric correction neural network model based on the satellite sensor can input the solar zenith angle, observation zenith angle, atmospheric water vapor, aerosol type, aerosol optical thickness, ozone content, and elevation of the image to be inverted pixel by pixel to quickly obtain the atmospheric correction coefficient for each pixel, thereby quickly realizing the inversion of surface reflectivity.

[0182] The atmospheric correction neural network model corresponding to the optical satellite sensor to be inverted is obtained through pre-training to generate the atmospheric correction coefficients, which solves the shortcomings of long lookup time in the traditional lookup table and unstable accuracy of the atmospheric correction results of optical satellite remote sensing data, improves the efficiency and accuracy of surface reflectivity inversion of optical satellite remote sensing data, and is suitable for surface reflectivity inversion of most optical satellite remote sensing data with strong universality.

[0183] In addition, by coordinating the multi-source MODIS aerosol optical depth products and adopting a spatial fusion method based on the universal kriging method, complete and continuous aerosol optical depth data can be obtained, which can solve the problems of MODIS aerosol optical depth products with a large number of invalid values ​​and incomplete coverage, as well as the problem of data distortion caused by the use of traditional interpolation algorithms in areas with a large number of invalid values, thereby further improving the accuracy of the surface reflectivity affected by optical satellite remote sensing to be inverted.

[0184] In summary, the surface reflection inversion method provided in this embodiment has the characteristics of strong universality, high inversion accuracy and fast calculation speed.

[0185] The disclosed embodiment also provides a surface reflectivity inversion system for optical satellite remote sensing data. The structure of the system is as follows: Figure 4 As shown, the system includes modules M11-M12 and modules M21-M24, wherein modules M11-M12 are used to obtain the atmospheric correction neural network model of the optical satellite sensor to be inverted. When the atmospheric correction corresponding neural network model corresponding to the satellite sensor exists, modules M11 and M12 can be skipped and modules M21-M24 can be directly run. The modules of the system can be deployed in different devices.

[0186] The lookup table construction module M11 is used to: obtain the spectral response function of the optical satellite sensor and the sample sensitive parameter data under different sensitive parameter conditions, wherein the sensitive parameter conditions include: observation geometry conditions, atmospheric conditions, and elevation conditions; for each spectral band in the spectral response function, calculate the sample atmospheric correction coefficient corresponding to the sample sensitive parameter data according to the sample sensitive parameter data and the radiation transfer model.

[0187] For example, the spectral response function of each band of the optical satellite sensor to be inverted is input. When traversing different solar zenith angles, satellite zenith angles, atmospheric water vapor content, ozone content, aerosol optical thickness and elevation values, the 6S radiation transfer model is used to construct the atmospheric correction coefficient x with the solar zenith angle, satellite zenith angle, atmospheric water vapor content, ozone content, aerosol optical thickness and elevation as independent variables. a 、x b 、x c A lookup table of multidimensional atmospheric correction coefficients for the optical satellite sensor to be inverted as the dependent variable.

[0188] The model training module M12 is used to: train the atmospheric correction neural network model using a deep learning model according to the constructed atmospheric correction coefficient lookup table, and obtain the atmospheric correction neural network model of the optical satellite sensor to be inverted. Specifically, the sample sensitive parameter data in the atmospheric correction coefficient lookup table is input into the initial neural network to obtain the initial atmospheric correction coefficient output by the initial neural network; based on the difference between the initial atmospheric correction coefficient and the sample atmospheric correction coefficient, a loss function is determined; according to the loss function, the initial neural network is trained until the loss function meets the preset stop training condition, and the atmospheric correction neural network model corresponding to the optical satellite sensor is obtained.

[0189] For example, the atmospheric correction coefficient lookup table constructed by module M11 is used as a sample data set, wherein 80% of the sample data is used as a training set to input into a deep learning model of a multilayer perceptron neural network for training and learning, and 20% of the sample data is used as a test set for verification, thereby obtaining the atmospheric correction neural network model of the optical satellite sensor to be inverted.

[0190] The auxiliary data acquisition and processing module M21 is used to obtain sensitive parameter data corresponding to pixels, where the pixels are the pixels in the optical satellite remote sensing data to be inverted.

[0191] The auxiliary data obtained above include observation geometry data, atmospheric auxiliary data, and elevation data; observation geometry data include solar zenith angle, solar azimuth angle, observation zenith angle, and observation azimuth angle data; atmospheric auxiliary data include atmospheric water vapor content, ozone content, aerosol type, and aerosol optical thickness data; elevation data include elevation data; observation geometry data are calculated and obtained by spatial interpolation using optical satellite remote sensing data to be inverted; atmospheric auxiliary data and elevation data can be obtained from a preset atmospheric auxiliary database; after obtaining atmospheric auxiliary data and elevation data, resampling, reprojection, and clipping are performed to obtain sensitive parameter data with consistent geographic range, resolution, and projection with the optical satellite remote sensing data to be inverted. It is worth noting that after reading a variety of aerosol optical thickness data, the universal kriging method is used to spatially fuse a variety of aerosol optical thickness data to obtain aerosol optical thickness data with complete coverage and spatial continuity.

[0192] The radiation calibration module M22 is used to perform radiation calibration on each pixel of the optical satellite remote sensing data to be inverted, and obtain the top atmosphere radiation brightness of the pixel;

[0193] The atmospheric correction coefficient module M23 is used to: input the sensitive parameter data of each pixel of the optical satellite remote sensing data to be inverted into the atmospheric correction neural network model of the optical satellite sensor to be inverted, and obtain the atmospheric correction coefficient of the pixel;

[0194] The surface reflectance calculation module M24 is used to obtain the surface reflectance of each pixel by combining the atmospheric top layer radiation brightness of each pixel of the optical satellite remote sensing data to be inverted with the atmospheric correction coefficient of the pixel.

[0195] In an embodiment of the present invention, a lookup table for the atmospheric correction coefficient of the optical satellite sensor to be inverted is obtained through a lookup table construction module M11; the constructed atmospheric correction coefficient lookup table is used as a sample data set and input into a model training module M12, and a neural network model for atmospheric correction of the optical satellite sensor to be inverted is obtained through learning and training; and sensitive parameter data having a consistent geographical range, resolution, projection and the same time as the optical satellite remote sensing data to be inverted is obtained through an auxiliary data acquisition and processing module M21, so as to obtain the solar zenith angle, solar azimuth angle, observation zenith angle, observation azimuth angle, atmospheric water vapor content, Ozone content, aerosol type and aerosol optical thickness, and elevation data values; obtain the top atmosphere radiation brightness of each pixel in the optical satellite remote sensing data to be inverted through the radiation calibration module M22; input the auxiliary data value of each pixel obtained by the module M21 into the atmospheric correction neural network model obtained by the module M12, and obtain the atmospheric correction coefficient of each pixel in the optical satellite remote sensing data to be inverted through the atmospheric correction module M23; input the obtained atmospheric correction coefficient and top atmosphere radiation brightness of each pixel of the optical satellite remote sensing data to be inverted into the surface reflectivity calculation module M24 to obtain the surface reflectivity of each pixel of the optical satellite remote sensing data to be inverted.

[0196] like Figure 5 As shown, Figure 5 is a block diagram of a device for inverting surface reflectivity of optical satellite remote sensing data shown in at least one embodiment of the present disclosure, the device comprising:

[0197] A data acquisition module 51 is used to acquire sensitive parameter data corresponding to a pixel, wherein the pixel is each pixel in the optical satellite remote sensing data to be inverted, wherein the optical satellite remote sensing data is acquired by an optical satellite sensor, and the sensitive parameter data includes observation geometry data, atmospheric auxiliary data, and elevation data;

[0198] An atmospheric correction module 52 is used to: input the sensitive parameter data into an atmospheric correction neural network model to obtain an atmospheric correction coefficient of the pixel, wherein the atmospheric correction neural network model is pre-trained based on an atmospheric correction coefficient lookup table of the optical satellite sensor;

[0199] The reflectivity calculation module 53 is used to calculate the surface reflectivity of the optical satellite remote sensing data according to the atmospheric correction coefficient and the radiation brightness of each pixel.

[0200] In some embodiments, the atmospheric correction neural network model is obtained by the following training method:

[0201] Acquire an atmospheric correction coefficient lookup table of the optical satellite sensor, wherein the atmospheric correction coefficient lookup table is constructed according to a mapping relationship between sample sensitive parameter data and sample atmospheric correction coefficients;

[0202] Inputting the sample sensitive parameter data in the atmospheric correction coefficient lookup table into the initial neural network to obtain the initial atmospheric correction coefficient output by the initial neural network;

[0203] determining a loss function based on a difference between the initial atmospheric correction coefficient and the sample atmospheric correction coefficient;

[0204] The initial neural network is trained according to the loss function until the loss function satisfies a preset training stop condition, thereby obtaining an atmospheric correction neural network model corresponding to the optical satellite sensor.

[0205] In some embodiments, obtaining an atmospheric correction coefficient lookup table of the optical satellite sensor includes:

[0206] Acquire the spectral response function of the optical satellite sensor and the sample sensitive parameter data under different sensitive parameter conditions, wherein the sensitive parameter conditions include: observation geometry conditions, atmospheric conditions, and elevation conditions;

[0207] For each spectral band in the spectral response function, a sample atmosphere correction coefficient corresponding to the sample sensitive parameter data is calculated according to the sample sensitive parameter data and a radiation transfer model.

[0208] In some embodiments, the atmospheric auxiliary data includes: aerosol optical thickness data;

[0209] The data acquisition module 51, when used to acquire aerosol optical thickness data corresponding to a pixel, is specifically used to:

[0210] Obtain aerosol optical depth data from multiple data sources;

[0211] The aerosol optical thickness data from the multiple data sources are spatially fused to obtain the aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data.

[0212] In some embodiments, the data acquisition module 51, when used to spatially fuse the aerosol optical thickness data from the multiple data sources to obtain the aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data, is specifically used to:

[0213] Determine primary data and secondary data, wherein the primary data is aerosol optical thickness data from a data source with the highest resolution among multiple data sources, and the secondary data is aerosol optical thickness data from data sources other than the primary data;

[0214] Obtaining, according to the secondary data, an estimated value of aerosol optical thickness data of an area not covered by the primary data in the optical satellite remote sensing data;

[0215] The aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data are determined according to the main data and the estimated value.

[0216] In some embodiments, the observation geometry data includes: a solar zenith angle and an observation zenith angle;

[0217] The atmospheric auxiliary data include: aerosol type, aerosol optical thickness, atmospheric water vapor content, and ozone content;

[0218] The elevation data is digital elevation model data.

[0219] In some embodiments, the data acquisition module 51, when used to acquire sensitive parameter data corresponding to a pixel, is specifically used to:

[0220] Based on the optical satellite remote sensing data, obtaining the observed geometric data corresponding to the pixel;

[0221] Processing the collected initial atmospheric auxiliary data to obtain atmospheric auxiliary data matching the optical satellite remote sensing data;

[0222] Based on the atmospheric auxiliary data matched with the optical satellite remote sensing data, obtaining the atmospheric auxiliary data of each pixel in the optical satellite remote sensing data;

[0223] Processing the collected initial elevation data to obtain elevation data matching the optical satellite remote sensing data;

[0224] Based on the elevation data matched with the optical satellite remote sensing data, the elevation data of each pixel in the optical satellite remote sensing data is obtained.

[0225] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.

[0226] The present disclosure also provides an electronic device, such as Figure 6As shown, the electronic device includes a memory 61 and a processor 62, wherein the memory 61 is used to store computer instructions that can be executed on the processor, and the processor 62 is used to implement the surface reflectivity inversion method of optical satellite remote sensing data described in any embodiment of the present disclosure when executing the computer instructions.

[0227] An embodiment of the present disclosure also provides a computer program product, which includes a computer program / instructions. When the computer program / instructions are executed by a processor, the method for inverting the surface reflectivity of optical satellite remote sensing data described in any embodiment of the present disclosure is implemented.

[0228] An embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the method for inverting the surface reflectivity of optical satellite remote sensing data as described in any embodiment of the present disclosure is implemented.

[0229] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this specification. A person of ordinary skill in the art can understand and implement it without paying creative labor.

[0230] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0231] Those skilled in the art will readily appreciate other embodiments of the specification after considering the specification and practicing the invention claimed herein. The specification is intended to cover any variations, uses or adaptations of the specification that follow the general principles of the specification and include common knowledge or customary techniques in the art that are not claimed in the specification. The specification and examples are to be considered exemplary only, and the true scope and spirit of the specification are indicated by the following claims.

[0232] It should be understood that the present description is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present description is limited only by the appended claims.

[0233] The above description is only a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this specification should be included in the scope of protection of this specification.

Claims

1. A method for inverting surface reflectivity of optical satellite remote sensing data, characterized in that: The method comprises: Acquire sensitive parameter data corresponding to a pixel, wherein the pixel is each pixel in the optical satellite remote sensing data to be inverted, wherein the optical satellite remote sensing data is acquired by an optical satellite sensor, and the sensitive parameter data includes observation geometry data, atmospheric auxiliary data, and elevation data; Inputting the sensitive parameter data into an atmospheric correction neural network model to obtain an atmospheric correction coefficient of the pixel, wherein the atmospheric correction neural network model is pre-trained based on an atmospheric correction coefficient lookup table of the optical satellite sensor; The surface reflectivity of the optical satellite remote sensing data is obtained by calculating according to the atmospheric correction coefficient and the radiation brightness of each pixel.

2. The method according to claim 1, characterized in that The atmospheric correction neural network model is obtained by the following training method: Acquire an atmospheric correction coefficient lookup table of the optical satellite sensor, wherein the atmospheric correction coefficient lookup table is constructed according to a mapping relationship between sample sensitive parameter data and sample atmospheric correction coefficients; Inputting the sample sensitive parameter data in the atmospheric correction coefficient lookup table into the initial neural network to obtain the initial atmospheric correction coefficient output by the initial neural network; determining a loss function based on a difference between the initial atmospheric correction coefficient and the sample atmospheric correction coefficient corresponding to the sample sensitive parameter data; The initial neural network is trained according to the loss function until the loss function satisfies a preset stop training condition, thereby obtaining the atmospheric correction neural network model.

3. The method according to claim 2, characterized in that Obtaining an atmospheric correction coefficient lookup table for the optical satellite sensor, comprising: Acquire the spectral response function of the optical satellite sensor and the sample sensitive parameter data under different sensitive parameter conditions, wherein the sensitive parameter conditions include: observation geometry conditions, atmospheric conditions, and elevation conditions; For each spectral band in the spectral response function, a sample atmosphere correction coefficient corresponding to the sample sensitive parameter data is calculated according to the sample sensitive parameter data and a radiation transfer model.

4. The method according to claim 1, characterized in that: The atmospheric auxiliary data include: aerosol optical thickness data; When obtaining the aerosol optical thickness data corresponding to the pixel, it includes: Obtain aerosol optical depth data from multiple data sources; The aerosol optical thickness data from the multiple data sources are spatially fused to obtain the aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data.

5. The method according to claim 4, characterized in that The aerosol optical thickness data from the multiple data sources are spatially fused to obtain the aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data, including: Determine primary data and secondary data, wherein the primary data is aerosol optical thickness data from a data source with the highest resolution among multiple data sources, and the secondary data is aerosol optical thickness data from data sources other than the primary data; Obtaining, according to the secondary data, an estimated value of aerosol optical thickness data of an area not covered by the primary data in the optical satellite remote sensing data; The aerosol optical thickness data corresponding to each pixel in the optical satellite remote sensing data are determined according to the main data and the estimated value.

6. The method according to claim 1, characterized in that The observation geometry data includes: solar zenith angle and observation zenith angle; The atmospheric auxiliary data include: aerosol type, aerosol optical thickness, atmospheric water vapor content, and ozone content; The elevation data is digital elevation model data.

7. The method according to any one of claims 1 to 6, characterized in that: The step of obtaining sensitive parameter data corresponding to the pixel includes: Based on the optical satellite remote sensing data, obtaining the observed geometric data corresponding to the pixel; Processing the collected initial atmospheric auxiliary data to obtain atmospheric auxiliary data matching the optical satellite remote sensing data; Based on the atmospheric auxiliary data matched with the optical satellite remote sensing data, obtaining the atmospheric auxiliary data of each pixel in the optical satellite remote sensing data; Processing the collected initial elevation data to obtain elevation data matching the optical satellite remote sensing data; Based on the elevation data matched with the optical satellite remote sensing data, the elevation data of each pixel in the optical satellite remote sensing data is obtained.

8. A surface reflectivity inversion device for optical satellite remote sensing data, characterized in that: The device comprises: A data acquisition module is used to acquire sensitive parameter data corresponding to a pixel, wherein the pixel is each pixel in the optical satellite remote sensing data to be inverted, wherein the optical satellite remote sensing data is acquired by an optical satellite sensor, and the sensitive parameter data includes observation geometry data, atmospheric auxiliary data, and elevation data; An atmospheric correction module, used for inputting the sensitive parameter data into an atmospheric correction neural network model to obtain an atmospheric correction coefficient of the pixel, wherein the atmospheric correction neural network model is pre-trained based on an atmospheric correction coefficient lookup table of the optical satellite sensor; The reflectivity calculation module is used to calculate the surface reflectivity of the optical satellite remote sensing data according to the atmospheric correction coefficient and the radiation brightness of each pixel.

9. An electronic device, characterized in that: The device includes a memory and a processor, wherein the memory is used to store computer instructions that can be executed on the processor, and the processor is used to implement the surface reflectivity inversion method of optical satellite remote sensing data as described in any one of claims 1 to 7 when executing the computer instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for inverting the surface reflectivity of optical satellite remote sensing data as described in any one of claims 1 to 7 is implemented.

Images