FY3D and MODIS data reflectivity correction algorithm based on machine learning

Through the FY3D and MODIS data reflectivity correction algorithm based on machine learning, the problem of insufficient vegetation index time series noise and data repair accuracy in MODIS data is solved, and higher accuracy and stable remote sensing vegetation index acquisition is achieved, enhancing the time and space of data Promotional.

CN118982491AInactive Publication Date: 2024-11-19青海省气象科学研究所
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411025779.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing MODIS data processing methods have noise problems in the remote sensing vegetation index time series, and the missing data repair scheme fails to fully utilize the characteristics of the data itself, resulting in low accuracy and insufficient spatial and temporal promotion.

Method used

Using the reflectivity correction algorithm of FY3D and MODIS data based on machine learning, the pixel-by-cell atmospheric correction algorithm is gradually formed through geometric correction, radiation positioning, solar zenith angle correction, 6S radiation transmission model and autonomous learning method model, and combined with deep learning The intelligent matching algorithm outputs the NDVI vegetation index.

Benefits of technology

It improves the accuracy and stability of remote sensing images, comprehensively considers the sensitivity of multiple parameters, and more accurately obtains the true reflectivity of land objects, enhancing the spatial and temporal generalization of data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982491A_ABST
    Figure CN118982491A_ABST
Patent Text Reader

Abstract

The invention relates to a reflectivity correction algorithm for FY3D and MODIS data based on machine learning, and belongs to the technical field of optical remote sensing image processing. By obtaining the surface reflectance, the solar zenith angle, the atmospheric apparent reflectance and the autonomous learning of the autonomous learning method model, a pixel-by-pixel atmospheric correction algorithm is formed step by step, so that the algorithm is more stable, various parameters with different sensitivities are comprehensively considered, and the accuracy of atmospheric correction is improved. Compared with the conventional common algorithm for obtaining the average parameter based on the whole-scene remote sensing image, the algorithm is closer to the actual situation, atmospheric correction can be more accurately carried out on the remote sensing image, and the real reflectivity of the ground feature can be obtained.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of optical remote sensing image processing, and in particular to a reflectivity correction algorithm for FY3D and MODIS data based on machine learning. Background Art

[0002] MODIS is the world's new generation of "map-in-one" optical remote sensing instrument, with 36 discrete spectral bands and a wide spectral range, covering the entire spectrum from 0.4 microns (visible light) to 14.4 microns (thermal infrared). MODIS's multi-band data can simultaneously provide information reflecting the land surface conditions, cloud boundaries, cloud characteristics, ocean water color, phytoplankton, biogeography, chemistry, atmospheric water vapor, aerosols, surface temperature, cloud top temperature, atmospheric temperature, ozone and cloud top height. It can be used for long-term global observations of the surface, biosphere, solid earth, atmosphere and ocean. Remote sensing vegetation index time series data reflects the dynamic changes in vegetation. However, it is affected by factors such as the atmosphere, cloud cover, and ice and snow cover, resulting in a large amount of noise in remote sensing vegetation index images, which limits the in-depth analysis and application of the data. At present, there are more than 20 methods for time series data reconstruction (filtering, denoising), mainly including the Best Index Slope Extraction (BISE), the median filtering method, and the MISE method.

[0003] (Median Filtering, MF), SG method (Savitzky Golay, SG), Fourier transform method (FourierTransform, FT), wavelet transform method (Wavelet Transform, WT), iterative interpolation data reconstruction method (Interpolation for Data Reconstruction, IDR), mean value iteration filter method (Mean ValueIteration filter, MVI), maximum value composite method (Maximum Value Composite, MVC), etc.

[0004] The above time series data reconstruction methods can generally be divided into three types, each of which has advantages and disadvantages in reconstructing NDVI time series data. The first is a local filtering method in the time domain, such as BISE. This method is based on the predictability of vegetation changes. The filtered results are often close to the original NDVI time series, but it requires a sliding period to determine the NDVI regeneration percentage, and it is also necessary to determine a smooth area to indicate whether the increase in the area is within an acceptable range. Therefore, when the NDVI time series shows a long-term downward trend, this method makes the extracted information unreliable. The second is a frequency domain denoising method, such as FT. This method fits the first three harmonics with a least squares solver. The NDVI curve obtained by this method is relatively smooth, and the vegetation growth characteristics are obvious. However, it is not suitable for irregular or asymmetric NDVI data. The third is a function-based method, which is based on the least squares of the piecewise function of the NDVI time series, such as the dual logistic function fitting method. However, this method is difficult to fit the NDVI time series with irregular seasonal time distribution, and its application is limited when the area is large. It should be pointed out that the vegetation index time series is a kind of time series data with strong periodicity, but due to cloud and snow occlusion, a certain number of mutation points will be generated. These mutation points are high-frequency components in the frequency domain. Therefore, proper smoothing of such high-frequency components is the key to denoising the vegetation index time series data. Wavelet transform filtering is an effective frequency domain filtering algorithm. It inherits and develops the localization idea of ​​short-time Fourier transform and overcomes certain shortcomings. However, the filtering effect is still not ideal for mutation points or continuous mutation points with obvious changes in vegetation index, and further optimization is needed. In addition, the above method can only focus on the change characteristics of NDVI in the time-frequency dimension, and cannot consider the spatial correlation of NDVI.

[0005] There are three main methods for repairing missing data of MODIS: fusion based on different algorithms, fusion based on multi-source sensor data, and data interpolation based on the spatiotemporal information of the image data itself. Data fusion based on different algorithms is based on the spatial complementarity of different algorithms, but the existing methods often use simple linear fusion without considering the applicability of the algorithm in a specific area; fusion based on multi-source sensor data includes the fusion of data between different satellite sensors and between satellites and ground observations. Due to the different data sources of this method, there is a problem of matching the spatiotemporal information of the data; spatiotemporal information fusion based on the image data itself often uses a simple spatial interpolation method, which does not make full use of the time information. And no further correction is performed. At present, most of the repair schemes for MODIS data only use one or two aspects of information, and do not make full use of the characteristics of MODIS data itself to design targeted and comprehensive improvement schemes. The repair results are not comprehensive enough, the accuracy is not high enough, and the spatiotemporal generalization is not strong. Summary of the invention

[0006] The present invention aims to solve the above technical problems and provides a reflectivity correction algorithm for FY3D and MODIS data based on machine learning.

[0007] In order to solve the above technical problems, the technical solution provided by the present invention is:

[0008] A reflectivity correction algorithm for FY3D and MODIS data based on machine learning, including the following steps:

[0009] S1. Get the original data:

[0010] Obtain FY3D and MODIS data, including images, satellite parameters, metadata, spatial coordinate data and latitude;

[0011] S2. Geometric correction:

[0012] Use bilinear interpolation or 2-4 spline curves to interpolate longitude and latitude coordinates to generate coordinate points that correspond to the image one by one, and then use the interpolated coordinate points to perform projection-free geometric correction on the MODIS image;

[0013] S3. Radiation positioning:

[0014] Substitute the satellite parameters into the equation to calculate the surface reflectivity, as shown in Formula 1;

[0015] f=s×(Yc) Formula 1;

[0016] S4, the solar zenith angle is corrected,

[0017] According to formula 2, the effect of different sun altitudes on reflectivity is eliminated.

[0018] F = f × cos(θ) Formula 2;

[0019] S5. Calibration:

[0020] The 6S radiation transfer model is established for atmospheric correction. After running the 6S radiation transfer model, the corrected atmospheric apparent reflectivity is calculated according to Formula 3:

[0021]

[0022] S6. Autonomous learning method model:

[0023] Establish an autonomous learning method model, match the feature points based on the deep learning intelligent matching algorithm, and calculate the matching rate. The intelligent matching algorithm is shown in Formula 4.

[0024]

[0025] S7. Vegetation detection:

[0026] The vegetation remote sensing model is established by using the spectral reflectance of the ground objects and biophysical quantities after considering the solar zenith angle and the apparent reflectance of the atmosphere. The vegetation remote sensing model outputs the NDVI vegetation index and image.

[0027] Preferably, the f is the reflectivity, the s is the scaling factor of the reflectivity, the c is the offset parameter of the reflectivity, and the Y is the original recorded value of the satellite data.

[0028] Preferably, the F is the corrected reflectivity, and the θ is the solar altitude angle.

[0029] Preferably, after the solar zenith angle is corrected, the solar reflection bands 1, 2, 3, 6, 7, 8, and 26 are calibrated to the emissivity, and the thermal radiation bands 29, 31, and 32 are calibrated to the radiant brightness. The radiant brightness is converted into the brightness temperature as shown in Formula 5.

[0030] Wherein T is the brightness temperature, h is the Planck constant, P is the speed of light in a vacuum, k is the Boltzmann constant, t is the corresponding channel wavelength, and L is the radiation brightness.

[0031] Preferably, F1 is the apparent reflectivity of the atmosphere, F2 is the equivalent reflectivity of the path radiation term of the atmosphere, V1 is the solar azimuth, V2 is the satellite azimuth, T1 is the downward transmittance of the atmosphere, T2 is the total upward radiation transmittance, and S is the albedo of the large balloon surface.

[0032] Preferably, the M is a substitution coefficient, the m is the number of correct matching point pairs, the p1 is the actual number of feature points, the P2 is the number of feature points extracted by the autonomous learning method model, the K1 is a correction coefficient, and the M is a matching rate.

[0033] Preferably, the value of K1 is 0.8-1.

[0034] Preferably, the larger the calculated value of M is, the closer the value of K1 is to 1.

[0035] After adopting the above method, the present invention has the following advantages:

[0036] The present invention obtains the surface reflectivity, the solar zenith angle for correction, the atmospheric apparent reflectivity and uses the autonomous learning method model to gradually form an algorithm for pixel-by-pixel atmospheric correction, so that the algorithm is more stable and comprehensively considers various parameters with different sensitivities. Compared with the commonly used algorithm for obtaining average parameters based on the whole remote sensing image, the algorithm is closer to the actual situation and can more accurately perform atmospheric correction on the remote sensing image and obtain the true reflectivity of the ground object.

[0037] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1 Schematic diagram of the reflectivity of the present invention.

[0040] Figure 2 It is a schematic diagram of the vegetation index of the present invention. DETAILED DESCRIPTION

[0041] Specific embodiments of the present invention will now be mentioned in detail. Although the present invention is described in conjunction with these specific embodiments, it should be appreciated that it is not intended to limit the present invention to these specific embodiments. On the contrary, these embodiments are intended to cover substitutions, changes or equivalent embodiments that may be included in the spirit and scope of the invention defined by the claims. In the following description, a large number of specific details are set forth in order to provide a comprehensive understanding of the present invention. The present invention can be implemented without some or all of these specific details. In other cases, in order not to make the present invention unnecessarily obscure, well-known process operations are not described in detail.

[0042] When used in conjunction with "including," "methods comprising," or similar language in this specification and the appended claims, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0043] The present invention is further described in detail below in conjunction with the full text.

[0044] Combined with Figure 1 ,

[0045] A reflectivity correction algorithm for FY3D and MODIS data based on machine learning, including the following steps:

[0046] S1. Get the original data:

[0047] Obtain FY3D and MODIS data, including images, satellite parameters, metadata, spatial coordinate data and latitude;

[0048] S2. Geometric correction:

[0049] Use bilinear interpolation or 2-4 spline curves to interpolate longitude and latitude coordinates to generate coordinate points that correspond to the image one by one, and then use the interpolated coordinate points to perform projection-free geometric correction on the MODIS image;

[0050] S3. Radiation positioning:

[0051] Substitute the satellite parameters into the equation to calculate the surface reflectivity, as shown in Formula 1;

[0052] f=s×(Yc) Formula 1;

[0053] S4, the solar zenith angle is corrected,

[0054] According to formula 2, the effect of different sun altitudes on reflectivity is eliminated.

[0055] F = f × cos(θ) Formula 2;

[0056] S5. Calibration:

[0057] The 6S radiation transfer model is established for atmospheric correction. After running the 6S radiation transfer model, the corrected atmospheric apparent reflectivity is calculated according to Formula 3:

[0058]

[0059] S6. Autonomous learning method model:

[0060] Establish an autonomous learning method model, match the feature points based on the deep learning intelligent matching algorithm, and calculate the matching rate. The intelligent matching algorithm is shown in Formula 4.

[0061]

[0062] S7. Vegetation detection:

[0063] The vegetation remote sensing model is established by using the spectral reflectance of the ground objects and biophysical quantities after considering the solar zenith angle and the apparent reflectance of the atmosphere. The vegetation remote sensing model outputs the NDVI vegetation index and image.

[0064] The f is the reflectivity, the s is the scaling factor of the reflectivity, the c is the offset parameter of the reflectivity, and the Y is the original recorded value of the satellite data.

[0065] The F is the corrected reflectivity, and the θ is the solar altitude angle.

[0066] After the solar zenith angle is corrected, the solar reflection bands 1, 2, 3, 6, 7, 8, and 26 are calibrated to the emissivity, and the thermal radiation bands 29, 31, and 32 are calibrated to the radiant brightness. The radiant brightness is converted into the brightness temperature as shown in Formula 5.

[0067] Where T is the brightness temperature, h is the Planck constant, P is the speed of light in a vacuum, k is the Boltzmann constant, t is the wavelength of the corresponding channel, L is the radiation brightness, and the Planck constant is 6.626x10 -34 j·s, the speed of light in a vacuum is 2.998x 10 8 m·s -1 , the value of the Boltzmann constant is 1.381x10 -23 j.K -1 .

[0068] The F1 is the apparent reflectivity of the atmosphere, the F2 is the equivalent reflectivity of the path radiation term of the atmosphere, the V1 is the solar azimuth, the V2 is the satellite azimuth, the T1 is the downward transmittance of the atmosphere, the T2 is the total upward radiation transmittance, and the S is the albedo of the large balloon surface.

[0069] The M is the substitution coefficient, the m is the number of correct matching points, the p1 is the actual number of feature points, the P2 is the number of feature points extracted by the autonomous learning method model, the K1 is the correction coefficient, and the M is the matching rate. After the image is styled, although the modal difference between the cross-modal images is reduced, the converted infrared image and the real infrared image are still heterogeneous images, and there are still certain differences in the resolution and detail texture of the image. The autonomous learning method model calculates the matching rate based on the feature points matched by the intelligent matching algorithm based on deep learning. The number of feature points matched by the intelligent matching algorithm increases significantly, and the matching results are more intensive, especially for sparse texture areas with fewer feature points. The higher the correct matching rate, the more feature points the algorithm correctly matches, and the better the matching performance. The larger the value calculated by M, the closer the value of K1 is to 1.

[0070] The value of K1 is 0.8-1.

[0071] Within a certain range, if the step length of the parameter is too long, the accuracy of the correction will be affected, while if the step length is too small, the data volume will be too large and there will be more redundant data. Therefore, the range of the parameters can be set according to the situation of the study area, and the step length of each parameter can be designed separately, and the lookup table can be established according to the study area, rather than using a fixed set of lookup tables.

[0072] The 6S radiation transfer model uses the scattering calculation method of SOS, which is based on the radiation transfer theory. It accurately simulates the atmospheric influence on the path of the sun, target, and sensor. It has a wide range of applications, is less affected by the characteristics of the study area and the type of target, and has high accuracy. The 6S radiation transfer model is suitable for simulating the atmospheric radiation transmission of electromagnetic waves in the wavelength range of 0.25 to 4.00 μm. The workflow of atmospheric correction using this model is to input the geometric parameters, atmospheric model, aerosol model, visibility, altitude, observation band and other parameters into the 6S radiation transfer model in sequence, and give the atmospheric correction coefficient by calculating and simulating the radiation status of solar radiation in the atmosphere.

[0073] Using the 6S radiation transfer model for atmospheric correction can significantly improve the accuracy of remote sensing images. Since the 6S radiation transfer model can accurately simulate the complex effects of the atmosphere on electromagnetic waves, it can more realistically reflect the actual situation of the surface.

[0074] Embodiment 1:

[0075] A reflectivity correction algorithm for FY3D and MODIS data based on machine learning, including the following steps:

[0076] S1. Get the original data:

[0077] Obtain FY3D and MODIS data, including images, satellite parameters, metadata, spatial coordinate data and latitude;

[0078] S2. Geometric correction:

[0079] Use bilinear interpolation or 2-4 spline curves to interpolate longitude and latitude coordinates to generate coordinate points that correspond to the image one by one, and then use the interpolated coordinate points to perform projection-free geometric correction on the MODIS image;

[0080] S3. Radiation positioning:

[0081] Substitute the satellite parameters into the equation to calculate the surface reflectivity, as shown in Formula 1;

[0082] f=s×(Yc) Formula 1;

[0083] S4, the solar zenith angle is corrected,

[0084] According to formula 2, the effect of different sun altitudes on reflectivity is eliminated.

[0085] F = f × cos(θ) Formula 2;

[0086] S5. Calibration:

[0087] The 6S radiation transfer model is established for atmospheric correction. After running the 6S radiation transfer model, the corrected atmospheric apparent reflectivity is calculated according to Formula 3:

[0088]

[0089] S6. Vegetation detection:

[0090] The vegetation remote sensing model is established by eliminating the reflectivity of the sun's altitude and combining it with biophysical quantities. The vegetation remote sensing model outputs the NDVI vegetation index and image to obtain the apparent reflectance and vegetation index.

[0091] Embodiment 2:

[0092] A reflectivity correction algorithm for FY3D and MODIS data based on machine learning, including the following steps:

[0093] S1. Get the original data:

[0094] Obtain FY3D and MODIS data, including images, satellite parameters, metadata, spatial coordinate data and latitude;

[0095] S2. Geometric correction:

[0096] Use bilinear interpolation or 2-4 spline curves to interpolate longitude and latitude coordinates to generate coordinate points that correspond to the image one by one, and then use the interpolated coordinate points to perform projection-free geometric correction on the MODIS image;

[0097] S3. Radiation positioning:

[0098] Substitute the satellite parameters into the equation to calculate the surface reflectivity, as shown in Formula 1;

[0099] f=s×(Yc) Formula 1;

[0100] S4, the solar zenith angle is corrected,

[0101] According to formula 2, the effect of different sun altitudes on reflectivity is eliminated.

[0102] F = f × cos(θ) Formula 2;

[0103] S5. Calibration:

[0104] The 6S radiation transfer model is established for atmospheric correction. After running the 6S radiation transfer model, the corrected atmospheric apparent reflectivity is calculated according to Formula 3:

[0105]

[0106] S6. Autonomous learning method model:

[0107] Establish an autonomous learning method model, match the feature points based on the deep learning intelligent matching algorithm, and calculate the matching rate. The intelligent matching algorithm is shown in Formula 4. At this time, M is small and K1 is 0.8.

[0108]

[0109] S7. Vegetation detection:

[0110] The vegetation remote sensing model is established by using the spectral reflectance of the ground object and biophysical quantities after considering the solar zenith angle and the apparent reflectance of the atmosphere. The vegetation remote sensing model outputs the NDVI vegetation index and image, and obtains the reflectance after the first unified parameter atmospheric correction and the vegetation index after the first unified parameter atmospheric correction.

[0111] Embodiment three:

[0112] A reflectivity correction algorithm for FY3D and MODIS data based on machine learning, including the following steps:

[0113] S1. Get the original data:

[0114] Obtain FY3D and MODIS data, including images, satellite parameters, metadata, spatial coordinate data and latitude;

[0115] S2. Geometric correction:

[0116] Use bilinear interpolation or 2-4 spline curves to interpolate longitude and latitude coordinates to generate coordinate points that correspond to the image one by one, and then use the interpolated coordinate points to perform projection-free geometric correction on the MODIS image;

[0117] S3. Radiation positioning:

[0118] Substitute the satellite parameters into the equation to calculate the surface reflectivity, as shown in Formula 1;

[0119] f=s×(Yc) Formula 1;

[0120] S4, the solar zenith angle is corrected,

[0121] According to formula 2, the effect of different sun altitudes on reflectivity is eliminated.

[0122] F = f × cos(θ) Formula 2;

[0123] S5. Calibration:

[0124] The 6S radiation transfer model is established for atmospheric correction. After running the 6S radiation transfer model, the corrected atmospheric apparent reflectivity is calculated according to Formula 3:

[0125]

[0126] S6. Autonomous learning method model:

[0127] An autonomous learning method model is established, and the feature points matched by the intelligent matching algorithm based on deep learning are used to calculate the matching rate. The intelligent matching algorithm is shown in Formula 4. At this time, M is large, K1 is 1, and the autonomous learning method model has improved its correction accuracy after multiple autonomous learning. The atmospheric correction weakens the red light band signal enhancement caused by atmospheric Rayleigh scattering and aerosol scattering, and the terrain factors make the corrected reflectivity greater than the apparent reflectivity. The corrected reflectivity is closer to the actual situation.

[0128]

[0129] S7. Vegetation detection:

[0130] The vegetation remote sensing model is established by using the spectral reflectance of the ground object and biophysical quantities after considering the solar zenith angle and the apparent reflectance of the atmosphere. The vegetation remote sensing model outputs the NDVI vegetation index and image to obtain the atmospherically corrected reflectance of each pixel and the vegetation index after atmospheric correction of each pixel.

[0131] like Figure 1 and Figure 2 As shown in the figure, atmospheric correction weakens the red band signal enhancement caused by atmospheric Rayleigh scattering and aerosol scattering, while the terrain factor makes the corrected reflectivity greater than the apparent reflectivity, and the corrected reflectivity is closer to the actual situation. The results of atmospheric correction were verified using MODIS surface reflectance products, and the relative errors of unified parameters and pixel-by-pixel atmospheric correction were controlled within 26.9% and 12.7%, respectively. Therefore, the pixel-by-pixel atmospheric correction method is more advantageous for mountainous areas or farmlands with large image ranges or uneven distribution of aerosols and ground elevation.

[0132] Taking vegetation index as an example, the influence of atmospheric correction accuracy on subsequent remote sensing quantitative calculation is demonstrated. Figure 2 ).from Figure 2 It can be seen that the distribution and change trends of the two NDVI curves of the same profile before and after correction are consistent, the peaks and valleys are basically the same, and the pixel-by-pixel NDVI calculated by the single parameter atmospheric correction method is 14.4% higher on average. The reason is that after atmospheric correction, the reflectivity of visible light decreases, the reflectivity of the near-infrared band increases, the contrast of the two bands increases, and the attenuation of light in the process of atmospheric propagation is compensated, so the vegetation index of the image shows an increasing trend.

[0133] The present invention and its embodiments are described above, and such description is not restrictive. What is shown in the full text is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A reflectivity correction algorithm for FY3D and MODIS data based on machine learning, characterized in that: The following steps are involved: S1. Get the original data: Obtain FY3D and MODIS data, including images, satellite parameters, metadata, spatial coordinate data and latitude; S2. Geometric correction: Use bilinear interpolation or 2-4 spline curves to interpolate longitude and latitude coordinates to generate coordinate points that correspond to the image one by one, and then use the interpolated coordinate points to perform projection-free geometric correction on the MODIS image; S3. Radiation positioning: Substitute the satellite parameters into the equation to calculate the surface reflectivity, as shown in Formula 1; f=s×(Yc) Formula 1; S4, the solar zenith angle is corrected, According to formula 2, the effect of different sun altitudes on reflectivity is eliminated. F = f × cos(θ) Formula 2; S5. Calibration: The 6S radiation transfer model is established for atmospheric correction. After running the 6S radiation transfer model, the corrected atmospheric apparent reflectivity is calculated according to Formula 3: S6. Autonomous learning method model: Establish an autonomous learning method model, match the feature points based on the deep learning intelligent matching algorithm, and calculate the matching rate. The intelligent matching algorithm is shown in Formula 4. S7. Vegetation detection: The vegetation remote sensing model is established by using the spectral reflectance of the ground objects and biophysical quantities after considering the solar zenith angle and the apparent reflectance of the atmosphere. The vegetation remote sensing model outputs the NDVI vegetation index and image.

2. The reflectivity correction algorithm for FY3D and MODIS data based on machine learning according to claim 1, characterized in that: The f is the reflectivity, the s is the scaling factor of the reflectivity, the c is the offset parameter of the reflectivity, and the Y is the original recorded value of the satellite data.

3. The reflectivity correction algorithm for FY3D and MODIS data based on machine learning according to claim 1, characterized in that: The F is the corrected reflectivity, and the θ is the solar altitude angle.

4. The reflectivity correction algorithm for FY3D and MODIS data based on machine learning according to claim 1, characterized in that: After the solar zenith angle is corrected, the solar reflection bands 1, 2, 3, 6, 7, 8, and 26 are calibrated to the emissivity, and the thermal radiation bands 29, 31, and 32 are calibrated to the radiant brightness. The radiant brightness is converted into the brightness temperature as shown in Formula 5. Wherein T is the brightness temperature, h is the Planck constant, P is the speed of light in a vacuum, k is the Boltzmann constant, t is the corresponding channel wavelength, and L is the radiation brightness.

5. The reflectivity correction algorithm for FY3D and MODIS data based on machine learning according to claim 1, characterized in that: The F1 is the apparent reflectivity of the atmosphere, the F2 is the equivalent reflectivity of the path radiation term of the atmosphere, the V1 is the solar azimuth, the V2 is the satellite azimuth, the T1 is the downward transmittance of the atmosphere, the T2 is the total upward radiation transmittance, and the S is the albedo of the large balloon surface.

6. The reflectivity correction algorithm for FY3D and MODIS data based on machine learning according to claim 1, characterized in that: The M is a substitution coefficient, the m is the number of correct matching point pairs, the p1 is the actual number of feature points, the P2 is the number of feature points extracted by the autonomous learning method model, the K1 is a correction coefficient, and the M is a matching rate.

7. The reflectivity correction algorithm for FY3D and MODIS data based on machine learning according to claim 6, characterized in that: The value of K1 is 0.8-1.

8. The reflectivity correction algorithm for FY3D and MODIS data based on machine learning according to claim 6, characterized in that: The larger the calculated value of M is, the closer the value of K1 is to 1.