Method for improving satellite data inversion soil salinity by using unmanned aerial vehicle data

Abstract

The invention relates to a method for improving satellite data inversion soil salinity by using hyperspectral data of an unmanned aerial vehicle. The method comprises the following steps: collecting a soil sample in a test area, and calculating soil salt content; acquiring and preprocessing satellite remote sensing image data of a test area; storing data, outputting a single-band satellite image, and calculating a spectral index; obtaining and preprocessing hyperspectral image data of the unmanned aerial vehicle in the test area; reconstructing and upscaling unmanned aerial vehicle image data pixels to a spatial resolution of 10m; reconstructing the narrow-band spectral data of the unmanned aerial vehicle into broadband spectral data consistent with a satellite image, storing and outputting a single band, and calculating a spectral index; screening a soil salt content characteristic variable, and constructing a soil salt content inversion model by taking the screened characteristic variable as an independent variable and the actually measured soil salt content as a dependent variable; and fitting a function relationship between the upscaling unmanned aerial vehicle soil salt content inversion model and the satellite inversion model by adopting a numerical regression function method to realize correction of the satellite inversion model.

Description

Technical Field

[0001] The invention relates to a method for inverting soil salinity, in particular to a method for improving satellite data by using unmanned aerial vehicle (UAV) hyperspectral data to achieve high-precision and large-scale inversion of soil salinity. Background Art

[0002] Soil salinization is an important ecological and environmental problem facing arid and semi-arid regions around the world. Ningxia Yinbei has a large area of ​​saline-alkali soil, which is typical in the study of soil salinization in arid regions around the world. The area has a deep soil layer and flat terrain, and is an important reserve land resource in my country. Using scientific methods to accurately invert and monitor the dynamics of regional soil salinization is of great significance to regional food production and sustainable agricultural development.

[0003] Using satellite remote sensing technology to obtain large-scale soil salinity data has become an important means of soil salinization monitoring. However, the low resolution of satellite remote sensing data often leads to spatial heterogeneity problems because multiple features may exist within a single pixel. Soil salinity data collected at a smaller scale cannot be well matched with satellite remote sensing data, resulting in a mismatch in spatial scales.

[0004] UAV remote sensing data has the advantages of high spatial resolution and strong timeliness, and has received extensive attention in the field of remote sensing research and application. Compared with satellite remote sensing data, UAV remote sensing data has less spatial heterogeneity within a pixel, and the spatial resolution of soil salinity data can be well matched with that of UAV remote sensing data. However, UAVs are limited by flight time and altitude, and image coverage is limited, making it difficult to achieve large-scale soil salinization monitoring such as satellite images. It can be seen that a single remote sensing data is difficult to simultaneously meet the needs of large-scale and high-precision soil salinity monitoring.

[0005] In previous studies on satellite and UAV spectral matching, UAV multispectral data was mainly used. However, the wavelength range and central wavelength of UAV multispectral data are often different from those of satellite multispectral data. This difference will cause certain spectral information to be ignored when combining UAV and satellite imagery, thus affecting the inversion accuracy. Hyperspectral sensors can usually capture hundreds of continuous spectral bands with narrow bandwidths, which can be flexibly combined to be consistent with the wavelength range of satellite multispectral bands. Therefore, using UAV hyperspectral data to correct satellite multispectral data can often achieve higher accuracy in the correction process.

[0006] Using UAV remote sensing data to correct satellite remote sensing data for soil salinization monitoring can effectively combine the high precision of UAV monitoring and the wide coverage of satellite monitoring. However, the spatial resolution of UAV images is at the centimeter level, while the spatial resolution of satellite images such as Sentinel-2 satellite images is 10 meters. Due to this huge difference in pixel scale, the number of image elements in two images of the same study area is quite different. If they are directly corrected, there will be a problem of spatial pixel scale mismatch. Therefore, an upscaling method is needed to reconstruct the pixel size so that the pixel size of the two platforms is of the same scale. Summary of the invention

[0007] In order to solve the above problems, the present invention provides a method for improving the inversion of soil salinity from satellite data using drone hyperspectral data. The method of reconstructing pixels by scale conversion is first used to scale the drone data to the same scale as the satellite data. Then, the spectrum conversion method is used to match the spectrum characteristics of the drone and satellite data. Finally, the numerical regression method is used to find the corresponding functional relationship between the two. The present invention uses drone data to correct satellite data, which can improve the inversion accuracy of soil salinity and provide technical support for soil salinization monitoring and control.

[0008] The technical solution adopted by the present invention is: a method for improving satellite data inversion of soil salinity using drone data, comprising the following steps:

[0009] S1. Collect soil samples at sampling points in the test area, measure the conductivity of each soil sample, and calculate the soil salt content based on the conductivity;

[0010] S2. Acquire and preprocess the Sentinel-2 10m spatial resolution satellite remote sensing image data of the test area; save and output single-band satellite images, and calculate the spectral index;

[0011] S3. Obtain and preprocess the UAV hyperspectral image data with a spatial resolution of 0.2m in the test area; reconstruct and scale the pixels of the UAV image data with a spatial resolution of 0.2m to a spatial resolution of 10m using the dominant variation weight method;

[0012] S4, reconstructing the narrow-band spectral data of the UAV into wide-band spectral data consistent with the satellite image, saving and outputting a single band, and calculating the spectral index;

[0013] S5. Use variable projection importance method to screen soil salinity characteristic variables;

[0014] S6. Using the random forest method, the screened characteristic variables were used as independent variables and the measured soil salinity was used as the dependent variable to construct a soil salinity inversion model.

[0015] S7. The numerical regression function method is used to fit the functional relationship between the upscaled UAV soil salinity inversion model and the satellite inversion model to realize the correction of the satellite inversion model.

[0016] Furthermore, step S1 includes the following sub-steps:

[0017] S11. Determine the test area, select representative locations with different soil salinity levels in the test area as sampling points, and collect soil samples at the sampling points;

[0018] S12, measuring the conductivity of each soil sample with a conductivity meter to obtain the soil conductivity of each sampling point;

[0019] S13, calculating the soil salt content at each sampling point according to the measured soil electrical conductivity, wherein the soil salt content is calculated using formula (1);

[0020] y=(x+41.2543)×5÷2120.76 (1)

[0021] Where y is the salt content (g / kg) and x is the conductivity (us / cm).

[0022] Further, step S2 includes the following sub-steps:

[0023] S21. Obtain the satellite remote sensing image data of Sentinel-210m spatial resolution of the test area, perform preprocessing, and save and output the single-band satellite images of the test area: blue band (Blue), green band (Green), red band (Red), and near infrared band (NIR);

[0024] S22, calculate the salinization index S1 according to formula (2), calculate the salinization index S2 according to formula (3), calculate the salinization index S3 according to formula (4), calculate the salt index SI1 according to formula (5), calculate the salt index SI2 according to formula (6), calculate the normalized vegetation index NDVI according to formula (7), calculate the normalized spectral index NDSI according to formula (8), calculate the difference vegetation index DVI according to formula (9), calculate the normalized green band difference vegetation index GNDVI according to formula (10), calculate the renormalized vegetation index RDVI according to formula (11), and save the calculation results in .TIF format, where formulas (2)-(11) are as follows:

[0025] S1=Blue / Red (2)

[0026] S2=(Blue-Red) / (Blue+Red) (3)

[0027] S3= (Green×Red) / Blue (4)

[0028] SI1=(Green ×Red)^0.5 (5)

[0029] SI2= (Green^2+Red^2+NIR^2)^0.5 (6)

[0030] NDVI=(NIR-Red) / (NIR+Red) (7)

[0031] NDSI=(Red -NIR) / (NIR+Red) (8)

[0032] DVI=NIR-Red (9)

[0033] GNDVI=(NIR-Green) / (NIR+Green) (10)

[0034] RDVI=(NIR-Red) / [(NIR+Red)^0.5] (11)

[0035] Further, step S3 includes the following sub-steps:

[0036] S31, obtain the UAV hyperspectral data of the test area with a spatial resolution of 0.2m and perform preprocessing;

[0037] S32. Use the dominant variation weight method to reconstruct the pixels of the drone image data. The calculation formula is as follows:

[0038]

[0039] in,

[0040]

[0041] In the formula, is the upscaled remote sensing information value in the basic geographical unit; y(u) is the remote sensing information value of a certain pixel in the basic geographical unit; w(u) is the weight; y(wd) is the remote sensing information value of the dominant spatial feature pixel; m is the total number of pixels in the basic geographical unit; n is the number of pixels, ranging from 1 to m.

[0042] Further, step S4 includes the following steps:

[0043] S41. Based on the spectral response function of Sentinel-2 10m spatial resolution satellite image, convolution operation is used to reconstruct the narrow-band spectral data of the drone into wide-band spectral data consistent with the satellite image. The calculation formula is as follows:

[0044]

[0045] In the formula, ρs is the reflectance of the UAV hyperspectral data fitted to the satellite image band, s(λ) is the satellite spectral response function, and λ min and λ max are the maximum and minimum wavelengths of the band, ρ(λ) is the reflectance of the original UAV hyperspectral image, and dλ is the spectral resolution of the UAV hyperspectral image.

[0046] S42, save and output single bands: blue band (Blue), green band (Green), red band (Red), near infrared band (NIR);

[0047] S43. Calculate the salinization index S1 according to formula (2), calculate the salinization index S2 according to formula (3), calculate the salinization index S3 according to formula (4), calculate the salt index SI1 according to formula (5), calculate the salt index SI2 according to formula (6), calculate the normalized vegetation index NDVI according to formula (7), calculate the normalized spectral index NDSI according to formula (8), calculate the difference vegetation index DVI according to formula (9), calculate the normalized green band difference vegetation index GNDVI according to formula (10), calculate the renormalized vegetation index RDVI according to formula (11), and save the calculation results in .TIF format.

[0048] Further, step S5 includes the following steps:

[0049] S51, in the ArcGIS platform, extract the index constructed in steps S2 and S4 to the sampling point through multi-value extraction to point, and transfer the table to Excel to obtain the corresponding spectral index attributes of the measured sampling point;

[0050] S52, calculating the variable projection importance of satellite spectral index for soil salinity in the statistical software SIMCA-P, and selecting characteristic variables with importance greater than or equal to 1 to participate in modeling;

[0051] S53. Calculate the variable projection importance of the drone spectral index for soil salinity in SIMCA-P, and select characteristic variables with importance greater than or equal to 1 to participate in modeling;

[0052] Among them, the variable projection importance of spectral index for soil salinity is calculated according to formula (15);

[0053]

[0054] Among them, VIP j is the importance of variable j; p is the number of independent variables; F is the total number of principal components; f is the principal component; SSY f is the sum of squares of variance of the f principal components; SSY total is the sum of squares of the dependent variable; W jf2 is the importance of variable j in principal component f.

[0055] Further, step S6 includes the following steps:

[0056] S61. Arrange the sampling points from high to low according to the salt content, extract data from every 3 sampling points as the validation set, and use the remaining data as the training set.

[0057] S62. In Python software, input the random forest regression algorithm, use the screened characteristic variables as independent variables, and the measured soil salinity as the dependent variable to construct a soil salinity inversion model.

[0058] S63. Use the Bayesian optimizer to continuously optimize the random forest hyperparameters, including the maximum depth of the decision tree (max_depth), the number of decision trees (n_estimators), and the minimum number of internal split samples (min_samples_split), and finally obtain the best model parameter combination.

[0059] Further, step S7 includes the following steps:

[0060] S71. The numerical regression function method is used to fit the upscaled UAV soil salinity inversion data and the satellite inversion data to obtain formula (16).

[0061] y=-0.02x 2 +1.090x+0.297 (16)

[0062] S72, substituting the satellite soil salinity inversion model as an independent variable into formula (16) to obtain a calibrated model and salinity estimation value;

[0063] S73, Satellite inversion model R after correction by UAV data 2 It increased from 0.630 to 0.787, and the RMSE decreased from 2.696 g / kg to 2.043 g / kg;

[0064] S74, using ArcGIS Fishnet tool to segment the test area image to obtain multiple segmented images;

[0065] S75, using the raster to point tool in ArcGIS to convert the segmented images into point shp data in turn according to the pixel size;

[0066] S76, loading the spectral index calculated and screened in steps S4 and S53 in ArcGIS, and using the Extract Multi-Value to Point tool to assign the spectral index attribute to the sampling point;

[0067] S77. Use the table-to-Excel tool in ArcGIS to export all pixel attribute tables of each segmented image;

[0068] S78. In Python software, the corrected satellite soil salinity inversion model is used to invert the salt information of all pixel points in the test area;

[0069] S79, in ArcGIS, connect all pixel point salinity data obtained by the inversion to the point shp data obtained in step S76 by using the attribute tool, and then use the ArcGIS point to raster tool to obtain the soil salinity of the test area, and set the value field to the inverted salinity;

[0070] S710. Use the ArcGIS mosaic to new raster tool to merge the block data of the experimental area, and classify them according to the soil salinization level to obtain the salinization level distribution map of the experimental area.

[0071] The beneficial effects of the present invention are:

[0072] (1) The method of using UAV data to improve satellite data to invert soil salinity proposed in the present invention can effectively solve the problem of low accuracy of the soil salinity inversion model using single remote sensing data. Combining the high precision of UAV data and the large-scale inversion advantages of satellite data, it can accurately extract the salt content of saline soil in a wide range of arid, semi-arid and coastal areas.

[0073] (2) The UAV pixel reconstruction method proposed in the present invention uses the dominant variation weight method to resample and upscale the 0.2m resolution UAV data to the same scale as the 10m resolution satellite data. Compared with the traditional pixel reconstruction method, it not only performs weighted processing, but also fully considers the heterogeneity of surrounding pixels, and can identify and adapt to the degree of variation in different regions. Therefore, when extracting the image spectral reflectance, the result more objectively reflects the actual salinization situation of the surface in the test area, which helps to improve the accuracy of the model inversion.

[0074] (3) The present invention proposes to use the satellite spectral response function to reconstruct the UAV hyperspectral band into a multispectral band consistent with the Sentinel-2 multispectral band. Hyperspectral sensors can usually capture hundreds of narrow bandwidth continuous spectral bands, which can be flexibly combined and can be consistent with the wavelength range of satellite multispectral bands.

[0075] (4) The present invention not only calculates the salinity index that can represent soil salinity, but also calculates five vegetation indices (NDVI, NDSI, DVI, GNDVI, RDVI). These vegetation indices effectively distinguish the effects of soil and vegetation, and enhance the adaptability of the model to different land cover types and vegetation conditions. This method helps to improve the performance of the model in complex environments, especially in areas with large changes in vegetation cover.

[0076] (5) The present invention uses the VIP method to screen soil salt-sensitive variables, which can effectively solve the problem of difficulty in extracting spectral information of salinized soil, effectively characterize the degree of soil salinization, and provide stable basic data guarantee for accurately monitoring the dynamic changes of soil salinization.

[0077] (6) The present invention uses a numerical regression function method to fit UAV data and satellite data. The quadratic polynomial fitting function constructed by the numerical regression function method can better reflect the complex relationship between satellite and aircraft data.

[0078] (7) The present invention inverts the salt content of the sampling points and divides the salt content into five categories: non-salinization, mild salinization, moderate salinization, severe salinization, and saline soil, which is representative. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 A flow chart of a method for using drone data to improve satellite data for inverting soil salinity provided in an embodiment of the present invention.

[0080] FIG. 2( a ) is a schematic diagram of the screening results of sensitive variables of soil salinity in drone data provided by an embodiment of the present invention.

[0081] FIG. 2( b ) is a schematic diagram of the screening results of sensitive variables of soil salinity in satellite remote sensing data provided by an embodiment of the present invention.

[0082] FIG3( a ) shows the inversion accuracy of the satellite soil salinity inversion model before correction provided by an embodiment of the present invention.

[0083] FIG3( b ) shows the inversion accuracy of the satellite soil salinity inversion model after correction provided by an embodiment of the present invention.

[0084] Figure 4 A schematic diagram of the distribution of different degrees of soil salinization in the test area provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0085] Now, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the accompanying drawings are only exemplary and are intended to explain the principles and spirit of the present invention, rather than to limit the scope of the present invention.

[0086] The embodiment of the present invention provides a method for improving the soil salinity inversion of satellite data using drone data, such as Figure 1 As shown, the following steps are included:

[0087] S1. Collect soil samples at sampling points in the test area, measure the conductivity of the soil samples, and calculate the soil salt content according to the conductivity method.

[0088] Step S1 includes the following sub-steps:

[0089] S11. Determine the test area, select representative locations with different soil salinity levels in the core test area to set up sampling points, and collect soil samples at the sampling points.

[0090] Using the five-point sampling method, soil samples from five sampling points near the positioning point are taken and placed in sampling bags for mixing as the measured data of the sampling points.

[0091] In the embodiment of the present invention, the distribution of sampling points is based on the team's previous research, and sampling points are selected in five different distribution areas of non-salinization, mild salinization, moderate salinization, severe salinization, and saline soil, and a total of 79 soil samples are selected.

[0092] The soil salinity levels are shown in Table 1.

[0093] Table 1

[0094] S12. Use a conductivity meter to measure the conductivity of each soil sample to obtain the soil conductivity of each sampling point.

[0095] The selected 79 soil samples were naturally air-dried and ground, then filtered through a 2 mm pore size sieve, and the extract was prepared with a water-soil ratio of 5:1. Finally, the extract was measured with a conductivity meter to obtain the conductivity of the soil samples.

[0096] S13. Calculate the soil salinity at each sampling point based on the measured soil electrical conductivity.

[0097] Specifically, the soil salinity is calculated using formula (1).

[0098] y=(x+41.2543)×5÷2120.76 (1)

[0099] Where y is the salt content (g / kg) and x is the conductivity (us / cm).

[0100] S2. Acquire and preprocess the Sentinel-2 10m spatial resolution satellite remote sensing image data of the test area; save and output the single-band satellite image data, and calculate the spectral index.

[0101] Step S2 includes the following sub-steps:

[0102] S21. In the remote sensing image processing platform ENVI (The Environment for Visualizing Images), obtain the Sentinel-2 10m spatial resolution satellite remote sensing image data of the test area, perform preprocessing such as radiation calibration, atmospheric correction, cropping and mosaicking, and save and output single-band satellite images of the test area: blue band (Blue), green band (Green), red band (Red) and near infrared band (NIR).

[0103] S22. Calculate the salinization index S1 according to formula (2), calculate the salinization index S2 according to formula (3), calculate the salinization index S3 according to formula (4), calculate the salt index SI1 according to formula (5), calculate the salt index SI2 according to formula (6), calculate the normalized vegetation index NDVI according to formula (7), calculate the normalized spectral index NDSI according to formula (8), calculate the difference vegetation index DVI according to formula (9), calculate the normalized green band difference vegetation index GNDVI according to formula (10), calculate the renormalized vegetation index RDVI according to formula (11), and save the calculation results in .TIF format.

[0104] S1=Blue / Red (2)

[0105] S2=(Blue-Red) / (Blue+Red) (3)

[0106] S3= (Green×Red) / Blue (4)

[0107] SI1=(Green ×Red)^0.5 (5)

[0108] SI2=(Green^2+Red^2+NIR^2)^0.5 (6)

[0109] NDVI=(NIR-Red) / (NIR+Red) (7)

[0110] NDSI=(Red -NIR) / (NIR+Red) (8)

[0111] DVI=NIR-Red (9)

[0112] GNDVI=(NIR-Green) / (NIR+Green) (10)

[0113] RDVI=(NIR-Red) / [(NIR+Red)^0.5] (11)

[0114] In the embodiment of the present invention, in order to maximize the inversion accuracy of soil salinity, the construction of the spectral index refers to the same research literature, and five salt indices that are more sensitive to soil salinity are selected, namely: S1, S2, S3, SI1, and SI2. Considering that there is a small amount of vegetation coverage in the test area, five vegetation indices are also constructed: NDVI, NDSI, DVI, GNDVI, and RDVI.

[0115] The specific documents are as follows:

[0116] [Document 1] Abbas, A.; Khan, S. Using remote sensing techniques for appraisal of irrigated soil salinity. In Proceedings of the Advances and Applications for Management and Decision Making Land, Water and Environmental Management: Integrated Systems for Sustainability MODSIM07; Modelling and Simulation Society of Australia and New Zealand: Canberra, Australia, 2007; pp. 2632–2638.

[0117] [Literature 2] Douaoui, AEK; Nicolas, H.; Walter, C. Detecting salinity hazardswithin a semiarid context by means of combining soil and remote-sensingdata.Geoderma.2006,134,pp.217–230.

[0118] [Literature 3] Zhang, TT; Zeng, SL; Gao, Y.; Ouyang, ZT; Li, B.; Fang, CM; Zhao, B. Using hyperspectral vegetation indices as a proxy to monitor soilsalinity.Ecol.Indic.2011,11,pp.1552-1562.

[0119] [Document 4] Allbed, A.; Kumar, L.; Aldakheel, YY

[0120] [Literature 5] Dong, C.; Zhao, GX; Su, BW; Chen, XN; Zhang, SMResearch on the Decision Model of Variable Nitrogen Application in Winter Wheat's GreeningPeriod Based on UAV Multispectral Image.Spectrosc.Spect.Anal.2019,39,pp.3599–3605.

[0121] [Literature 6] Dong, JJ; Niu, JM; Zhang, Q.; Kang Sa, RL; Han, F.Remote sensing yield estimation of typical grassland based on multi-source satellitedata.Chin.J.Grassl.2013,35,pp.64–69.

[0122] S3. Obtain and preprocess the UAV hyperspectral image data with a spatial resolution of 0.2 m in the test area; reconstruct and upscale the pixels of the UAV image data with a spatial resolution of 0.2 m to a spatial resolution of 10 m using the dominant variation weight method.

[0123] Step S3 includes the following sub-steps:

[0124] S31. Obtain the UAV hyperspectral data of the test area with a spatial resolution of 0.2m, and perform pre-processing such as radiation calibration, atmospheric correction, splicing and cropping;

[0125] S32. Use the dominant variation weight method to reconstruct the pixels of the drone image data and resample the pixel size to 10m. The calculation formula is as follows:

[0126]

[0127] in,

[0128]

[0129] In the formula, is the upscaled remote sensing information value in the basic geographical unit; y(u) is the remote sensing information value of a certain pixel in the basic geographical unit; w(u) is the weight; y(wd) is the remote sensing information value of the dominant spatial feature pixel; m is the total number of pixels in the basic geographical unit; n is the number of pixels, ranging from 1 to m.

[0130] S4. Reconstruct the narrow-band spectral data of the UAV into wide-band spectral data consistent with the satellite image, save and output a single band, and calculate the spectral index.

[0131] Step S4 includes the following sub-steps:

[0132] S41. Based on the spectral response function of the Sentinel-210m spatial resolution satellite image, the convolution operation is used to reconstruct the narrow-band spectral data of the drone into wide-band spectral data consistent with the satellite image. The calculation formula is as follows:

[0133]

[0134] In the formula, ρ s is the reflectance of the UAV hyperspectral band fitted to the satellite image band, s(λ) is the satellite spectral response function, and λ min and λ max are the maximum and minimum wavelengths of the band, ρ(λ) is the reflectance of the original UAV hyperspectral image, and dλ is the spectral resolution of the UAV hyperspectral image.

[0135] S42, save and output single bands: blue band (Blue), green band (Green), red band (Red), near infrared band (NIR);

[0136] S43. Calculate the salinization index S1 according to formula (2), calculate the salinization index S2 according to formula (3), calculate the salinization index S3 according to formula (4), calculate the salt index SI1 according to formula (5), calculate the salt index SI2 according to formula (6), calculate the normalized vegetation index NDVI according to formula (7), calculate the normalized spectral index NDSI according to formula (8), calculate the difference vegetation index DVI according to formula (9), calculate the normalized green band difference vegetation index GNDVI according to formula (10), calculate the renormalized vegetation index RDVI according to formula (11), and save the calculation results in .TIF format.

[0137] S5. Use the variable projection importance method to screen the characteristic variables of soil salinity.

[0138] Step S5 includes the following sub-steps:

[0139] S51. In the ArcGIS platform, the spectral index constructed in steps S2 and S4 is extracted to the sampling points using the Multi-Value Extraction to Points tool, and then the table to Excel tool is used to export the measured point attributes to Excel to obtain the corresponding spectral index attributes of the measured sampling points.

[0140] S52. In the statistical software SIMCA-P, calculate the variable projection importance of the satellite spectral index for soil salinity, and select characteristic variables with importance greater than or equal to 1 to participate in modeling.

[0141] S53. Calculate the variable projection importance of the drone spectral index for soil salinity in SIMCA-P, and select characteristic variables with importance greater than or equal to 1 to participate in modeling.

[0142] Among them, the variable projection importance of the spectral index for soil salinity is calculated according to formula (15).

[0143]

[0144] Among them, VIP j is the importance of variable j; p is the number of independent variables; F is the total number of principal components; f is the principal component; SSY f is the sum of squares of variance of the f principal components; SSY total is the sum of squares of the dependent variable; W jf 2 is the importance of variable j in principal component f.

[0145] In this embodiment, 7 and 6 variables with importance greater than or equal to 1 are selected based on the upscaled UAV data and satellite data respectively for modeling. As shown in Figure 2(a), in the screening results of soil salinity sensitive variables, the satellite spectral indexes include: S1, S2, NDVI, NDSI, DVI, and RDVI; as shown in Figure 2(b), in the screening results of soil salinity sensitive variables, the UAV spectral indexes include: S1, S2, S3, SI1, NDVI, NDSI, and GNDVI.

[0146] S6. Using the random forest regression method, the screened characteristic variables were used as independent variables and the measured soil salinity as the dependent variable to construct a soil salinity inversion model.

[0147] Step S6 includes the following sub-steps:

[0148] S61. Arrange the sampling points from high to low according to the salt content, extract data from every 3 sampling points as the validation set, and use the remaining data as the training set.

[0149] S62. In Python software, the random forest regression algorithm was used to construct a soil salinity inversion model with the screened characteristic variables as independent variables and the measured soil salinity as the dependent variable.

[0150] S63. Use the Bayesian optimizer to continuously optimize the random forest hyperparameters, including the maximum depth of the decision tree (max_depth), the number of decision trees (n_estimators), and the minimum number of internal split samples (min_samples_split), and finally obtain the best model parameter combination. Among them, the optimal hyperparameter combination for the drone model is (10, 33, 9), and the optimal hyperparameter combination for the satellite model is (10, 31, 11).

[0151] S7. The numerical regression function method is used to fit the functional relationship between the upscaled UAV soil salinity inversion model and the satellite inversion model to achieve the correction of the satellite model.

[0152] Step S7 includes the following sub-steps:

[0153] S71. The numerical regression function method is used to fit the upscaled UAV soil salinity inversion data and the satellite inversion data to obtain formula (16).

[0154] y=-0.02x 2 +1.090x+0.297 (16)

[0155] S72. Substitute the satellite soil salinity inversion model as an independent variable into formula (16) to obtain the calibrated model and salinity estimation value.

[0156] S73, Satellite inversion model R after correction by UAV data 2 It increased from 0.630 to 0.787, and RMSE decreased from 2.696 g / kg to 2.043 g / kg. As shown in Figure 3, Figure 3(a) shows the satellite image inversion value before correction, where R2 is 0.630 and RMSE is 2.696 g / kg, and Figure 3(b) shows the satellite image inversion value after correction, where R2 is 0.787 and RMSE is 2.043 g / kg.

[0157] S74. Since the number of pixels output exceeds the maximum output value specified by ArcGIS, the ArcGIS fishnet tool is used to segment the image to obtain multiple segmented images.

[0158] S75. Use the raster to point tool in ArcGIS to convert the segmented images into point shp data according to the pixel size.

[0159] S76. Load steps S4 and S53 in ArcGIS, calculate and filter the spectral index, and use the Extract Multivalues ​​to Points tool to assign the spectral index attributes to the sampling points.

[0160] S77. Use the table-to-Excel tool in ArcGIS to export all pixel attribute tables for each segmented image.

[0161] S78. In Python software, the corrected satellite soil salinity inversion model is used to invert the salt information of all pixel points in the test area.

[0162] S79. In ArcGIS, connect the inverted pixel point salinity to the attribute and then convert the point to raster. The value field is the inverted salinity, and the soil salinity of the test area is obtained.

[0163] S710, using ArcGIS to mosaic the block pixels into a new grid, and reclassify the soil according to the soil salt content level shown in Table 1 to obtain the salinization level distribution map of the test area, as shown in Figure 4 shown.

[0164] In the embodiment of the present invention, the drone image data is resampled and upscaled to the same scale as the satellite image data, and then the numerical regression method is used to fit the two, so as to use the drone data to correct the satellite data to improve its inversion accuracy. The results show that the corrected satellite salinity inversion model R 2 From 0.630 to 0.787. Secondly, the soil salinity distribution in the test area was explored by inverting the soil salinity at the scale of each satellite pixel in the test area, which is representative and feasible.

[0165] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for improving satellite data inversion of soil salinity using drone data, characterized in that: The following steps are involved: S1. Collect soil samples at sampling points in the test area, measure the conductivity of the soil samples, and calculate the soil salt content according to the conductivity method; S2. Acquire and preprocess the Sentinel-2 10m spatial resolution satellite remote sensing image data of the test area; save and output single-band satellite images, and calculate the spectral index; S3. Obtain and preprocess the UAV hyperspectral image data with a spatial resolution of 0.2m in the test area; reconstruct and scale the pixels of the UAV image data with a spatial resolution of 0.2m to a spatial resolution of 10m using the dominant variation weight method; S4, reconstructing the narrow-band spectral data of the UAV into wide-band spectral data consistent with the satellite image, saving and outputting a single band, and calculating the spectral index; S5. Use variable projection importance method to screen soil salinity characteristic variables; S6. Using the random forest method, the screened characteristic variables were used as independent variables and the measured soil salinity was used as the dependent variable to construct a soil salinity inversion model. S7. The numerical regression function method is used to fit the functional relationship between the upscaled UAV soil salinity inversion model and the satellite inversion model to realize the correction of the satellite inversion model.

2. The method for inverting soil salinity by using drone data to improve satellite data according to claim 1, characterized in that: Step S1 includes the following sub-steps: S11. Determine the test area, select representative locations with different soil salinity levels in the test area as sampling points, and collect soil samples at the sampling points; S12, measuring the conductivity of each soil sample with a conductivity meter to obtain the soil conductivity of each sampling point; S13, calculating the soil salt content at each sampling point according to the measured soil electrical conductivity, wherein the soil salt content is calculated using formula (1); y=(x+41.2543)×5÷2120.76 (1) Where y is the salt content (g / kg) and x is the conductivity (us / cm).

3. The method for inverting soil salinity by using drone data to improve satellite data according to claim 1, characterized in that: Step S2 includes the following sub-steps: S21. Acquire the Sentinel-2 10m spatial resolution satellite remote sensing image data of the test area, perform preprocessing, and save and output the single-band satellite images of the test area: the blue band (Blue), green band (Green), red band (Red) and near infrared band (NIR) are single-band outputs; S22, calculate the salinization index S1 according to formula (2), calculate the salinization index S2 according to formula (3), calculate the salinization index S3 according to formula (4), calculate the salt index SI1 according to formula (5), calculate the salt index SI2 according to formula (6), calculate the normalized vegetation index NDVI according to formula (7), calculate the normalized spectral index NDSI according to formula (8), calculate the difference vegetation index DVI according to formula (9), calculate the normalized green band difference vegetation index GNDVI according to formula (10), calculate the renormalized vegetation index RDVI according to formula (11), and save the calculation results in .TIF format, where formulas (2)-(11) are as follows: S1=Blue / Red (2) S2=(Blue-Red) / (Blue+Red) (3) S3= (Green×Red) / Blue (4) SI1=(Green×Red)^0.5 (5) SI2= (Green^2+Red^2+NIR^2)^0.5 (6) NDVI=(NIR-Red) / (NIR+Red) (7) NDSI=(Red -NIR) / (NIR+Red) (8) DVI=NIR-Red (9) GNDVI=(NIR-Green) / (NIR+Green) (10) RDVI=(NIR-Red) / [(NIR+Red)^0.5] (11).

4. The method for inverting soil salinity by using drone data to improve satellite data according to claim 1, characterized in that: Step S3 includes the following sub-steps: S31, obtain the UAV data with 0.2m spatial resolution in the test area and perform preprocessing; S32. Use the dominant variation weight method to reconstruct the pixels of the drone image data and resample the pixel size to 10m. The calculation formula is as follows: in, In the formula, is the upscaled remote sensing information value in the basic geographical unit; y(u) is the remote sensing information value of a certain pixel in the basic geographical unit; w(u) is the weight; y(wd) is the remote sensing information value of the dominant spatial feature pixel; m is the total number of pixels in the basic geographical unit; n is the number of pixels, ranging from 1 to m.

5. The method for inverting soil salinity by using drone data to improve satellite data according to claim 2, characterized in that: Step S4 includes the following sub-steps: S41. Based on the spectral response function of Sentinel-2 10m spatial resolution satellite image, convolution operation is used to reconstruct the narrow-band spectral data of the drone into wide-band spectral data consistent with the satellite image. The calculation formula is as follows: In the formula, ρ s is the reflectance of the UAV hyperspectral data fitted to the satellite image band, s(λ) is the satellite spectral response function, and λ min and λ max are the maximum and minimum wavelengths of the band, ρ(λ) is the reflectance of the original UAV hyperspectral image, and dλ is the spectral resolution of the UAV hyperspectral image. S42, save and output single bands: blue band (Blue), green band (Green), red band (Red), near infrared band (NIR); S43. Calculate the salinization index S1 according to formula (2), calculate the salinization index S2 according to formula (3), calculate the salinization index S3 according to formula (4), calculate the salt index SI1 according to formula (5), calculate the salt index SI2 according to formula (6), calculate the normalized vegetation index NDVI according to formula (7), calculate the normalized spectral index NDSI according to formula (8), calculate the difference vegetation index DVI according to formula (9), calculate the normalized green band difference vegetation index GNDVI according to formula (10), calculate the renormalized vegetation index RDVI according to formula (11), and save the calculation results in .TIF format.

6. The method for inverting soil salinity by using drone data to improve satellite data according to claim 1, characterized in that: Step S5 The following steps are included: S51, in the ArcGIS platform, extract the spectral index constructed in steps S2 and S4 to the sampling point through multi-value extraction to point, and transfer the table to Excel to obtain the corresponding spectral index attributes of the measured sampling point; S52. In the statistical software SIMCA-P, the variable projection importance of the satellite spectral index for soil salinity is calculated, and characteristic variables with importance greater than or equal to 1 are selected to participate in modeling; S53. In SIMCA-P, calculate the variable projection importance of the drone spectral index for soil salinity, and select characteristic variables with importance greater than or equal to 1 to participate in modeling; Among them, the variable projection importance of spectral index for soil salinity is calculated according to formula (15): Among them, VIP j is the importance of variable j; p is the number of independent variables; F is the total number of principal components; f is the principal component; SSY f is the sum of squares of variance of the f principal components; SSY total is the sum of squares of the dependent variable; W jf 2 is the importance of variable j in principal component f.

7. The method for inverting soil salinity by using drone data to improve satellite data according to claim 6, characterized in that: According to step S52, 6 characteristic variables are screened out for modeling, namely: S1, S2, NDVI, NDSI, DVI, and RDVI; according to step S53, 7 characteristic variables are screened out for modeling, namely: S1, S2, S3, SI1, NDVI, NDSI, and GNDVI.

8. The method for inverting soil salinity by using drone data to improve satellite data according to claim 1, characterized in that: Step S6 includes the following sub-steps: S61, arranging the sampling points from high to low according to the salt content, extracting data from every 3 sampling points as a validation set, and the remaining data as a training set; S62. In Python software, a random forest regression algorithm was used to construct a soil salinity inversion model with the screened characteristic variables as independent variables and the measured soil salinity as the dependent variable. S63. Use the Bayesian optimizer to continuously optimize the random forest hyperparameters, including the maximum depth of the decision tree (max_depth), the number of decision trees (n_estimators), and the minimum number of internal split samples (min_samples_split), and finally obtain the best model parameter combination.

9. The method for inverting soil salinity by using drone data to improve satellite data according to claim 8, characterized in that: The optimal hyperparameter combination of the drone model in step S63 is (10, 33, 9), and the optimal hyperparameter combination of the satellite model is (10, 31, 11).

10. The method for inverting soil salinity by using drone data to improve satellite data according to claim 1, characterized in that: Step S7 includes the following sub-steps: S71. Using the numerical regression function method, the upscaled UAV soil salinity inversion data and the satellite soil salinity inversion data are fitted to obtain formula (16); y=-0.02x 2 +1.090x+0.297 (16) S72, substituting the satellite soil salinity inversion model as an independent variable into formula (16) to obtain a calibrated model and salinity estimation value; S73, Satellite inversion model R after correction by UAV data 2 It increased from 0.630 to 0.787, and the RMSE decreased from 2.696 g / kg to 2.043 g / kg; S74, using ArcGIS Fishnet tool to segment the test area image to obtain multiple segmented images; S75, using the raster to point tool in ArcGIS to convert the segmented images into point shp data in turn according to the pixel size; S76, loading the spectral index calculated and screened in steps S4 and S53 in ArcGIS, and using the Extract Multi-Value to Point tool to assign the spectral index attribute to the sampling point; S77. Use the table-to-Excel tool in ArcGIS to export all pixel attribute tables of each segmented image; S78. In Python software, the corrected satellite soil salinity inversion model is used to invert the salt information of all pixel points in the test area.

11. The method for inverting soil salinity by using drone data to improve satellite data according to claim 10, characterized in that: Further comprising the following steps: S79, in ArcGIS, connect all pixel point salinity data obtained by the inversion to the point shp data obtained in step S76 by using the attribute tool, and then use the ArcGIS point to raster tool to obtain the soil salinity of the test area, and set the value field to the inverted salinity; S710. Use the ArcGIS mosaic to new raster tool to merge the test area block data into grids, and classify them according to the soil salinization level to obtain a salinization level distribution map of the test area.

Images