Method for accurately and rapidly measuring organic matters in soil in large range

Abstract

The invention discloses a method for accurately and rapidly measuring soil organic matters in a large range, and belongs to the field of hyperspectral remote sensing inversion of surface parameters. The method comprises the following steps: acquiring soil surface reflectivity data by using a hyperspectral sensor carried by an unmanned aerial vehicle, collecting samples by dividing a 50m * 50m rectangular grid, measuring the content of organic matters, performing 0-2-order fractional order differential calculation and principal component analysis on the hyperspectral data, and extracting the first four principal components to form a feature data set; and establishing a random forest prediction model after normalization, and finally realizing the calculation of the organic matter content of the whole region. The method solves the problems of time and labor consumption, high cost, high risk and the like of traditional measurement, has the advantages of low cost, no damage, high efficiency, strong maneuverability and the like, is high in measurement precision, can realize batch automatic treatment, and is suitable for accurate and rapid measurement of large-area soil organic matters.

Description

Technical Field

[0001] The present invention belongs to the field of hyperspectral remote sensing inversion of surface parameters, and specifically relates to a large-scale, accurate and rapid measurement method for soil organic matter. Background Art

[0002] Traditional methods for measuring soil organic matter primarily include potassium dichromate oxidation, high-temperature calcination, spectrophotometry, and total organic carbon. These traditional methods determine soil organic matter content through chemical oxidation or high-temperature decomposition of organic carbon, combined with titration, weight difference, or colorimetric analysis. However, traditional organic matter measurement methods have a number of drawbacks. For example, organic matter measurement results using traditional methods often rely on empirical conversion coefficients, which can introduce systematic errors. The organic matter measurement process using traditional methods is cumbersome and time-consuming, involving the use of hazardous reagents such as concentrated sulfuric acid and potassium dichromate, posing safety risks. Furthermore, high-temperature calcination can mismeasure CO2 released from carbonates and interfere with the accuracy of organic matter measurements in calcareous soils. Furthermore, traditional methods are time-consuming and costly, making them impractical for large-scale soil organic matter measurement. Summary of the Invention

[0003] Based on the above shortcomings, the present invention provides a large-scale, accurate and rapid measurement method for soil organic matter. It uses drones to obtain high-spectral reflectance data of the soil surface, and establishes a model relationship between the fractional-order differential coefficients of different soil spectral reflectances and soil moisture content to achieve rapid, accurate and large-area measurement of soil organic matter content, which is used to solve the shortcomings of traditional soil organic matter measurement methods such as time-consuming, labor-intensive, high-cost and high-risk.

[0004] The technical solution adopted by the present invention is as follows: a method for accurately and quickly measuring soil organic matter over a large range, comprising the following steps:

[0005] Step 1: Soil sample collection:

[0006] Map software was used to determine the area to be measured, which was then divided into rectangular grids. The center of each grid was used as a sampling point, and the longitude and latitude of the sampling point were recorded. Soil samples 20 cm thick were collected from the surface of the sampling point, and the soil organic matter content of the samples was determined using the potassium dichromate oxidation method.

[0007] Step 2: Surface UAV measurement: Use a drone equipped with a hyperspectral sensor to hover directly above each sampling point to collect hyperspectral reflectance data of the soil surface at that point;

[0008] Step 3: Soil surface spectrum extraction:

[0009] The hyperspectral reflectance data were preprocessed and fractional differential spectra of order 0-2 were calculated with a step size of 0.25. Principal component analysis was performed on the differential spectra of each order, and the first four principal components were extracted to form a soil surface spectral feature dataset.

[0010] Step 4: Establishment of soil organic matter prediction model:

[0011] The spectral feature data set was normalized, and a random forest prediction model was established with the normalized principal component as the independent variable and the measured organic matter content as the dependent variable.

[0012] Step 5: Soil organic matter UAV remote sensing measurement:

[0013] UAVs are used to obtain hyperspectral remote sensing images of the area to be measured, and the spectral characteristics of each pixel in the image are extracted. After normalization, they are input into the random forest model to calculate the soil organic matter content in the entire area.

[0014] Furthermore, the rectangular grid in step 1 is divided as follows: in ArcGIS software, grid lines of 50 m in both horizontal and vertical directions are set for the high spatial resolution remote sensing image, and the area to be measured is divided into a rectangular grid of 50 m × 50 m.

[0015] Furthermore, in step 2, the bandwidth range of the hyperspectral sensor carried by the UAV is 350nm-2500nm, the flight altitude is set to 50m, and it flies in a serpentine trajectory, with the overlap ratio of the lateral and heading images both being 80%.

[0016] Furthermore, the calculation formula for the fractional differential in step 3 is:

[0017]

[0018] Where Rij(λ) is the reflectance data value of the soil sample Pij at the band λ; q is the differential order; h is the band resampling width, h = 10 nm; t and a are the upper and lower limits of the differential, t = 2500 nm, a = 350 nm, and n = (ta) / h = 215.

[0019] Furthermore, the specific operation of the principal component analysis in step 3 is: performing dimensionality reduction on each order differential spectrum and retaining the first four principal components with cumulative contribution rates ≥ 95%.

[0020] Furthermore, the normalization processing formula in step 4 is: GP = PCmax-PCminPC-PCmin, where PC is the original principal component value, PCmin and PCmax are the minimum and maximum values ​​of the component respectively; the normalization result range is [0,1].

[0021] Furthermore, the processing of the hyperspectral remote sensing image in step 5 includes: using RTK positioning information to perform geometric correction, forming a complete regional image through image stitching, and extracting the full-band reflectance data of each pixel from 350nm to 2500nm.

[0022] Furthermore, the above method also includes placing high-reflectivity metal calibration plates at the four corners of the area to be measured for radiation calibration of the UAV hyperspectral data.

[0023] The outstanding beneficial effects and advantages of the present invention are mainly reflected in the establishment of a new large-scale rapid measurement method for soil organic matter content, which obtains soil surface spectral characteristic parameters through unmanned aerial vehicle hyperspectral remote sensing technology combined with fractional-order differential algorithm, thereby realizing rapid, synchronous, and large-scale measurement of soil organic matter. The present invention is used to solve the shortcomings of existing traditional soil organic matter measurement methods, such as time-consuming, labor-intensive, high-risk factors, soil damage, and delayed measurement results, as well as the limitation that soil organic matter measurement results are easily interfered with by environmental factors and human factors. Compared with traditional soil organic matter measurement methods, the organic matter measurement method based on unmanned aerial vehicle hyperspectral remote sensing involved in the present invention has low cost and low loss, can realize accurate, real-time, non-destructive, and efficient large-scale synchronous observation of soil organic matter content over a large area, and has the characteristics of strong maneuverability. In addition, the present invention has very low requirements on the professional knowledge of measurement personnel, and the processes of remote sensing image preprocessing, fractional-order spectral characteristic parameter extraction, principal component analysis, and random forest model establishment can all be software-based through computer programming, so that users can complete batch automated processing operations. Since the hyperspectral reflectance data of the soil surface is directly affected by the soil organic matter content, the fractional-order differential processing increases the spectral characteristic differences of the soil surface with different organic matter contents. Therefore, the parameter relationship of the prediction model is very clear, which ensures the measurement accuracy of the UAV hyperspectral inversion method for soil organic matter content involved in the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 UAV flight path map of hyperspectral data from sampling points. DETAILED DESCRIPTION

[0025] In order to more clearly illustrate the purpose, technical solutions and advantages of the present invention, the following will be combined with the accompanying drawings and specific implementations to further describe the specific embodiments of the present invention. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0026] Example 1

[0027] A method for rapid and accurate measurement of soil organic matter over a large area, the steps are as follows:

[0028] Step 1: The software used includes ArcGIS, map software (Google Maps, Baidu Maps, Tencent Maps, AutoNavi Maps, etc.), and Google Earth Engine platform; the hardware includes a ring cutter, a ziplock bag, a shovel, and a handheld GPS. First, download the high-spatial-resolution satellite optical remote sensing image of the area to be measured on the Google Earth Engine platform. After loading the remote sensing image into the ArcGIS software, determine the rectangular area to be measured in the software, and set a grid line of 50m horizontally and vertically for the rectangular area. Divide the rectangular area to be measured into a regular grid, and divide the area to be monitored into a 50m*50m rectangular grid area with m rows and n columns. The divided rectangular grid area is numbered A. ij , where i is the row number of the grid area after the area to be measured is divided (i = 1, 2, 3, ..., m), and j is the column number of the grid area after the area to be measured is divided (j = 1, 2, 3, ..., n). Determine each rectangular grid area A in ArcGIS software. ij The coordinates of the center position W ij , input the coordinate information of all sampling points into the map software. According to the location coordinates W of each sampling point ij Plan the route in the map software and use GPS to find and determine the actual location of each sampling point in the field according to the route. ij The soil sample is taken from the surface of the sampling point at 20 cm, and the soil sample is placed in a ziplock bag and numbered P. ij Repeat the above steps to collect soil samples from all sampling points. Laboratory measurements of soil organic matter were performed on soil samples from all sampling points using the potassium dichromate oxidation method. Excess potassium dichromate solution was used to dissolve the organic matter in the soil. The potassium dichromate remaining in the solution was titrated with standard ferrous sulfate, and the organic matter content of the sample was finally calculated using the consumed ferrous sulfate. The calculation formula is as follows:

[0029]

[0030] Among them, O ij Represents soil sample P ij The organic matter content, T ij Represents soil sample P ij where c is the mass of the soil sample, v is the volume of the consumed ferrous sulfate solution, and v0 is the volume of the silica solution used to standardize the ferrous sulfate solution.

[0031] Step 2: The software used is map software (Google Maps, Baidu Maps, Tencent Maps, AutoNavi Maps, etc.), and the hardware used includes a hovering multi-rotor drone with an RTK precision positioning module, an airborne hyperspectral imaging spectrometer with a bandwidth range of 350nm-2500nm, and four high-reflectivity metal calibration plates. The four high-reflectivity metal calibration plates are placed at the center point W of the rectangular grid area at the four corners of the monitored area. 11 , W 1n , W m1 and W mn Position. Then, as Figure 1 As shown, on the map software, move the center point of the rectangular grid area in the upper left corner of the area to be measured to the position W 11 As the starting point, the center point W of the rectangular grid area in the lower right corner of the area to be measured is mn As the end point, the coordinates of the center points of all rectangular grid areas determine the serpentine flight trajectory. The UAV is set to fly at an altitude of 50m and flies along the serpentine trajectory. The RTK positioning module is used to set the hovering position point so that the UAV is in each rectangular grid area A. ij The center coordinate W ij Hover just above the position and acquire the hyperspectral image of the position point, perform geometric correction on the hyperspectral image, and calculate the value of the hyperspectral image according to the sampling point coordinate W. ij The soil sample P was further extracted ij The high spectral reflectance of each band in the range of 350nm-2500nm is resampled into 10nm spectral reflectance data of the full spectrum of 350nm-2500nm R ij Repeat this operation to obtain the 10nm hyperspectral reflectance data of the full spectrum of 350nm-2500nm for all sampling points.

[0032] Step 3: Take the soil sample P in the area where soil organic matter is to be measured. ij 10nm resampled spectral reflectance data R ij Perform fractional differential calculation from 0 to 2 with a step size of 0.25 for the entire wavelength range from 350nm to 2500nm to obtain the soil sample P ij The full spectrum spectral reflectance differential data of 0th, 0.25th, 0.5th, 0.75th, 1st, 1.25th, 1.5th, 1.75th and 2nd orders from 350nm to 2500nm, soil sample P ij The calculation formulas for each fractional order of reflectivity are as follows:

[0033]

[0034] Wherein, q is the differential order; h is the band resampling width, and the differential order step size h=10nm in the present invention; t and a are the upper and lower limits of the differential respectively, and in the present invention, t=2500nm and a=350nm.

[0035] The Gamma function calculation formula is as follows:

[0036]

[0037] R ij (λ) is the soil sample P ij The reflectivity data value at the wavelength range of λ. The wavelength space is a = 350nm to t = 2500nm, h = 10nm, and n represents P ij Reflectivity data R of the location point ij The number of bands in the present invention is n=(ta) / h=215. Substituting formula (3) into formula (2) and further expanding it, we can get the soil sample P at band λ. ij The fractional differential calculation expression of is:

[0038]

[0039] According to the above formula, the reflectance data of soil samples at all sampling points are calculated by fractional differential. ij , the result of differential calculation of the 0th order reflectivity data is reflectivity R ij In order to unify the expression with other fractional orders, the 0th order reflectivity differential calculation result is further expressed as R 0-ij , the result of differential calculation of 0.25 order reflectivity data is the reflectivity itself R 0.25-ij , the result of the differential calculation of the 0.5-order reflectivity data is the reflectivity itself R 0.5-ij , the result of differential calculation of 0.75 order reflectivity data is the reflectivity itself R 0.75-ij , the result of the differential calculation of the first-order reflectivity data is the reflectivity itself R 1-ij , the result of the differential calculation of the 1.25-order reflectivity data is the reflectivity itself R 1.25-ij , the result of the 1.5-order reflectivity data differential calculation is the reflectivity itself R 1.5-ij , the result of the differential calculation of the 1.75-order reflectivity data is the reflectivity itself R 1.75-ij , the result of the differential calculation of the second-order reflectivity data is the reflectivity itself R 2-ij For the soil samples at all sampling points, principal component analysis is performed on the full spectrum spectral differential data of 350nm-2500nm at each fractional order, and the first four spectral differential principal components of the principal component transformation results of each fractional order differential data are obtained. ij , the first four principal components of its 0th order spectral differential are PC1 0-ij、PC2 0-ij ,PC3 0-ij 、PC4 0-ij , the first four principal components of the 0.25 order spectral differential are PC1 0.25-ij 、PC2 0.25-ij ,PC3 0.25-ij 、PC4 0.25-ij , the first four principal components of the 0.5-order spectral differential are PC1 0.5-ij 、PC2 0.5-ij ,PC3 0.5-ij 、PC4 0.5-ij , the first four principal components of the 0.75-order spectral differential are PC1 0.75-ij 、PC2 0.75-ij ,PC3 0.75-ij 、PC4 0.75-ij , the first four principal components of the first-order spectral differential are PC1 1-ij 、PC2 1-ij ,PC3 1-ij 、PC4 1-ij , the first four principal components of the 1.25-order spectral differential are PC1 1.25-ij 、PC2 1.25-ij ,PC3 1.25-ij 、PC4 1.25-ij , the first four principal components of its 1.5-order spectral differential are PC1 1.5-ij 、PC2 1.5-ij ,PC3 1.5-ij 、PC4 1.5-ij , the first four principal components of the 1.75-order spectral differential are PC1 1.75-ij 、PC2 1.75-ij ,PC3 1.75-ij 、PC4 1.75-ij , the first four principal components of its second-order spectral differential are PC1 2-ij 、PC2 2-ij ,PC3 2-ij 、PC4 2-ij The first four principal components of each fractional spectral differential of soil samples at all sampling points are saved as the surface spectral feature dataset of all soil samples in the area to be measured.

[0040] Step 4, soil organic matter prediction model establishment: normalize the surface spectral feature data set of soil samples at all sampling points in the measurement area. The normalization process is to use the value of a certain surface spectral feature principal component in all soil samples at a certain fractional order minus the minimum value of this surface spectral feature principal component in all samples at that fractional order, and then divide the calculated difference by the difference between the maximum and minimum values ​​of this surface spectral feature principal component in all samples at that fractional order, so that the normalized result values ​​of each principal component of the surface spectral feature at different fractional orders of all soil samples are between 0 and 1. ij The first four principal components of the 0th order spectral differential normalization are GP1 0-ij GP2 0-ij GP3 0-ij GP4 0-ij , the first four principal components of the 0.25-order spectral differential normalization are GP1 0.25-ij GP2 0.25-ij GP3 0.25-ij GP4 0.25-ij , the first four principal components of the 0.5-order spectral differential normalization are GP1 0.5-ij GP2 0.5-ij GP3 0.5-ij GP4 0.5-ij , the first four principal components of the 0.75-order spectral differential normalization are GP1 0.75-ij GP2 0.75-ij GP3 0.75-ij GP4 0.75-ij , the first four principal components of the first-order spectral differential normalization are GP1 1-ij GP2 1-ij GP3 1-ij GP4 1-ij , the first four principal components of the 1.25-order spectral differential normalization are PC1 1.25-ij 、PC2 1.25-ij ,PC3 1.25-ij 、PC4 1.25-ij , the first four principal components of the 1.5-order spectral differential normalization are PC1 1.5-ij 、PC2 1.5-ij ,PC3 1.5-ij 、PC4 1.5-ij , the first four principal components of the 1.75-order spectral differential normalization are PC1 1.75-ij 、PC2 1.75-ij ,PC3 1.75-ij 、PC4 1.75-ij , the first four principal components of the second-order spectral differential normalization are PC1 2-ij 、PC2 2-ij ,PC3 2-ij 、PC4 2-ijThe normalized first four principal components of each fractional spectral differential were used as independent variables, and the actual measurement results of soil organic matter at all sampling points were used as independent variables. A random forest prediction model for the organic matter content of soil samples in the measured area was established in the MALAB software. ij =RF(PCk s-ij ), where k = 1, 2, 3, 4; s = 0, 0.25, 0.5, 0.75, 1, 1.25, 1.5, 1.75, 2; i = 1, 2, 3, ..., m; j = 1, 2, 3, ..., n. This model can be used to accurately measure soil organic matter content using the normalized first four principal components of different fractional spectral derivatives.

[0041] Step 5. UAV remote sensing measurement of soil organic matter: The software used is mapping software (Google Maps, Baidu Maps, Tencent Maps, AutoNavi Maps, etc.) and ArcGIS software. The hardware used includes a hovering multi-rotor drone equipped with an RTK precision positioning module, an airborne hyperspectral imaging spectrometer with a bandwidth range of 350-2500nm, and four high-reflectivity metal calibration plates. The four high-reflectivity metal calibration plates are placed at the four corners of the entire monitored area, points W1, W2, W3, and W4. Then, on the mapping software, point W1 in the upper left corner of the measured area is used as the starting point, and point W4 in the lower right corner of the measured area is used as the end point. The drone is then used to fly over the entire area of ​​soil organic matter to obtain hyperspectral remote sensing images of the soil surface within the entire monitoring area. The UAV flight process starts from the upper left corner position point W1 of the entire area to be measured and ends at the lower right corner position point W4 of the entire area to be measured. It flies in a serpentine route with a flight altitude of 50 meters. The overlap ratio of the lateral and heading hyperspectral image measurements is 80% to ensure that the collected hyperspectral images cover the entire area to be measured. The RTK position information of the four corner points and the center point of each hyperspectral image is used to correct the geometric distortion. All the hyperspectral images after geometric correction are stitched in ArcGIS software to form a complete hyperspectral image I with M rows and N columns of the entire area to be measured. The full-band spectral reflectance data of the pixel I(x,y) at position (x,y) in the image I is R xy , where x and y represent the row and column numbers of the complete hyperspectral image, x = 1, 2, 3, ..., M; y = 1, 2, 3, ...., N. For the image pixel I(x, y) at position (x, y) in the hyperspectral image, use step 3 to calculate the full-band reflectance data R of the pixel from 350nm to 2500nm. xy Perform spectral differentiation processing of each fractional order, and the differential calculation result of the 0-order reflectivity data is the reflectivity R xy In order to unify the expression with other fractional orders, the 0th order reflectivity differential calculation result is further expressed as R 0-xy, the result of differential calculation of 0.25 order reflectivity data is R 0.25-xy , the result of differential calculation of 0.5-order reflectivity data is R 0.5-xy , the result of differential calculation of 0.75 order reflectivity data is R 0.75-xy , the result of the first-order reflectivity data differential calculation is R 1-xy , the result of differential calculation of 1.25-order reflectivity data is R 1.25-xy , the result of the differential calculation of the 1.5-order reflectivity data is R 1.5-xy , the result of differential calculation of 1.75-order reflectivity data is R 1.75-xy , the result of the differential calculation of the second-order reflectivity data is R 2-xy For any pixel I(x,y) in the hyperspectral image I of the entire area to be measured, perform principal component analysis on the 350nm-2500nm full spectrum spectral differential data of each fractional order of the pixel using step 3, and obtain the first four spectral differential principal components of the principal component transformation result of each fractional order differential data. The first four principal components of the 0-order spectral differential of the pixel I(x,y) at position (x,y) in the hyperspectral image are PC1, 0-xy 、PC2 0-xy ,PC3 0-xy 、PC4 0-xy , the first four principal components of the 0.25 order spectral differential are PC1 0.25-xy 、PC2 0.25-xy ,PC3 0.25-xy 、PC4 0.25-xy , the first four principal components of the 0.5-order spectral differential are PC1 0.5-xy 、PC2 0.5-xy ,PC3 0.5-xy 、PC4 0.5-xy , the first four principal components of the 0.75-order spectral differential are PC1 0.75-xy 、PC2 0.75-xy ,PC3 0.75-xy 、PC4 0.75-xy , the first four principal components of the first-order spectral differential are PC1 1-xy 、PC2 1-xy ,PC3 1-xy 、PC4 1-xy , the first four principal components of the 1.25-order spectral differential are PC1 1.25-xy 、PC2 1.25-xy ,PC3 1.25-xy 、PC4 1.25-xy , the first four principal components of its 1.5-order spectral differential are PC1 1.5-xy 、PC2 1.5-xy ,PC3 1.5-xy 、PC4 1.5-xy , the first four principal components of the 1.75-order spectral differential are PC11.75-xy 、PC2 1.75-xy ,PC3 1.75-xy 、PC4 1.75-xy , the first four principal components of its second-order spectral differential are PC1 2-xy 、PC2 2-xy ,PC3 2-xy 、PC4 2-xy Next, the first four principal components of the spectral differential of each order of the pixel I(x,y) at position (x,y) in the hyperspectral image I are normalized using step 4. The first four principal components of the 0th order spectral differential normalization are GP1 0-xy GP2 0-xy GP3 0-xy GP4 0-xy , the first four principal components of the 0.25-order spectral differential normalization are GP1 0.25-xy GP2 0.25-xy GP3 0.25-xy GP4 0.25-xy , the first four principal components of the 0.5-order spectral differential normalization are GP1 0.5-xy GP2 0.5-xy GP3 0.5-xy GP4 0.5-xy , the first four principal components of the 0.75-order spectral differential normalization are GP1 0.75-xy GP2 0.75-xy GP3 0.75-xy GP4 0.75-xy , the first four principal components of the first-order spectral differential normalization are GP1 1-xy GP2 1-xy GP3 1-xy GP4 1-xy , the first four principal components of the 1.25-order spectral differential normalization are PC1 1.25-xy 、PC2 1.25-xy ,PC3 1.25-xy 、PC4 1.25-xy , the first four principal components of the 1.5-order spectral differential normalization are PC1 1.5-xy 、PC2 1.5-xy ,PC3 1.5-xy 、PC4 1.5-xy , the first four principal components of the 1.75-order spectral differential normalization are PC1 1.75-xy 、PC2 1.75-xy ,PC3 1.75-xy 、PC4 1.75-xy , the first four principal components of the second-order spectral differential normalization are PC1 2-xy 、PC2 2-xy ,PC3 2-xy 、PC4 2-xyThe normalized first four principal components of the fractional order spectral differentials of each pixel are brought into the random forest model established in step 4. xy =RF(PCk s-xy ), where k = 1, 2, 3, 4; s = 0, 0.25, 0.5, 0.75, 1, 1.25, 1.5, 1.75, 2; x = 1, 2, 3, ..., M; y = 1, 2, 3, ..., N, calculate the soil organic matter content value O at the pixel location (x, y) xy , repeat this operation to calculate the soil organic matter content of each pixel of the hyperspectral image of the entire area to be measured, and finally realize large-area rapid measurement of the soil surface organic matter content in the entire area to be measured.

Claims

1. A method for rapid and accurate measurement of soil organic matter over a large area, characterized in that: The following steps are involved: Step 1: Soil sample collection: Map software was used to determine the area to be measured, which was then divided into rectangular grids. The center of each grid was used as a sampling point, and the longitude and latitude of the sampling point were recorded. Soil samples 20 cm thick were collected from the surface of the sampling point, and the soil organic matter content of the samples was determined using the potassium dichromate oxidation method. Step 2: Surface UAV measurement: Use a drone equipped with a hyperspectral sensor to hover directly above each sampling point to collect hyperspectral reflectance data of the soil surface at that point; Step 3: Soil surface spectrum extraction: The hyperspectral reflectance data were preprocessed and fractional differential spectra of order 0-2 were calculated with a step size of 0.

25. Principal component analysis was performed on the differential spectra of each order, and the first four principal components were extracted to form a soil surface spectral feature dataset. Step 4: Establishment of soil organic matter prediction model: The spectral feature data set was normalized, and a random forest prediction model was established with the normalized principal component as the independent variable and the measured organic matter content as the dependent variable. Step 5: Soil organic matter UAV remote sensing measurement: UAVs are used to obtain hyperspectral remote sensing images of the area to be measured, and the spectral characteristics of each pixel in the image are extracted. After normalization, they are input into the random forest model to calculate the soil organic matter content in the entire area.

2. The measuring method according to claim 1, wherein The rectangular grid division method in step 1 is as follows: set grid lines of 50 m in both horizontal and vertical directions for the high spatial resolution remote sensing image in ArcGIS software, and divide the area to be measured into a 50 m × 50 m rectangular grid.

3. The measuring method according to claim 1, wherein In step 2, the bandwidth of the hyperspectral sensor carried by the UAV is 350nm-2500nm, the flight altitude is set to 50m, and it flies in a serpentine trajectory. The overlap ratio of the lateral and heading images is 80%.

4. The measuring method according to claim 1, wherein: The calculation formula for the fractional differential in step 3 is: Where Rij(λ) is the reflectance data value of the soil sample Pij at the band λ; q is the differential order; h is the band resampling width, h = 10 nm; t and a are the upper and lower limits of the differential, t = 2500 nm, a = 350 nm, and n = (ta) / h = 215.

5. The measuring method according to claim 1, wherein: The specific operation of principal component analysis in step 3 is: reduce the dimension of each order differential spectrum and retain the first four principal components with cumulative contribution rate ≥ 95%.

6. The measuring method according to claim 1, characterized in that The normalization formula in step 4 is: GP = PCmax-PCminPC-PCmin Where PC is the original principal component value, PCmin and PCmax are the minimum and maximum values ​​of the component respectively; the normalized result range is [0,1].

7. The measuring method according to claim 1, characterized in that The processing of hyperspectral remote sensing images in step 5 includes: using RTK positioning information for geometric correction, forming a complete regional image through image stitching, and extracting the full-band reflectance data of 350nm-2500nm for each pixel.

8. The measuring method according to any one of claims 1 to 7, characterized in that: The method also includes placing high-reflectivity metal calibration plates at the four corners of the area to be measured for radiation calibration of the UAV hyperspectral data.

Images