A method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from unmanned aerial vehicles

Abstract

The present invention discloses a method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from unmanned aerial vehicles (UAVs), and belongs to the field of remote sensing measurement. The present invention utilizes UAVs to obtain remote sensing images of the surface of saline-alkali soil at different heights, utilizes the phenomenon of water loss, shrinkage, and cracking of the surface of viscous saline-alkali soil, and extracts characteristic parameters of soil surface cracks as soil indexes in combination with optical images. The diagnostic spectral characteristic reflectance of the main salt minerals in the saline-alkali soil and the mathematical transformation results of the reflectance are used as soil salt content response factors. According to an artificial neural network model, a quantitative relationship between the response factor and the soil electrical conductivity is established, thereby realizing accurate measurement of the electrical conductivity of saline-alkali soil using UAV low-altitude remote sensing image data. The method involved in the present invention has low cost, low cost, low loss, can realize real-time and synchronous observation of soil electrical conductivity over a large area, and has the advantage of strong maneuverability.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing measurement, and in particular relates to a method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from unmanned aerial vehicles (UAVs). Background Art

[0002] Rapid and accurate measurement of soil salt content and its distribution is of great practical significance for optimizing soil management practices, promoting rational use and improving soil condition, ensuring food security, and enhancing the ecological environment. Soil electrical conductivity numerically represents the electrical conductivity of a soil solution—that is, the soil's ability to conduct electric current. Soil electrical conductivity is generally considered an accurate quantitative indicator of soil salinity and is one of the international assessment standards for soil salinization. This is because when soil moisture remains constant, salt minerals are considered strong electrolytes in the soil. Dissolved salts in the soil solution exist as ions, all of which are conductive. An increase in the salt mineral content increases the osmotic pressure of the soil solution, leading to an increase in electrical conductivity. Therefore, electrical conductivity can be used to directly reflect soil salinity.

[0003] Currently, soil conductivity measurement methods are primarily categorized into contact and non-contact methods. Contact soil conductivity measurement methods primarily rely on electrode sensors, which require a voltmeter, electrodes, a constant current power supply, and soil to form a circuit for conductivity measurement. However, the electrode method requires digging pits or drilling holes in the soil to bury the salt sensor or electrodes used for measurement at varying depths. Furthermore, good contact between the instrument probe and the soil must be ensured during measurement. This method is often time-consuming and labor-intensive, and is significantly affected by soil moisture content and meteorological conditions. The instrument is also difficult to use and prone to operational errors. Non-contact conductivity measurement is primarily achieved using a geodesic conductivity meter. Based on electromagnetic field theory, this instrument measures and determines the relative relationship between primary and secondary magnetic fields to determine soil conductivity. However, in practical applications, the instrument itself is expensive, and its large size makes it difficult to carry and operate. Furthermore, the accuracy of conductivity measurements from geodetic conductivity meters is easily affected by numerous factors, including soil physical properties such as soil texture, moisture, and temperature. It is also susceptible to interference from the measurement environment, such as air temperature and pressure. Furthermore, conductivity measurements from geodetic conductivity meters often provide a general representation of the salinity of the soil profile at a specific depth, making them poorly sensitive to the commonly used salt content of the soil surface. Summary of the Invention

[0004] Based on the above shortcomings, the present invention provides a method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from drones, which solves the shortcomings of the hysteresis of existing traditional laboratory measurements of soil conductivity and the limitations of the conductivity method for measuring soil conductivity, which is easily affected by environmental factors and human factors during measurement. It realizes the accurate measurement of the electrical conductivity of saline-alkali soil using low-altitude remote sensing image data from drones.

[0005] The technology used in this invention is as follows: a method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from unmanned aerial vehicles, the steps of which are as follows:

[0006] Step 1: Acquisition of UAV remote sensing images

[0007] The equipment used included a multi-rotor drone, a handheld GPS, and a rectangular metal calibration frame with an inner diameter of 1m×1m. The drone was equipped with a high-definition CCD lens and a hyperspectral imaging spectrometer. The rectangular study area was divided into four sub-areas for capturing high-definition remote sensing images of the soil surface using the drone's CCD and hyperspectral remote sensing images. Sampling points were selected for soil sample collection, and the true value of the electrical conductivity of each soil sample was measured. All conductivity measurement results were saved as a conductivity dataset E, which served as the modeling sample set for the soil conductivity prediction model for the entire study area.

[0008] Step 2: Preprocessing of UAV remote sensing images

[0009] The CCD high-definition remote sensing images taken in the four rectangular sub-areas are mosaicked, geometrically corrected, and cropped to generate a rectangular CCD high-definition remote sensing image map Z1 corresponding to the entire rectangular study area; the hyperspectral remote sensing images taken in the four rectangular sub-areas are mosaicked, geometrically corrected, and cropped to generate a rectangular hyperspectral remote sensing image map Z2 corresponding to the entire rectangular study area;

[0010] Step 3: Extract spectral characteristic parameters

[0011] According to the location information of the sampling point, the full-band hyperspectral reflectance data of the sampling point is found, and on this basis, the characteristic spectral band of the sample point is calculated, and the first-order derivative, second-order derivative, logarithm, reciprocal and square root mathematical transformation parameters of the characteristic band spectral reflectance data are further calculated as spectral characteristic parameters to form a spectral reflectance characteristic parameter data set of the soil sample point; Step 4: Extract soil surface characteristic parameters

[0012] According to the location information of the sampling points, the number of image pixel rows and columns corresponding to a 1m×1m surface area is determined according to the scale. With each sample point as the center, the image size of the soil sample surface is determined according to the size of the surface area. All sample point data are cropped according to this size, and the cropped image is grayscaled. Based on the processed grayscale image, a grayscale co-occurrence matrix with 256 gray levels and a step size of 1 in the four directions of 0°, 45°, 90°, and 135° is calculated. Four statistical texture feature parameters, namely contrast, angular second-order moment, consistency, and energy, are extracted based on the grayscale co-occurrence matrix and synthesized into a crack feature parameter dataset for all soil sample points in the study area.

[0013] Step 5: Establish conductivity prediction model

[0014] Using the neural network toolbox of MATLAB software, the spectral characteristic parameter dataset T of all sample points and the texture characteristic parameter dataset of all sample points were used as independent variables, all sample points were used as training samples, and the measured conductivity data of the training samples were used as dependent variables. A neural network prediction model for salinized soil conductivity based on UAV remote sensing image data was established.

[0015] Step 6: Conduct large-scale remote sensing inversion of soil electrical conductivity

[0016] A sliding window is determined according to the number of rows and columns corresponding to the sampling area of ​​the soil sample point. Convolution calculation is performed on the UAV remote sensing image covering the entire measurement area. The central pixel in each sliding window is extracted as the prediction sample. The spectral feature parameter dataset and texture feature parameter dataset of all prediction samples are brought into the neural network prediction model to realize UAV remote sensing measurement of the salt content of soda salinized soil.

[0017] Furthermore, step 1 is as follows:

[0018] The equipment used includes a multi-rotor drone, a handheld GPS, and a rectangular metal calibration frame with an inner diameter of 1m×1m. The drone is equipped with a high-definition CCD lens and a hyperspectral imaging spectrometer. First, an online map platform is used to determine a rectangular study area S0 where the conductivity of saline-alkali soil is to be measured. The four vertices P of the rectangular study area S0 are extracted according to the online map platform. 01 、P 02 、P 03 、P 04 The longitude and latitude positions, as well as the actual side length L of the rectangular study area S0, are then divided evenly into four rectangular sub-areas of equal size, and the side length L1 of each rectangular sub-area, where L1 = 0.5L. At the same time, the four rectangular sub-areas S are determined according to the row and column distribution. 11 、S 12 、S 13 and S 14, treat each sub-area as an inscribed square of the circular area covered by the CCD lens, and calculate the radius of the circular area At the same time, according to the vertical shooting field angle A of the CCD lens, the fixed aerial photography height of the UAV is calculated According to the four vertices P 01 、P 02 、P 03 and P 04 The longitude and latitude positions and the boundaries of each rectangular sub-area are used to determine the flight altitude H, where the take-off point of the drone is the position point P. 10 , the location point is the rectangular sub-area S 11 The middle point on the left boundary, point P 20 is a rectangular subregion S 12 The middle point on the right side of the border, point P 30 is a rectangular subregion S 13 The middle point on the left boundary, point P 40 is a rectangular subregion S 14 The middle point on the right side of the boundary is the flight endpoint; at the same time, the position point P 10 →P 20 →P 30 →P 40 The order of the flight route marking points is used as the UAV flight route marking points, and the UAV is controlled to cover the entire rectangular research area S0 according to the flight path of the flight route marking points, and the soil surface UAV CCD high-definition remote sensing image and hyperspectral remote sensing image are taken; because the lens coverage range is the circumscribed circle of the rectangular sub-area when the CCD lens is shooting, it can be guaranteed that the area of ​​the CCD high-definition UAV remote sensing image in each shooting is overlapped with the next shooting area, and in each rectangular sub-area S 11 、S 12 、S 13 and S 14 The center point is used as the sampling point P1, P2, P3 and P4, in the rectangular sub-area S 11 With rectangular subregion S 12 Set four sampling points P in the overlapping area 11 、P 12 、P 21 and P 22 ; In the rectangular sub-area S 12 With rectangular subregion S 14 Set four sampling points P in the overlapping area 23 、P 24 、P 41 and P 42 ; In the rectangular sub-area S 14 With rectangular subregion S 13 Set four sampling points P in the overlapping area 43 、P 44、P 31 and P 32 ; and rectangular sub-area S 13 With rectangular subregion S 11 Set four sampling points P in the overlapping area 33 、P 34 、P 13 and P 14 The longitude and latitude positions of the sampling points in each overlapping area are recorded, and a rectangular metal calibration frame is placed at each sampling point for the subsequent geometric correction and stitching of the UAV CCD high-definition remote sensing images and hyperspectral remote sensing images. Finally, after the UAV flight is completed, soil samples are collected at the center point of the rectangular sub-area and the sampling points in the overlapping area. After the collected soil samples are dried in the laboratory, a suspension with a water-soil mass ratio of 5:1 is prepared. The actual conductivity value of each soil sample is then measured in the laboratory, and all conductivity measurement results are saved as the conductivity dataset E, which serves as the modeling sample set for the soil conductivity prediction model of the entire rectangular study area.

[0019] Furthermore, step 2 is as follows:

[0020] Step 2.1, perform image mosaicking, geometric correction and cropping operations on the CCD high-definition remote sensing images of the four rectangular sub-areas acquired by the UAV at a fixed height H: 11 With rectangular subregion S 12 The captured CCD high-definition remote sensing images i1 and i2 are stitched together to generate a stitching result I1, and then the stitching result I1 is combined with the rectangular sub-area S 14 The captured CCD high-definition remote sensing image i3 is stitched to generate a stitching result I2, and then the stitching result I2 is combined with the rectangular sub-area S 13 The captured CCD high-definition remote sensing image i4 is stitched to generate a stitching result I3. Finally, the stitching result I3 is stitched again with the CCD high-definition remote sensing image i1 for stitching verification. If there is no ghosting, it proves that the stitching result I3 has high accuracy and can be used as the entire CCD high-definition remote sensing image corresponding to the final rectangular study area S0.

[0021] The splicing result I3 is geometrically corrected using the position points P at the four vertices of the rectangular study area S0. 01 、P 02 、P 03 and P 04 GPS latitude and longitude measured data, and each rectangular sub-area S 11 、S 12 、S 13 and S 14The GPS longitude and latitude measured data of the center points P1, P2, P3 and P4 are used to perform polynomial-based geometric correction on the corresponding position points on the splicing result I3, thereby achieving geometric distortion correction of all pixel points in the entire rectangular study area S0 and generating the corrected remote sensing image I4; finally, in ARCGIS software, a vector rectangular boundary ROI is generated according to the four vertices of the entire rectangular study area S0, the vector rectangular boundary ROI is overlapped with the remote sensing image I4, and all pixel points inside the vector rectangular boundary ROI are retained to achieve the cropping of the remote sensing image I4, and finally a rectangular CCD high-definition remote sensing image Z1 corresponding to the entire rectangular study area S0 is generated;

[0022] Step 2.2: Perform image mosaicking, geometric correction, and cropping operations on the hyperspectral remote sensing images of the four rectangular sub-areas acquired by the UAV at a fixed height: 11 and S 12 The measured hyperspectral remote sensing images p1 and p2 are stitched together to generate a stitching result Q1, and then the stitching result Q1 is combined with the rectangular sub-area S 14 The measured hyperspectral remote sensing image p3 is stitched to generate the stitching result Q2, and then the stitching result Q2 is combined with the rectangular sub-area S 13 The measured hyperspectral remote sensing image p4 is stitched to generate the stitching result Q3, and finally the stitching result Q3 is stitched again with the stitching result Q1 for verification. If there is no ghosting between the stitching result Q3 and the stitching result Q1, it proves that the stitching result Q3 has high accuracy and can be used as the entire high spectral resolution remote sensing image corresponding to the final rectangular study area S0. Secondly, the stitching result Q3 is geometrically corrected using the position points P at the four vertices of the rectangular study area S0. 01 , P0, P 03 and P 04 The GPS longitude and latitude measured data, as well as the GPS longitude and latitude measured data of the center points P1, P2, P3 and P4 of each rectangular sub-area, are used to perform geometric correction based on polynomials on the corresponding position points on the splicing result Q3, thereby achieving geometric distortion correction of all pixel points on the remote sensing image of the entire rectangular study area S0, and generating the corrected remote sensing image Q4; finally, in ARCGIS software, a vector rectangular boundary ROI is generated according to the four vertices of the entire rectangular study area S0, the vector rectangular boundary ROI is overlapped with the remote sensing image Q4, and all pixel points inside the vector rectangular boundary ROI are retained to achieve the cropping of the remote sensing image Q4, and finally a rectangular hyperspectral remote sensing image map Z2 corresponding to the entire rectangular study area S0 is generated;

[0023] Furthermore, step 3 is as follows:

[0024] For 20 sampling points P1, P2, P3, P4, P11 、P 12 、P 13 、P 14 、P 21 、P 22 、P 23 、P 24 、P 31 、P 32 、P 33 、P 34 、P 41 、P 42 、P 43 and P 44 , calculate the correlation coefficients between the conductivity measurement values ​​of the 20 sampling points and the reflectance values ​​of these sampling points in each band, draw the correlation coefficient curves of the 20 sampling points in all bands, and select the 5 bands with the highest correlation between reflectance and conductivity as the characteristic spectral bands according to the correlation coefficient curves. The wavelengths corresponding to the characteristic spectral bands are λ1, λ2, λ3, λ4 and λ5 respectively; then, extract the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance data set T1; calculate the first-order derivative of the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance first-order derivative data set T2; calculate the first-order derivative of the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance first-order derivative data set T2; 2, λ3, λ4 and λ5 and save it as a dataset of second-order derivatives of reflectance T3; calculate the logarithm of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of logarithms of reflectance T4; calculate the inverse of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of inverse reflectance T5; calculate the square root of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of square roots of reflectance T6; synthesize the datasets T1, T2, T3, T4, T5 and T6 into a dataset T of spectral reflectance characteristic parameters of all soil sample points in the rectangular study area S0;

[0025] Furthermore, step 4 is as follows:

[0026] According to the two vertices P on the left boundary of the rectangular research area S0 01 、P 03 The latitude W1 and W3 of the location, and the row number H of the image Z1 corresponding to the entire rectangular study area S0, calculate the actual size of each pixel At the same time, calculate the number of pixels of the side length of the rectangular image area corresponding to the inner diameter of 1m×1m in the rectangular metal calibration frame Then for the 20 sampling points P1, P2, P3, P4, P 11 、P12 、P 13 、P 14 、P 21 、P 22 、P 23 、P 24 、P 31 、P 32 、P 33 、P 34 、P 41 、P 42 、P 43 and P 44 As the center, 20 rectangular CCD high-definition remote sensing sub-images B1, B2, B3, B4, B 11 、B 12 、B 13 、B 14 、B 21 、B 22 、B 23 、B 24 、B 31 、B 32 、B 33 、B 34 、B 41 、B 42 、B 43 and B 44 , grayscale processing is performed on the cropped image, and 256 gray levels are calculated based on the processed grayscale image. The gray level co-occurrence matrices M1, M2, M3 and M4 of the rectangular CCD high-definition remote sensing sub-image corresponding to each sampling point at 0°, 45°, 90° and 135° with a step size of 1 are calculated. In each direction, the four-directional average contrast texture feature of the rectangular CCD high-definition remote sensing sub-image corresponding to all sample points is calculated and saved as dataset C1, the four-directional average energy value texture feature is saved as dataset C2, the four-directional average entropy value texture feature is saved as dataset C3, and the four-directional average consistency texture feature is saved as dataset C4; the datasets C1, C2, C3 and C4 are synthesized into the crack characteristic parameter dataset C of all soil sample points in the study area;

[0027] Furthermore, step 5 is as follows:

[0028] Using the neural network toolbox of MATLAB software and the spectral reflectance characteristic parameter dataset T and crack characteristic parameter dataset C of all sample points, various spectral reflectance characteristic parameter datasets in dataset T are standardized to generate a standardized spectral reflectance characteristic parameter dataset T', and various soil crack characteristic parameter datasets in dataset C are standardized to generate a standardized crack characteristic parameter dataset C', thereby establishing a soil conductivity prediction model; using all 20 sampling points as training samples, the standardized spectral reflectance characteristic parameter dataset T' and the standardized crack characteristic parameter dataset C' of the training samples as independent variables, and the soil conductivity measurement dataset E as the dependent variable, an artificial neural network prediction model is established, and the number of neural network iterations k1 = 100, the error threshold k2 = 0.4, and the initial learning rate k3 = 0.2 are set. On this basis, a neural network prediction model for soil conductivity is established, and the model form is E = f(T', C'); further, step 6 is specifically as follows:

[0029] According to the training sample points, that is, the number of pixels U1 of the side length of the rectangular image area corresponding to the inner diameter of the 1m×1m in the rectangular metal calibration frame of the sampling point, a rectangular sliding window is established, and the side length of the rectangular sliding window is U1; then, starting from the first pixel in the upper left corner of the image map Z2 and the image map Z1, respectively, the sliding window covers the image map Z1 and the image map Z2, and traverses the entire image map Z1 and the image map Z2 element by element;

[0030] For image Z2, each time a pixel is slid, according to step 3, the reflectivity t1 of the central pixel point in the sliding window at the spectral bands λ1, λ2, λ3, λ4, and λ5 is extracted respectively; the first-order derivative t2 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; the second-order derivative t3 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; the logarithm t4 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; The reciprocal t5 of the reflectance of the central pixel point in the window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; the square root t6 of the reflectance of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated, and the reflectance t1, first-order derivative t2, second-order derivative t3, logarithm t4, reciprocal t5, and square root t6 are combined into the reflectance characteristic parameter data set t(i1, j1) of the central pixel in the window, where i1 is the row number of the central pixel of the current sliding window in the image map Z2, and j1 is the column number of the central pixel of the current sliding window in the image map Z2;

[0031] For the image map Z1, each time a pixel is slid, according to step 4, the four-directional average contrast texture feature c1 of the central pixel of the sliding window is extracted, the four-directional average energy value texture feature c2 of the central pixel of the sliding window is extracted, the four-directional average entropy value texture feature c3 of the central pixel of the sliding window is extracted, and the four-directional average consistency texture feature c4 of the central pixel of the sliding window is extracted. The contrast texture feature c1, the energy value texture feature c2, the entropy value texture feature c3 and the consistency texture feature c4 are combined into the crack feature parameter data set c(i2, j2) of the central pixel in the window, where i2 is the row number of the central pixel of the current sliding window in the image map Z1, and j2 is the column number of the central pixel of the current sliding window in the image map Z1;

[0032] The reflectivity characteristic parameter dataset t(i1,j1)=T' and the crack characteristic parameter dataset c(i2,j2)=C' are introduced into the artificial neural network prediction model E=f(T′,C′) to calculate the conductivity prediction value of the central pixel of the sliding window; then this method is used to traverse the entire image map Z1 and image map Z2, and finally realize the remote sensing measurement of the soil conductivity values ​​in all pixels within the rectangular study area S0.

[0033] The present invention offers significant benefits and advantages: Compared with traditional laboratory conductivity measurements and field conductivity measurements using geodetic conductivity meters, the method disclosed herein offers low cost, low production cost, minimal losses, and the ability to achieve real-time, synchronous observation of soil conductivity over large areas. Furthermore, it offers high mobility. The method objectively reflects the precipitation of salt minerals on the soil surface. The spectral characteristic parameters extracted from the diagnostic spectral bands of salt minerals are directly affected by soil salinity, resulting in a clear parameter relationship in the prediction model and high measurement accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a schematic diagram of UAV ground measurement;

[0035] Figure 2 This is a schematic diagram of the UAV's flight trajectory and flight range;

[0036] Figure 3 Schematic diagram of the rectangular overall area, rectangular sub-areas and related location points of the study area. DETAILED DESCRIPTION

[0037] The present invention is further described with examples below:

[0038] Example 1

[0039] A method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from drones, the steps are as follows:

[0040] Step 1: Acquisition of UAV remote sensing images

[0041] like Figure 1-3 As shown in the figure, the equipment used includes a multi-rotor drone, a handheld GPS, and a rectangular metal calibration frame with an inner diameter of 1m×1m. The drone is equipped with a high-definition CCD lens and a hyperspectral imaging spectrometer. First, an online map platform is used to determine a rectangular study area S0 where the conductivity of saline-alkali soil is to be measured. The four vertices P of the rectangular study area S0 are extracted according to the online map platform. 01 、P 02 、P 03 、P 04 The longitude and latitude positions, as well as the actual side length L of the rectangular study area S0, are then divided evenly into four rectangular sub-areas of equal size, and the side length L1 of each rectangular sub-area, where L1 = 0.5L. At the same time, the four rectangular sub-areas S are determined according to the row and column distribution. 11 、S 12 、S 13 and S 14 , treat each sub-area as an inscribed square of the circular area covered by the CCD lens, and calculate the radius of the circular area At the same time, according to the vertical shooting field angle A of the CCD lens, the fixed aerial photography height of the UAV is calculated According to the four vertices P 01 、P 02 、P 03 and P 04 The longitude and latitude positions and the boundaries of each rectangular sub-area are used to determine the flight altitude H, where the take-off point of the drone is the position point P. 10 , the location point is the rectangular sub-area S 11 The middle point on the left boundary, point P 20 is a rectangular subregion S 12 The middle point on the right side of the border, point P 30 is a rectangular subregion S 13 The middle point on the left boundary, point P 40 is a rectangular subregion S 14 The middle point on the right side of the boundary is the flight endpoint; at the same time, the position point P 10 →P 20 →P 30 →P 40The order of the flight route marking points is used as the UAV flight route marking points, and the UAV is controlled to cover the entire rectangular research area S0 according to the flight path of the flight route marking points, and the soil surface UAV CCD high-definition remote sensing image and hyperspectral remote sensing image are taken; because the lens coverage range is the circumscribed circle of the rectangular sub-area when the CCD lens is shooting, it can be guaranteed that the area of ​​the CCD high-definition UAV remote sensing image in each shooting is overlapped with the next shooting area, and in each rectangular sub-area S 11 、S 12 、S 13 and S 14 The center point is used as the sampling point P1, P2, P3 and P4, in the rectangular sub-area S 11 With rectangular subregion S 12 Set four sampling points P in the overlapping area 11 、P 12 、P 21 and P 22 ; In the rectangular sub-area S 12 With rectangular subregion S 14 Set four sampling points P in the overlapping area 23 、P 24 、P 41 and P 42 ; In the rectangular sub-area S 14 With rectangular subregion S 13 Set four sampling points P in the overlapping area 43 、P 44 、P 31 and P 32 ; and rectangular sub-area S 13 With rectangular subregion S 11 Set four sampling points P in the overlapping area 33 、P 34 、P 13 and P 14 , record the longitude and latitude positions of the sampling points in each overlapping area, and place a rectangular metal calibration frame at each sampling point for subsequent geometric correction and stitching of the UAV CCD high-definition remote sensing images and hyperspectral remote sensing images; finally, after the UAV flight is completed, soil samples are collected at the center point of the rectangular sub-area and the sampling points in the overlapping area. After the collected soil samples are dried in the laboratory, a suspension with a water-soil mass ratio of 5:1 is prepared. Then, the true conductivity value of each soil sample is measured in the laboratory, and all conductivity measurement results are saved as the conductivity dataset E, which serves as the modeling sample set for the soil conductivity prediction model of the entire rectangular study area;

[0042] Step 2: Preprocessing of UAV remote sensing images

[0043] Step 2.1, perform image mosaicking, geometric correction and cropping operations on the CCD high-definition remote sensing images of the four rectangular sub-areas acquired by the UAV at a fixed height H: 11 With rectangular subregion S 12 The captured CCD high-definition remote sensing images i1 and i2 are stitched together to generate a stitching result I1, and then the stitching result I1 is combined with the rectangular sub-area S 14 The captured CCD high-definition remote sensing image i3 is stitched to generate a stitching result I2, and then the stitching result I2 is combined with the rectangular sub-area S 13 The captured CCD high-definition remote sensing image i4 is stitched to generate a stitching result I3. Finally, the stitching result I3 is stitched again with the CCD high-definition remote sensing image i1 for stitching verification. If there is no ghosting, it proves that the stitching result I3 has high accuracy and can be used as the entire CCD high-definition remote sensing image corresponding to the final rectangular study area S0.

[0044] The splicing result I3 is geometrically corrected using the position points P at the four vertices of the rectangular study area S0. 01 、P 02 、P 03 and P 04 GPS latitude and longitude measured data, and each rectangular sub-area S 11 、S 12 、S 13 and S 14 The GPS longitude and latitude measured data of the center points P1, P2, P3 and P4 are used to perform polynomial-based geometric correction on the corresponding position points on the splicing result I3, thereby achieving geometric distortion correction of all pixel points in the entire rectangular study area S0 and generating the corrected remote sensing image I4; finally, in ARCGIS software, a vector rectangular boundary ROI is generated according to the four vertices of the entire rectangular study area S0, the vector rectangular boundary ROI is overlapped with the remote sensing image I4, and all pixel points inside the vector rectangular boundary ROI are retained to achieve the cropping of the remote sensing image I4, and finally a rectangular CCD high-definition remote sensing image Z1 corresponding to the entire rectangular study area S0 is generated;

[0045] Step 2.2: Perform image mosaicking, geometric correction, and cropping operations on the hyperspectral remote sensing images of the four rectangular sub-areas acquired by the UAV at a fixed height: 11 and S 12 The measured hyperspectral remote sensing images p1 and p2 are stitched together to generate a stitching result Q1, and then the stitching result Q1 is combined with the rectangular sub-area S 14 The measured hyperspectral remote sensing image p3 is stitched to generate the stitching result Q2, and then the stitching result Q2 is combined with the rectangular sub-area S 13The measured hyperspectral remote sensing image p4 is stitched to generate the stitching result Q3, and finally the stitching result Q3 is stitched again with the stitching result Q1 for verification. If there is no ghosting between the stitching result Q3 and the stitching result Q1, it proves that the stitching result Q3 has high accuracy and can be used as the entire high spectral resolution remote sensing image corresponding to the final rectangular study area S0. Secondly, the stitching result Q3 is geometrically corrected using the position points P at the four vertices of the rectangular study area S0. 01 , P0, P 03 and P 04 The GPS longitude and latitude measured data, as well as the GPS longitude and latitude measured data of the center points P1, P2, P3 and P4 of each rectangular sub-area, are used to perform geometric correction based on polynomials on the corresponding position points on the splicing result Q3, thereby achieving geometric distortion correction of all pixel points on the remote sensing image of the entire rectangular study area S0, and generating the corrected remote sensing image Q4; finally, in ARCGIS software, a vector rectangular boundary ROI is generated according to the four vertices of the entire rectangular study area S0, the vector rectangular boundary ROI is overlapped with the remote sensing image Q4, and all pixel points inside the vector rectangular boundary ROI are retained to achieve the cropping of the remote sensing image Q4, and finally a rectangular hyperspectral remote sensing image map Z2 corresponding to the entire rectangular study area S0 is generated;

[0046] Step 3: Extract spectral characteristic parameters

[0047] For 20 sampling points P1, P2, P3, P4, P 11 、P 12 、P 13 、P 14 、P 21 、P 22 、P 23 、P 24 、P 31 、P 32 、P 33 、P 34 、P 41 、P 42 、P 43 and P 44, calculate the correlation coefficients between the conductivity measurement values ​​of the 20 sampling points and the reflectance values ​​of these sampling points in each band, draw the correlation coefficient curves of the 20 sampling points in all bands, and select the 5 bands with the highest correlation between reflectance and conductivity as the characteristic spectral bands according to the correlation coefficient curves. The wavelengths corresponding to the characteristic spectral bands are λ1, λ2, λ3, λ4 and λ5 respectively; then, extract the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance data set T1; calculate the first-order derivative of the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance first-order derivative data set T2; calculate the first-order derivative of the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance first-order derivative data set T2; 2, λ3, λ4 and λ5 and save it as a dataset of second-order derivatives of reflectance T3; calculate the logarithm of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of logarithms of reflectance T4; calculate the inverse of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of inverse reflectance T5; calculate the square root of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of square roots of reflectance T6; synthesize the datasets T1, T2, T3, T4, T5 and T6 into a dataset T of spectral reflectance characteristic parameters of all soil sample points in the rectangular study area S0;

[0048] Step 4: Extract soil surface characteristic parameters

[0049] According to the two vertices P on the left boundary of the rectangular research area S0 01 、P 03 The latitude W1 and W3 of the location, and the row number H of the image Z1 corresponding to the entire rectangular study area S0, calculate the actual size of each pixel At the same time, calculate the number of pixels of the side length of the rectangular image area corresponding to the inner diameter of 1m×1m in the rectangular metal calibration frame Then for the 20 sampling points P1, P2, P3, P4, P 11 、P 12 、P 13 、P 14 、P 21 、P 22 、P 23 、P 24 、P 31 、P 32 、P 33 、P 34 、P 41 、P 42 、P 43 and P 44As the center, 20 rectangular CCD high-definition remote sensing sub-images B1, B2, B3, B4, B 11 、B 12 、B 13 、B 14 、B 21 、B 22 、B 23 、B 24 、B 31 、B 32 、B 33 、B 34 、B 41 、B 42 、B 43 and B 44 , grayscale processing is performed on the cropped image, and 256 gray levels are calculated based on the processed grayscale image. The gray level co-occurrence matrices M1, M2, M3 and M4 of the rectangular CCD high-definition remote sensing sub-image corresponding to each sampling point at 0°, 45°, 90° and 135° with a step size of 1 are calculated. In each direction, the four-directional average contrast texture feature of the rectangular CCD high-definition remote sensing sub-image corresponding to all sample points is calculated and saved as dataset C1, the four-directional average energy value texture feature is saved as dataset C2, the four-directional average entropy value texture feature is saved as dataset C3, and the four-directional average consistency texture feature is saved as dataset C4; the datasets C1, C2, C3 and C4 are synthesized into the crack characteristic parameter dataset C of all soil sample points in the study area;

[0050] Step 5: Establish conductivity prediction model

[0051] Using the neural network toolbox of MATLAB software and the spectral reflectance characteristic parameter dataset T and crack characteristic parameter dataset C of all sample points, the various spectral reflectance characteristic parameter datasets in the dataset T are standardized to generate a standardized spectral reflectance characteristic parameter dataset T', and the various soil crack characteristic parameter datasets in the dataset C are standardized to generate a standardized crack characteristic parameter dataset C', thereby realizing the establishment of a soil conductivity prediction model; using all 20 sampling points as training samples, the standardized spectral reflectance characteristic parameter dataset T' and the standardized crack characteristic parameter dataset C' of the training samples as independent variables, and the soil conductivity measurement dataset E as the dependent variable, an artificial neural network prediction model is established, and the number of neural network iterations k1 = 100, the error threshold k2 = 0.4, and the initial learning rate k3 = 0.2 are set. On this basis, a neural network prediction model for soil conductivity is established, and the model form is E = f(T', C');

[0052] Step 6: Conduct large-scale remote sensing inversion of soil electrical conductivity

[0053] According to the training sample points, that is, the number of pixels U1 of the side length of the rectangular image area corresponding to the inner diameter of the 1m×1m in the rectangular metal calibration frame of the sampling point, a rectangular sliding window is established, and the side length of the rectangular sliding window is U1; then, starting from the first pixel in the upper left corner of the image map Z2 and the image map Z1, respectively, the sliding window covers the image map Z1 and the image map Z2, and traverses the entire image map Z1 and the image map Z2 element by element;

[0054] For image Z2, each time a pixel is slid, according to step 3, the reflectivity t1 of the central pixel point in the sliding window at the spectral bands λ1, λ2, λ3, λ4, and λ5 is extracted respectively; the first-order derivative t2 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; the second-order derivative t3 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; the logarithm t4 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; The reciprocal t5 of the reflectance of the central pixel point in the window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; the square root t6 of the reflectance of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated, and the reflectance t1, first-order derivative t2, second-order derivative t3, logarithm t4, reciprocal t5, and square root t6 are combined into the reflectance characteristic parameter data set t(i1, j1) of the central pixel in the window, where i1 is the row number of the central pixel of the current sliding window in the image map Z2, and j1 is the column number of the central pixel of the current sliding window in the image map Z2;

[0055] For the image map Z1, each time a pixel is slid, according to step 4, the four-directional average contrast texture feature c1 of the central pixel of the sliding window is extracted, the four-directional average energy value texture feature c2 of the central pixel of the sliding window is extracted, the four-directional average entropy value texture feature c3 of the central pixel of the sliding window is extracted, and the four-directional average consistency texture feature c4 of the central pixel of the sliding window is extracted. The contrast texture feature c1, the energy value texture feature c2, the entropy value texture feature c3 and the consistency texture feature c4 are combined into the crack feature parameter data set c(i2, j2) of the central pixel in the window, where i2 is the row number of the central pixel of the current sliding window in the image map Z1, and j2 is the column number of the central pixel of the current sliding window in the image map Z1;

[0056] The reflectivity characteristic parameter dataset t(i1,j1)=T' and the crack characteristic parameter dataset c(i2,j2)=C' are introduced into the artificial neural network prediction model E=f(T′,C′) to calculate the conductivity prediction value of the central pixel of the sliding window; then this method is used to traverse the entire image map Z1 and image map Z2, and finally realize large-scale, rapid and synchronous remote sensing measurement of soil conductivity values ​​in all pixels in the study area S0.

Claims

1. A method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from unmanned aerial vehicles, characterized in that: Here are the steps: Step 1: Acquisition of UAV remote sensing images The equipment used included a multi-rotor drone, a handheld GPS, and a rectangular metal calibration frame with an inner diameter of 1m×1m. The drone was equipped with a high-definition CCD lens and a hyperspectral imaging spectrometer. The rectangular study area was divided into four sub-areas for capturing high-definition UAV CCD remote sensing images and hyperspectral remote sensing images of the soil surface. Sampling points were selected for soil sample collection. A rectangular metal calibration frame was placed at each sampling point for subsequent geometric correction and stitching of the UAV CCD high-definition remote sensing images and hyperspectral remote sensing images. The true value of the electrical conductivity of each soil sample was measured, and all conductivity measurement results were saved as a conductivity dataset E, which served as the modeling sample set for the soil conductivity prediction model for the entire study area. Step 2: Preprocessing of UAV remote sensing images The CCD high-definition remote sensing images taken in the four rectangular sub-areas are mosaicked, geometrically corrected, and cropped to generate a rectangular CCD high-definition remote sensing image map Z1 corresponding to the entire rectangular study area; the hyperspectral remote sensing images taken in the four rectangular sub-areas are mosaicked, geometrically corrected, and cropped to generate a rectangular hyperspectral remote sensing image map Z2 corresponding to the entire rectangular study area; Step 3: Extract spectral characteristic parameters According to the location information of the sampling point, the full-band hyperspectral reflectance data of the sampling point is found. On this basis, the characteristic spectral band of the sampling point is calculated, and the first-order derivative, second-order derivative, logarithm, reciprocal and square root mathematical transformation parameters of the characteristic band spectral reflectance data are further calculated as spectral characteristic parameters to form a spectral reflectance characteristic parameter dataset of the soil sample point; Step 4: Extract soil surface characteristic parameters According to the location information of the sampling points, the number of image pixel rows and columns corresponding to a 1m×1m surface area is determined according to the scale. With each sample point as the center, the image size of the soil sample surface is determined according to the size of the surface area. All sample point data are cropped according to this size, and the cropped image is grayscaled. Based on the processed grayscale image, a grayscale co-occurrence matrix with 256 gray levels and a step size of 1 in the four directions of 0°, 45°, 90°, and 135° is calculated. Four statistical texture feature parameters, namely contrast, angular second-order moment, consistency, and energy, are extracted based on the grayscale co-occurrence matrix and synthesized into a crack feature parameter dataset for all soil sample points in the study area. Step 5: Establish conductivity prediction model Using the neural network toolbox of MATLAB software, a neural network prediction model for salinized soil conductivity based on UAV remote sensing image data was established, with the spectral reflectance characteristic parameter dataset of all sample points and the crack characteristic parameter dataset of all sample points as independent variables, all sample points as training samples, and the measured conductivity data of the training samples as the dependent variable. Step 6: Conduct large-scale remote sensing inversion of soil electrical conductivity A sliding window is determined according to the number of rows and columns corresponding to the sampling area of ​​the soil sample point. Convolution calculation is performed on the UAV remote sensing image covering the entire measurement area. The central pixel in each sliding window is extracted as the prediction sample. The spectral feature parameter dataset and texture feature parameter dataset of all prediction samples are brought into the neural network prediction model to realize UAV remote sensing measurement of the salt content of soda salinized soil.

2. The method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images of unmanned aerial vehicles according to claim 1, characterized in that: Step 1 is as follows: The equipment used includes a multi-rotor drone, a handheld GPS, and a rectangular metal calibration frame with an inner diameter of 1m×1m. The drone is equipped with a high-definition CCD lens and a hyperspectral imaging spectrometer. First, an online map platform is used to determine a rectangular study area S0 where the conductivity of saline-alkali soil is to be measured. The four vertices P of the rectangular study area S0 are extracted according to the online map platform. 01 、P 02 、P 03 、P 04 The longitude and latitude positions, as well as the actual side length L of the rectangular study area S0, are then divided evenly into four rectangular sub-areas of equal size, and the side length L1 of each rectangular sub-area, where L1 = 0.5L. At the same time, the four rectangular sub-areas S are determined according to the row and column distribution. 11 、S 12 、S 13 and S 14 , treat each sub-area as an inscribed square of the circular area covered by the CCD lens, and calculate the radius of the circular area At the same time, according to the vertical shooting field angle A of the CCD lens, the fixed aerial photography height of the UAV is calculated According to the four vertices P 01 、P 02 、P 03 and P 04 The longitude and latitude positions and the boundaries of each rectangular sub-area are used to determine the flight altitude H, where the take-off point of the drone is the position point P. 10 , the location point is the rectangular sub-area S 11 The middle point on the left boundary, point P 20 is a rectangular subregion S 12 The middle point on the right side of the border, point P 30 is a rectangular subregion S 13 The middle point on the left boundary, point P 40 is a rectangular subregion S 14 The middle point on the right side of the boundary is the flight endpoint; at the same time, the position point P 10 →P 20 →P 30 →P 40 The order of the flight route marking points is used as the UAV flight route marking points, and the UAV is controlled to cover the entire rectangular research area S0 according to the flight path of the flight route marking points, and the soil surface UAV CCD high-definition remote sensing image and hyperspectral remote sensing image are taken; because the lens coverage range is the circumscribed circle of the rectangular sub-area when the CCD lens is shooting, it can be guaranteed that the area of ​​the CCD high-definition UAV remote sensing image in each shooting is overlapped with the next shooting area, and in each rectangular sub-area S 11 、S 12 、S 13 and S 14 The center point is used as the sampling point P1, P2, P3 and P4, in the rectangular sub-area S 11 With rectangular subregion S 12 Set four sampling points P in the overlapping area 11 、P 12 、P 21 and P 22 ; In the rectangular sub-area S 12 With rectangular subregion S 14 Set four sampling points P in the overlapping area 23 、P 24 、P 41 and P 42 ; In the rectangular sub-area S 14 With rectangular subregion S 13 Set four sampling points P in the overlapping area 43 、P 44 、P 31 and P 32 ; and rectangular sub-area S 13 With rectangular subregion S 11 Set four sampling points P in the overlapping area 33 、P 34 、P 13 and P 14 The longitude and latitude positions of the sampling points in each overlapping area are recorded, and a rectangular metal calibration frame is placed at each sampling point for the subsequent geometric correction and stitching of the UAV CCD high-definition remote sensing images and hyperspectral remote sensing images. Finally, after the UAV flight is completed, soil samples are collected at the center point of the rectangular sub-area and the sampling points in the overlapping area. After the collected soil samples are dried in the laboratory, a suspension with a water-soil mass ratio of 5:1 is prepared. The actual conductivity value of each soil sample is then measured in the laboratory, and all conductivity measurement results are saved as the conductivity dataset E, which serves as the modeling sample set for the soil conductivity prediction model of the entire rectangular study area.

3. The method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images of unmanned aerial vehicles according to claim 2, characterized in that: Step 2 is as follows: Step 2.1, perform image mosaicking, geometric correction and cropping operations on the CCD high-definition remote sensing images of the four rectangular sub-areas acquired by the UAV at a fixed height H: 11 With rectangular subregion S 12 The captured CCD high-definition remote sensing images i1 and i2 are stitched together to generate a stitching result I1, and then the stitching result I1 is combined with the rectangular sub-area S 14 The captured CCD high-definition remote sensing image i3 is stitched to generate a stitching result I2, and then the stitching result I2 is combined with the rectangular sub-area S 13 The captured CCD high-definition remote sensing image i4 is stitched to generate a stitching result I3. Finally, the stitching result I3 is stitched again with the CCD high-definition remote sensing image i1 for stitching verification. If there is no ghosting, it proves that the stitching result I3 has high accuracy and can be used as the entire CCD high-definition remote sensing image corresponding to the final rectangular study area S0. The splicing result I3 is geometrically corrected using the position points P at the four vertices of the rectangular study area S0. 01 、P 02 、P 03 and P 04 GPS latitude and longitude measured data, and each rectangular sub-area S 11 、S 12 、S 13 and S 14 The GPS longitude and latitude measured data of the center points P1, P2, P3 and P4 are used to perform polynomial-based geometric correction on the corresponding position points on the splicing result I3, thereby achieving geometric distortion correction of all pixel points in the entire rectangular study area S0 and generating the corrected remote sensing image I4; finally, in ARCGIS software, a vector rectangular boundary ROI is generated according to the four vertices of the entire rectangular study area S0, the vector rectangular boundary ROI is overlapped with the remote sensing image I4, and all pixel points inside the vector rectangular boundary ROI are retained to achieve the cropping of the remote sensing image I4, and finally a rectangular CCD high-definition remote sensing image Z1 corresponding to the entire rectangular study area S0 is generated; Step 2.2: Perform image mosaicking, geometric correction, and cropping operations on the hyperspectral remote sensing images of the four rectangular sub-areas acquired by the UAV at a fixed height: 11 and S 12 The measured hyperspectral remote sensing images q1 and q2 are stitched together to generate the stitching result Q1, and then the stitching result Q1 is combined with the rectangular sub-area S 14 The measured hyperspectral remote sensing image q3 is stitched to generate the stitching result Q2, and then the stitching result Q2 is combined with the rectangular sub-area S 13 The measured hyperspectral remote sensing image q4 is stitched to generate the stitching result Q3, and finally the stitching result Q3 is stitched again with the stitching result Q1 for verification. If there is no ghosting between the stitching result Q3 and the stitching result Q1, it proves that the stitching result Q3 has high accuracy and can be used as the entire high spectral resolution remote sensing image corresponding to the final rectangular study area S0. Secondly, the stitching result Q3 is geometrically corrected using the position points P at the four vertices of the rectangular study area S0. 01 , P0, P 03 and P 04 The GPS longitude and latitude measured data, as well as the GPS longitude and latitude measured data of the center points P1, P2, P3 and P4 of each rectangular sub-area, are used to perform polynomial-based geometric correction on the corresponding position points on the stitching result Q3, so as to achieve geometric distortion correction of all pixel points on the remote sensing image of the entire rectangular study area S0, and generate the corrected remote sensing image Q4; finally, in ARCGIS software, a vector rectangular boundary ROI is generated according to the four vertices of the entire rectangular study area S0, the vector rectangular boundary ROI is overlapped with the remote sensing image Q4, and all pixel points inside the vector rectangular boundary ROI are retained to achieve the cropping of the remote sensing image Q4, and finally a rectangular hyperspectral remote sensing image Z2 corresponding to the entire rectangular study area S0 is generated.

4. The method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images of unmanned aerial vehicles according to claim 3, characterized in that: Step 3 is as follows: For 20 sampling points P1, P2, P3, P4, P 11 、P 12 、P 13 、P 14 、P 21 、P 22 、P 23 、P 24 、P 31 、P 32 、P 33 、P 34 、P 41 、P 42 、P 43 and P 44 , calculate the correlation coefficients between the conductivity measurement values ​​of the 20 sampling points and the reflectance values ​​of these sampling points in each band, draw the correlation coefficient curves of the 20 sampling points in all bands, and select the 5 bands with the highest correlation between reflectance and conductivity as the characteristic spectral bands according to the correlation coefficient curves. The wavelengths corresponding to the characteristic spectral bands are λ1, λ2, λ3, λ4 and λ5 respectively; then, extract the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance data set T1; calculate the first-order derivative of the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance first-order derivative data set T2; calculate the first-order derivative of the reflectance of each sampling point in the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save them as the reflectance first-order derivative data set T2; 2, λ3, λ4 and λ5 and save it as a dataset of second-order derivatives of reflectance T3; calculate the logarithm of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of logarithms of reflectance T4; calculate the inverse of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of inverse reflectance T5; calculate the square root of the reflectance of each sampling point at the characteristic spectral bands λ1, λ2, λ3, λ4 and λ5 and save it as a dataset of square roots of reflectance T6; the datasets T1, T2, T3, T4, T5 and T6 are synthesized into a dataset T of spectral reflectance characteristic parameters of all soil sample points in the rectangular study area S0.

5. The method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images of unmanned aerial vehicles according to claim 4, characterized in that: Step 4 is as follows: According to the two vertices P on the left boundary of the rectangular research area S0 01 、P 03 The latitude W1 and W3 of the location, and the row number H of the image Z1 corresponding to the entire rectangular study area S0, calculate the actual size of each pixel At the same time, calculate the number of pixels of the side length of the rectangular image area corresponding to the inner diameter of 1m×1m in the rectangular metal calibration frame Then for the 20 sampling points P1, P2, P3, P4, P 11 、P 12 、P 13 、P 14 、P 21 、P 22 、P 23 、P 24 、P 31 、P 32 、P 33 、P 34 、P 41 、P 42 、P 43 and P 44 As the center, 20 rectangular CCD high-definition remote sensing sub-images B1, B2, B3, B4, B 11 、B 12 、B 13 、B 14 、B 21 、B 22 、B 23 、B 24 、B 31 、B 32 、B 33 、B 34 、B 41 、B 42 、B 43 and B 44 , grayscale processing is performed on the cropped result image, and 256 gray levels are calculated based on the processed grayscale image. The gray level co-occurrence matrices M1, M2, M3 and M4 of the rectangular CCD high-definition remote sensing sub-image corresponding to each sampling point at 0°, 45°, 90° and 135° with a step size of 1 are calculated. In each direction, the four-directional average contrast texture feature of the rectangular CCD high-definition remote sensing sub-image corresponding to all sample points is calculated and saved as dataset C1, the four-directional average energy value texture feature is saved as dataset C2, the four-directional average entropy value texture feature is saved as dataset C3, and the four-directional average consistency texture feature is saved as dataset C4; the datasets C1, C2, C3 and C4 are synthesized into the crack characteristic parameter dataset C of all soil sample points in the study area.

6. The method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from unmanned aerial vehicles according to claim 5, characterized in that: Step 5 is as follows: Using the neural network toolbox of MATLAB software, as well as the spectral reflectance characteristic parameter dataset T and crack characteristic parameter dataset C of all sample points, the various spectral reflectance characteristic parameter datasets in the dataset T are standardized to generate a standardized spectral reflectance characteristic parameter dataset T', and the various soil crack characteristic parameter datasets in the dataset C are standardized to generate a standardized crack characteristic parameter dataset C', thereby realizing the establishment of a soil conductivity prediction model; using all 20 sampling points as training samples, the standardized spectral reflectance characteristic parameter dataset T' and the standardized crack characteristic parameter dataset C' of the training samples as independent variables, and the soil conductivity measurement dataset E as the dependent variable, an artificial neural network prediction model is established, and the number of neural network iterations k1 = 100, the error threshold k2 = 0.4, and the initial learning rate k3 = 0.2 are set. On this basis, a neural network prediction model for soil conductivity is established, and the model form is E = f(T′, C′).

7. The method for measuring the electrical conductivity of cracked saline-alkali soil based on low-altitude remote sensing images from unmanned aerial vehicles according to claim 6, characterized in that: Step 6 is as follows: According to the training sample points, that is, the number of pixels U1 of the side length of the rectangular image area corresponding to the inner diameter of the 1m×1m in the rectangular metal calibration frame of the sampling point, a rectangular sliding window is established, and the side length of the rectangular sliding window is U1; then, starting from the first pixel in the upper left corner of the image map Z2 and the image map Z1, respectively, the sliding window covers the image map Z1 and the image map Z2, and traverses the entire image map Z1 and the image map Z2 element by element; For image Z2, each time a pixel is slid, according to step 3, the reflectance t1 of the central pixel point in the sliding window at the spectral bands λ1, λ2, λ3, λ4, and λ5 is extracted respectively, and the first-order derivative t2 of the reflectance of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; the second-order derivative t3 of the reflectance of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5 is calculated; Calculate the logarithm t4 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5; calculate the reciprocal t5 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5; calculate the square root t6 of the reflectivity of the central pixel point in the sliding window at the characteristic spectral bands λ1, λ2, λ3, λ4, and λ5, and combine the reflectivity t1, first-order derivative t2, second-order derivative t3, logarithm t4, reciprocal t5, and square root t6 into the reflectivity characteristic parameter data set t(i1, j1) of the central pixel in the window, where i1 is the row number of the central pixel of the current sliding window in the image map Z2, and j1 is the column number of the central pixel of the current sliding window in the image map Z2; For the image map Z1, each time a pixel is slid, according to step 4, the four-directional average contrast texture feature c1 of the central pixel of the sliding window is extracted, the four-directional average energy value texture feature c2 of the central pixel of the sliding window is extracted, the four-directional average entropy value texture feature c3 of the central pixel of the sliding window is extracted, and the four-directional average consistency texture feature c4 of the central pixel of the sliding window is extracted. The contrast texture feature c1, the energy value texture feature c2, the entropy value texture feature c3 and the consistency texture feature c4 are combined into the crack feature parameter data set c(i2, j2) of the central pixel in the window, where i2 is the row number of the central pixel of the current sliding window in the image map Z1, and j2 is the column number of the central pixel of the current sliding window in the image map Z1; The reflectivity characteristic parameter dataset t(i1,j1)=T' and the crack characteristic parameter dataset c(i2,j2)=C' are introduced into the artificial neural network prediction model E=f(T′,C′) to calculate the conductivity prediction value of the central pixel of the sliding window; then this method is used to traverse the entire image map Z1 and image map Z2, and finally realize the remote sensing measurement of the soil conductivity values ​​in all pixels within the rectangular study area S0.

Images