Farmland soil moisture content fine integration inversion method cooperating with unmanned aerial vehicle multispectral and thermal infrared images

Abstract

The invention discloses a farmland soil moisture content fine integration inversion method cooperating with unmanned aerial vehicle multispectral and thermal infrared images. The invention relates to the technical field of farmland soil moisture content integrated inversion, and solves the problems of insufficient inversion precision and weak stability of a conventional machine learning inversion algorithm. The method comprises the steps of obtaining and preprocessing unmanned aerial vehicle multispectral and thermal infrared image data, calculating a vegetation index, constructing a multi-modal data feature set, collecting farmland soil moisture content sample points, constructing a farmland soil moisture content integrated inversion model with residual error optimization, performing farmland soil moisture content inversion and performing farmland soil moisture content inversion result precision evaluation. According to the method, the characteristics and advantages of optical and thermal infrared multi-mode remote sensing data are effectively fused, the farmland soil moisture content integrated inversion model with the residual error optimization is constructed, and the precision and stability of high-resolution soil moisture content dynamic inversion are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated inversion of farmland soil moisture, and particularly to a fine integrated inversion method for farmland soil moisture that synergistically uses unmanned aerial vehicle (UAV) multi-spectral and thermal infrared images. Background Art

[0002] Soil water content, as an important indicator reflecting farmland soil moisture, plays a key role in the material and energy cycle of the farmland ecosystem and is an important variable in the research fields of agricultural yield increase, hydrological cycle, etc. Therefore, constructing an efficient and accurate inversion method for soil water content with high spatial resolution at the field scale is of great significance for improving crop irrigation management level, agricultural water resource utilization efficiency, crop growth and drought monitoring ability, etc.

[0003] Satellite remote sensing receives electromagnetic wave signals reflected or emitted by the earth's surface through a payload, and then calculates the information on the soil moisture status of the earth's surface in combination with an algorithm. It has achieved large-area and rapid monitoring of soil moisture to a certain extent, but there are still problems such as insufficient timeliness, poor resolution, and low accuracy. UAV remote sensing has the advantages of high timeliness, high resolution, and the ability to carry multiple types of sensors, and can effectively realize the dynamic monitoring of soil moisture at the field block scale. Research shows that comprehensively using UAV surface temperature, multi-spectral image data and their derived vegetation indices can better characterize soil moisture information and is conducive to the fine-grained and accurate inversion of farmland soil moisture. The remote sensing inversion of soil moisture information is affected by various factors such as vegetation cover, surface roughness, and soil type, and the relationship between soil water content and data characteristic variables is complex. Machine learning algorithms have obvious advantages in solving non-linearity, heteroscedasticity, etc. and are currently effective solutions. However, when agricultural scenarios are complex and multi-modal data are used in combination, conventional machine learning inversion algorithms often face problems such as insufficient inversion accuracy and weak stability. Therefore, it is necessary to propose a fine integrated inversion method for farmland soil moisture that synergistically uses UAV multi-spectral and thermal infrared images to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a fine integrated inversion method for farmland soil moisture that synergistically uses UAV multi-spectral and thermal infrared images to solve the problems that conventional machine learning inversion algorithms often face, such as insufficient inversion accuracy and weak stability.

[0005] The present invention provides a fine integrated inversion method for farmland soil moisture that synergistically uses UAV multi-spectral and thermal infrared images, including: Step 1, collecting multi-spectral image data and thermal infrared image data of a target area through a UAV; preprocessing the multi-spectral image data and thermal infrared image data; Step 2: Based on the preprocessed multi-spectral image data, calculate the Normalized Difference Vegetation Index (NDVI), Normalized Difference Red Edge Index (NDRE), Leaf Chlorophyll Index (LCI), and Green Band Normalized Difference Vegetation Index (GNDVI) respectively; Step 3: Based on the preprocessed multi-spectral image data and thermal infrared image data in Step 1, and the NDVI, NDRE, LCI, and GNDVI calculated in Step 2, use the data stacking method to construct a multi-modal data feature set for soil moisture status inversion; Step 4: Based on the principle of spatial uniform distribution, use the soil moisture monitoring system to synchronously and evenly select ground farmland points in an air-ground collaborative manner and measure the soil moisture content of the farmland. The measurement time is consistent with the acquisition time of the multi-spectral image data and thermal infrared image data in Step 1; combine with the multi-modal data feature set in Step 3 to construct a farmland soil moisture content sample set. The multi-modal data features in the sample set are used as independent variables for the input of the inversion model, and the corresponding farmland soil moisture content is used as the dependent variable for the input of the inversion model; divide the sample set into a training set and a test set; Step 5: Construct a residual-optimized integrated inversion model for farmland soil moisture; among them, in the first stage, based on three inversion algorithms of Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR), use the training set in Step 4 as the input, train the three inversion algorithms respectively, evaluate the trained inversion algorithms on the test set, and calculate the coefficient of determination; design weight coefficients with the coefficient of determination as the measurement index, and perform weighted summation integration on the inversion results of the three inversion algorithms to obtain the basic integrated inversion result of the soil moisture content; in the second stage, calculate the residual between the basic integrated inversion result in the training set and the true value of the soil moisture content, and use the Extreme Gradient Boosting Regression algorithm as the residual correction model to fit the independent variables and residuals in the training set; finally, sum the basic integrated inversion result in the first stage and the residuals generated by the residual correction model in the second stage to obtain the final farmland soil moisture content, which reflects the soil moisture status; Step 6: Use the residual-optimized integrated inversion model for farmland soil moisture constructed in Step 5, with the multi-modal data features of the target area as the input, to perform farmland soil moisture inversion.

[0006] Furthermore, in Step 1, the preprocessing of the multi-spectral image data includes band registration, image mosaicking, radiometric calibration, and standard deviation normalization; the preprocessing of the thermal infrared image data includes image mosaicking, bad value interpolation, and standard deviation normalization; after preprocessing the multi-spectral image data and thermal infrared image data, through image registration and image cropping operations, achieve spatial alignment between the two, and finally obtain the available image data covering the target area.

[0007] Further, in step one, the spatial resolutions of the multispectral image data and the thermal infrared image data are both 0.2 meters. Among them, the bands of the multispectral image data include the blue band, the green band, the red band, the red edge band, the first near-infrared band, and the second near-infrared band, a total of six bands.

[0008] Further, the wavelengths of the six bands are 450nm, 550nm, 660nm, 680nm, 840nm, and 900nm in sequence.

[0009] Further, in step two, the calculation formula is as follows:

[0010]

[0011]

[0012]

[0013] In the formula, is the normalized difference vegetation index, is the normalized difference red edge index, is the leaf area chlorophyll index, is the green band normalized difference vegetation index; 、 、 respectively represent the reflectance values of the red band, the green band, the red edge band, and the first near-infrared band of the UAV multispectral image data.

[0014] Further, in step five, the calculation method of the weight coefficient is as follows:

[0015] In the formula, represents the determination coefficient of the i th inversion algorithm, is the weight coefficient of the i th inversion algorithm when weighted summation integration.

[0016] Further, the method further includes: Step seven, based on the farmland soil moisture content results obtained by the integrated inversion model of farmland soil moisture based on residual optimization, calculate the evaluation indexes of determination coefficient, root mean square error, mean relative error, and mean absolute error, and evaluate the farmland soil moisture inversion ability of the fine integrated inversion method of farmland soil moisture by synergistic UAV multispectral and thermal infrared images.

[0017] Further, in step seven, the calculation formulas of the evaluation indexes of determination coefficient, root mean square error, mean relative error, and mean absolute error are as follows:

[0018]

[0019]

[0020]

[0021] In the formula, is the determination coefficient of the evaluation index, is the root mean square error, is the average relative error, is the average absolute error; , , respectively represent the inversion prediction value, true value and average value of the soil moisture content in farmland, and n is the number of samples; the closer the determination coefficient of the evaluation index is to 1, and the closer the root mean square error, average relative error and average absolute error are to 0, the higher the inversion accuracy of the model and the better the prediction performance.

[0022] The present invention has the following beneficial effects: The fine integrated inversion method for farmland soil moisture conditions by synergistically using unmanned aerial vehicle multi-spectral and thermal infrared images of the present invention constructs a residual-optimized integrated inversion model for farmland soil moisture conditions based on multi-spectral and thermal infrared image data, breaks through the limitations of weak dynamics and low spatial resolution in satellite remote sensing monitoring, and improves the accuracy and stability of soil moisture condition inversion. This method effectively integrates the characteristics and advantages of optical and thermal infrared multi-modal remote sensing data, and can realize low-cost, fast and accurate inversion of soil moisture conditions at the field block scale with high spatial resolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0024] Figure 1 is a flowchart of the fine integrated inversion method for farmland soil moisture conditions by synergistically using unmanned aerial vehicle multi-spectral and thermal infrared images of the present invention; Figure 2 is a diagram of the inversion result of soil moisture conditions in the field; Figure 3 is a diagram of the true value and inversion result of soil moisture content in farmland in the test set; Figure 4 is a scatter diagram of the true value and inversion result of soil moisture content in farmland in the test set. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention. The following will describe in detail the technical solutions provided by each embodiment of the present invention with reference to the drawings.

[0026] Please refer to Figure 1 , an embodiment of the present invention provides a fine integrated inversion method for farmland soil moisture by coordinating multi-spectral and thermal infrared images of unmanned aerial vehicles, including: Step 1: Collect multi-spectral image data and thermal infrared image data of the target area by an unmanned aerial vehicle; preprocess the multi-spectral image data and thermal infrared image data.

[0027] Specifically, during the period from 11 am to 3 pm on a certain day, use a DJI M350RTK unmanned aerial vehicle equipped with an MS600PRO multi-spectral and Zenmuse H20T thermal infrared imager to collect multi-spectral image data and thermal infrared image data of the target area respectively. The preprocessing of the multi-spectral image data includes band registration, image mosaicking, radiometric calibration, and standard deviation normalization; the preprocessing of the thermal infrared image data includes image mosaicking, bad value interpolation, and standard deviation normalization; after preprocessing the multi-spectral image data and thermal infrared image data, through image registration and image cropping operations, spatial alignment between the two is achieved, and finally available image data covering the target area is obtained.

[0028] The spatial resolution of both the multi-spectral image data and the thermal infrared image data is 0.2 meters. Among them, the bands of the multi-spectral image data include a blue band, a green band, a red band, a red edge band, a first near-infrared band, and a second near-infrared band, a total of six bands. Specifically, the wavelengths of the six bands are 450nm, 550nm, 660nm, 680nm, 840nm, and 900nm in sequence.

[0029] Step 2: Based on the preprocessed multi-spectral image data, calculate the Normalized Difference Vegetation Index (NDVI), Normalized Difference Red Edge (NDRE), Leaf Chlorophyll Index (LCI), and Green-band Normalized Vegetation Index (GNDVI) respectively.

[0030] Specifically, the calculation formula is as follows:

[0031]

[0032]

[0033]

[0034] In the formula, is the normalized difference vegetation index, is the normalized difference red-edge index, is the leaf chlorophyll index, is the green-band normalized difference vegetation index; , , respectively represent the reflectance values of the red band, green band, red-edge band and first near-infrared band of the UAV multispectral image data.

[0035] Step 3: Based on the multispectral image data, thermal infrared image data preprocessed in Step 1, and the normalized difference vegetation index, normalized difference red-edge index, leaf chlorophyll index and green-band normalized difference vegetation index calculated in Step 2, a multimodal data feature set for soil moisture status inversion is constructed by means of data stacking.

[0036] Step 4: Based on the principle of spatial uniform distribution, using the soil moisture monitoring system, the ground farmland points are synchronously and evenly selected in an air-ground collaborative manner and the soil moisture content of the farmland is measured. The measurement time is the same as the acquisition time of the multispectral image data and thermal infrared image data in Step 1; combined with the multimodal data feature set in Step 3, a farmland soil moisture content sample set is constructed. The multimodal data features in the sample set are used as the independent variables input to the inversion model, and the corresponding farmland soil moisture content is used as the dependent variable input to the inversion model; the sample set is divided into a training set and a test set.

[0037] In this embodiment, the soil moisture content value is measured by using the soil moisture monitoring system MiniTrase. After data processing, a total of 93 farmland soil moisture content sample points are obtained. The sample set is divided into a training set and a test set according to the ratio of 7:3, which are respectively used for the training and accuracy evaluation of the farmland soil moisture status inversion model. MiniTrase is a portable measuring device for soil moisture status based on the principle of time domain reflectometry (TDR); the soil moisture content refers to the soil volume moisture content, and the unit is %.

[0038] Step 5: Construct a residual-optimized integrated inversion model for farmland soil moisture status; Among them, in the first stage, based on three inversion algorithms of Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR), using the training set in step four as the input, train the three inversion algorithms respectively, evaluate the trained inversion algorithms on the test set, and calculate the coefficient of determination (R 2 ); design the weight coefficient with R 2 as the measurement index, perform weighted summation integration on the inversion results of the three inversion algorithms to obtain the basic integrated inversion result of soil moisture content; In the second stage, calculate the residual between the basic integrated inversion result in the training set and the true value of soil moisture content, and use the XGBoost regression algorithm as the residual correction model to fit the independent variable and the residual in the training set; finally, add the basic integrated inversion result in the first stage and the residual generated by the residual correction model in the second stage to obtain the final farmland soil moisture content, reflecting the soil moisture condition.

[0039] The calculation method of the weight coefficient is as follows:

[0040] In the formula, represents the coefficient of determination of the i th inversion algorithm, and is the weight coefficient of the i th inversion algorithm during weighted summation integration.

[0041] Step six, using the integrated inversion model of farmland soil moisture optimized by residuals constructed in step five, with the multi-modal data features of the target area as the input, perform inversion of farmland soil moisture. The field soil moisture inversion result is as Figure 2 shown.

[0042] Step seven, based on the farmland soil moisture content result obtained by the integrated inversion model of farmland soil moisture optimized by residuals, calculate the evaluation indexes of coefficient of determination (R 2 ), root mean square error (RMSE), mean relative error (MRE), and mean absolute error (MAE) to evaluate the farmland soil moisture inversion ability of the fine integrated inversion method of farmland soil moisture combining unmanned aerial vehicle multi-spectral and thermal infrared images.

[0043] In this embodiment, the R 2 , RMSE, MRE, and MAE of the farmland soil moisture inversion result are 0.80, 2.41, 0.05, and 1.46 respectively. The comparison with the true value of soil moisture is shown in Figure 3 and Figure 4 .

[0044] Specifically, the calculation formulas for the evaluation indicators of the coefficient of determination, root mean square error, mean relative error, and mean absolute error are as follows:

[0045]

[0046]

[0047]

[0048] In the formula, is the coefficient of determination of the evaluation indicator, is the root mean square error, is the mean relative error, is the mean absolute error; , , respectively represent the inversion prediction value, true value, and average value of the soil moisture content in farmland, and n is the number of samples; the closer the coefficient of determination of the evaluation indicator is to 1, and the closer the root mean square error, mean relative error, and mean absolute error are to 0, the higher the inversion accuracy of the model and the better the prediction performance.

[0049] In summary, the present invention utilizes the characteristics of high timeliness, high resolution, and the ability to carry multiple types of sensors of UAV remote sensing, effectively overcoming the deficiencies of weak dynamics and low spatial resolution existing in satellite remote sensing soil moisture monitoring; comprehensively utilizes UAV surface temperature, multi-spectral image data, and their derived vegetation indices to enhance the characterization of soil moisture information, constructs a residual-optimized integrated inversion model for farmland soil moisture, and improves the accuracy and stability of high-resolution soil moisture dynamic inversion. Aiming at the problems of insufficient inversion accuracy and weak stability often faced by conventional machine learning inversion algorithms, the present invention proposes a fine integrated inversion method for farmland soil moisture by coordinating UAV multi-spectral and thermal infrared images, comprehensively utilizes multi-modal data such as UAV surface temperature, multi-spectral images, and their derived vegetation indices, and based on RF, SVR, and XGBoost algorithms, constructs a residual-optimized integrated inversion model for farmland soil moisture, and improves the accuracy and stability of high-resolution soil moisture dynamic inversion.

[0050] The embodiments of the present invention described above do not constitute a limitation on the protection scope of the present invention.

Claims

1. A method for fine integrated inversion of farmland soil moisture using cooperative UAV multispectral and thermal infrared images, characterized in that: include: Step 1: Collect multispectral image data and thermal infrared image data of the target area through drones; Preprocessing the multispectral image data and thermal infrared image data; Step 2: Based on the preprocessed multispectral image data, the normalized difference vegetation index, the normalized difference red edge index, the leaf chlorophyll index and the green band normalized difference vegetation index are calculated respectively; Step 3: Based on the multispectral image data and thermal infrared image data preprocessed in step 1 and the normalized difference vegetation index, normalized difference red edge index, leaf chlorophyll index and green band normalized difference vegetation index calculated in step 2, a multimodal data feature set for soil moisture condition inversion is constructed by data stacking; Step 4: Based on the principle of uniform spatial distribution, the soil moisture monitoring system is used to synchronously and evenly select ground farmland points and measure the farmland soil moisture content in an air-ground collaborative manner. The measurement time is consistent with the acquisition time of the multispectral image data and thermal infrared image data in step 1. Combined with the multimodal data feature set in step 3, a farmland soil moisture sample set is constructed. The multimodal data features in the sample set are used as independent variables for the inversion model input, and the corresponding farmland soil moisture content is used as the dependent variable for the inversion model input. The sample set is divided into a training set and a test set. Step five, construct an integrated inversion model of farmland soil moisture with residual optimization; wherein, in the first stage, based on the three inversion algorithms of random forest, extreme gradient boosting and support vector regression, the training set in step four is used as input, and the three inversion algorithms are trained respectively, and the trained inversion algorithms are evaluated on the test set to calculate the determination coefficient; the weight coefficient is designed with the determination coefficient as the measurement indicator, and the inversion results of the three inversion algorithms are weighted summed and integrated to obtain the basic integrated inversion results of soil moisture content; in the second stage, the residual of the basic integrated inversion result in the training set and the true value of the soil moisture content are calculated, and the extreme gradient boosting regression algorithm is used as the residual correction model to fit the independent variables and residuals in the training set; finally, the basic integrated inversion result in the first stage and the residual generated by the residual correction model in the second stage are added to obtain the final farmland soil moisture content, reflecting the soil moisture condition; Step six, using the residual optimized farmland soil moisture integrated inversion model constructed in step five, and taking the multimodal data characteristics of the target area as input, perform farmland soil moisture inversion.

2. The method for fine integrated inversion of farmland soil moisture conditions based on cooperative UAV multispectral and thermal infrared images as claimed in claim 1, characterized in that: In step one, the preprocessing of multispectral image data includes band registration, image stitching, radiometric determination and standard deviation normalization; the preprocessing of thermal infrared image data includes image stitching, bad value interpolation and standard deviation normalization; after preprocessing the multispectral image data and thermal infrared image data, spatial alignment between the two is achieved through image registration and image cropping operations, and finally usable image data covering the target area is obtained.

3. The method for fine integrated inversion of farmland soil moisture conditions based on cooperative UAV multispectral and thermal infrared images as claimed in claim 1, characterized in that: In step one, the spatial resolution of the multispectral image data and the thermal infrared image data are both 0.2 meters, wherein the bands of the multispectral image data include a blue light band, a green light band, a red light band, a red edge band, a first near infrared band, and a second near infrared band, totaling six bands.

4. The method for fine integrated inversion of farmland soil moisture conditions based on cooperative UAV multispectral and thermal infrared images as claimed in claim 3 is characterized in that: The wavelengths of the six bands are 450nm, 550nm, 660nm, 680nm, 840nm and 900nm respectively.

5. The method for fine integrated inversion of farmland soil moisture conditions based on cooperative UAV multispectral and thermal infrared images as claimed in claim 4, characterized in that: In step 2, the calculation formula is as follows: In the formula, is the normalized difference vegetation index, is the normalized difference red edge index, is the leaf chlorophyll index, is the normalized difference vegetation index of the green band; , , They respectively represent the reflectance values ​​of the red band, green band, red edge band and first near-infrared band of the UAV multispectral image data.

6. The method for fine integrated inversion of farmland soil moisture conditions based on cooperative UAV multispectral and thermal infrared images as claimed in claim 1, characterized in that: In step 5, the weight coefficient is calculated as follows: In the formula, Indicates i The determination coefficient of the inversion algorithm is When the weighted sum is integrated i The weight coefficient of the inversion algorithm.

7. The method for fine integrated inversion of farmland soil moisture conditions based on cooperative UAV multispectral and thermal infrared images as claimed in claim 1, characterized in that: Also includes: Step seven, based on the farmland soil moisture results obtained by the residual optimized farmland soil moisture integrated inversion model, calculate the evaluation index determination coefficient, root mean square error, mean relative error and mean absolute error, and evaluate the farmland soil moisture inversion capability of the farmland soil moisture fine integrated inversion method based on collaborative UAV multispectral and thermal infrared images.

8. The method for fine integrated inversion of farmland soil moisture conditions based on cooperative UAV multispectral and thermal infrared images as claimed in claim 7, characterized in that: In step 7, the calculation formulas for the evaluation index determination coefficient, root mean square error, mean relative error, and mean absolute error are as follows: In the formula, is the determination coefficient of the evaluation index, is the root mean square error, is the average relative error, is the mean absolute error; , , They represent the inversion prediction value, true value and average value of farmland soil moisture content respectively, and n is the number of samples; the closer the determination coefficient of the evaluation index is to 1, and the closer the root mean square error, mean relative error and mean absolute error are to 0, the higher the inversion accuracy of the model and the better the prediction performance.

Images