Red soil organic matter retrieval method based on unmanned aerial vehicle hyperspectral image and spectral index
By combining UAV hyperspectral imagery and spectral indices, the problem of low accuracy in soil organic matter inversion has been solved, achieving efficient and accurate soil organic matter inversion, which is suitable for soil quality monitoring in complex terrain areas.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KUNMING UNIV OF SCI & TECH
- Filing Date
- 2026-03-26
- Publication Date
- 2026-06-05
AI Technical Summary
In existing technologies, satellite remote sensing data has low resolution and diverse land types, resulting in low accuracy of soil organic matter inversion and insufficient utilization of spectral information.
A method combining UAV hyperspectral imagery and spectral indices was adopted. Hyperspectral images were acquired through primary and secondary preprocessing, spectral indices were constructed, highly correlated spectral feature variables were screened, and support vector regression models were used to optimize parameters through the whale optimization algorithm to retrieve organic matter in red soil.
It improves the accuracy and efficiency of soil organic matter inversion, reduces costs, is applicable to soil quality monitoring in complex terrain areas, and enhances the model's generalization ability and prediction accuracy.
Smart Images
Figure CN121899047B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for inverting organic matter in red soil based on UAV hyperspectral imagery and spectral indices, belonging to the field of agricultural soil organic matter content measurement technology. Background Technology
[0002] Soil is a crucial component of the geographical environment and a natural resource essential for human survival. Soil organic matter consists of carbon-containing organic substances, primarily including plant and animal remains, microbial matter, and the organic matter resulting from their decomposition or synthesis. Humus in soil organic matter can improve soil permeability and porosity. Furthermore, the complexes formed by humus with ions such as iron, phosphorus, and aluminum can inhibit the formation of insoluble phosphates, thereby increasing the available nutrients in the soil and serving as a key indicator of soil fertility. Accurate measurement of soil organic matter content is of great significance for evaluating soil fertility.
[0003] Patent CN120853040A discloses a method for soil organic matter inversion based on multi-pixel combination of satellite imagery. It mainly constructs a soil organic matter inversion model using satellite remote sensing data and a number of sampling verification points. Finally, based on the predicted values of the verification points, spatial interpolation is performed on the target area to obtain a spatial distribution map of soil organic matter. Compared with traditional laboratory measurement methods, this method can achieve rapid inversion of soil organic matter content in large target areas, improving measurement efficiency and reducing measurement costs. However, satellite imagery data has low resolution, diverse land types, and a large number of mixed pixels, which is not conducive to high-precision organic matter inversion. Furthermore, the accuracy of the spatial interpolation-based method depends on the density and accuracy of the verification points, and it does not fully utilize the spectral and geographic information of each pixel.
[0004] In recent years, unmanned aerial vehicles (UAVs), as a highly flexible remote sensing platform, have been widely used, and the use of UAV remote sensing technology for agricultural soil organic matter inversion has become a major focus. Therefore, how to utilize UAV remote sensing technology to improve the accuracy and efficiency of agricultural soil organic matter content inversion is a pressing technical problem to be solved in current soil organic matter measurement. Summary of the Invention
[0005] The purpose of this invention is to provide a method for inverting organic matter in red soil based on UAV hyperspectral imagery and spectral indices, aiming to solve the technical problems of insufficient utilization of spectral information and low accuracy of spatial inversion of soil organic matter in existing technologies.
[0006] To achieve the above objectives, the technical solution of the present invention is: a method for inverting organic matter in red soil based on UAV hyperspectral imagery and spectral indices, comprising the following steps:
[0007] Step 1: Acquire UAV hyperspectral images and perform initial preprocessing; wherein, the initial preprocessing includes image stitching, radiometric correction, orthorectification and reflectance data extraction in sequence;
[0008] Step 2: Perform secondary preprocessing on the UAV hyperspectral image after the initial preprocessing; wherein, the secondary preprocessing includes first-order differentiation, second-order differentiation, standard normal variable transformation and multivariate scattering correction;
[0009] Step 3: Obtain the spectral indices of the UAV hyperspectral image after secondary preprocessing; wherein, the spectral indices include two-band spectral indices and three-band spectral indices;
[0010] Step 4: Obtain the Pearson correlation coefficient between each spectral index and the organic matter content of red soil in the UAV hyperspectral image after secondary preprocessing. Sort the spectral indices according to the absolute value of the Pearson correlation coefficient and select the spectral indices with high correlation as spectral feature variables. The spectral indices with high correlation are those whose absolute value of the Pearson correlation coefficient is greater than a preset threshold.
[0011] Step 5: Construct a red soil organic matter inversion model and perform parameter optimization to obtain an optimized red soil organic matter inversion model;
[0012] Step 6: Input the spectral feature variables into the optimized red soil organic matter inversion model to obtain the red soil organic matter inversion results.
[0013] Optionally, the reflectance data extraction specifically involves:
[0014] Acquire multi-band reflectance data and corresponding spatial reference information of orthorectified UAV hyperspectral images;
[0015] Based on the spatial reference information, establish the conversion relationship between image pixel coordinates and geographic coordinates;
[0016] By traversing the pixel positions in the orthorectified UAV hyperspectral image, the reflectance values of the corresponding pixels in each band are extracted to form the spectral reflectance vector of each pixel.
[0017] Based on the row and column positions of the pixels in the orthorectified UAV hyperspectral image, the corresponding geographic coordinates are calculated using the spatial reference information.
[0018] The geographic coordinates and corresponding multi-band reflectance data of each pixel are recorded and stored to complete the reflectance data extraction.
[0019] Optionally, the two-band spectral indices include a difference index, a ratio index, a sum index, and a normalized index, specifically:
[0020]
[0021]
[0022]
[0023]
[0024] in, It is the difference index. It is the ratio index. For sum and index, The normalized index, , The first one in the wavelength range Band, First Reflectivity of the band.
[0025] Optionally, the three-band spectral indices include six types of three-band spectral indices, specifically:
[0026]
[0027]
[0028]
[0029]
[0030]
[0031]
[0032] in, , , , , and These represent the three-band spectral indices with coded serial numbers 1, 2, 3, 4, 5, and 6, respectively. , , The first one in the wavelength range Band, First Band, First Reflectivity of the band.
[0033] Optionally, the red soil organic matter inversion model is a support vector regression model.
[0034] Optionally, the parameter optimization employs the whale optimization algorithm, specifically:
[0035] Each individual whale is represented as a set of spectral characteristic variables. ,in, As a penalty factor, For insensitive loss function parameters, The parameters are set as kernel function parameters, and the spectral feature variables are randomly initialized within a preset parameter range;
[0036] The mean square error of the five-fold cross-validation of the support vector regression model corresponding to the spectral feature variables on the training sample set was used as the fitness evaluation index to calculate the fitness value of each individual whale.
[0037] In each iteration, the position of the whale is updated based on the distance between the individual whale and the current global best individual, and the dynamic optimization of the support vector regression model parameters is achieved by gradually narrowing the search space.
[0038] When the maximum number of iterations is reached or the optimal fitness value no longer changes during multiple consecutive iterations, the corresponding global optimal parameter combination is output as the optimal parameters of the support vector regression model, thereby completing the parameter optimization; wherein, the multiple times refers to a pre-set number of times.
[0039] The beneficial effects of this invention are:
[0040] (1) This invention uses a drone carrying a camera to quickly and flexibly acquire hyperspectral image data of the target area, and the appropriate time can be selected according to the weather and phenological period. Compared with satellite image data, drones have more flexible acquisition time and area and higher resolution; compared with traditional field laboratory measurements, they can acquire a larger target area, saving time and economic costs.
[0041] (2) This invention, by combining UAV hyperspectral imagery and spectral indices, can fully mine and utilize the spectral information of pixels, and use a support vector regression model optimized by the whale optimization algorithm to retrieve soil organic matter (SOM) in red soil. Compared with traditional remote sensing soil organic matter estimation methods, this invention introduces an optimization algorithm with higher accuracy, which effectively improves the accuracy of soil organic matter retrieval and the generalization ability of the model.
[0042] (3) In this invention, the parameters of the SVR model are dynamically optimized through the iterative process of the whale optimization algorithm, which has better adaptability and higher flexibility than the traditional static parameter selection optimization method. The fitness of each individual whale is obtained based on the mean square error of the five-fold cross-validation of the model, which can comprehensively evaluate the model performance and dynamically adjust the model parameters.
[0043] (4) Through precise optimization and training of model parameters, this invention can extract soil organic matter information from hyperspectral image data more quickly, thereby achieving an efficient and accurate inversion process. Compared with existing models, this invention has significant advantages in data processing time and prediction accuracy. Attached Figure Description
[0044] Figure 1 This is a flowchart of the steps of the present invention;
[0045] Figure 2 This is a Pearson correlation coefficient diagram showing the relationship between the two-band spectral indices of this invention and the organic matter content of red soil. Figure 2 (a) is a Pearson correlation coefficient diagram between the NDI index and the organic matter content of red soil. Figure 2 (b) is a Pearson correlation coefficient diagram between the index DI and the organic matter content of red soil. Figure 2 (c) is a Pearson correlation coefficient diagram between the index SI and the organic matter content of red soil. Figure 2 (d) is a Pearson correlation coefficient diagram between the index RI and the organic matter content of red soil;
[0046] Figure 3 This is a Pearson correlation coefficient diagram showing the relationship between the three-band spectral indices of this invention and the organic matter content of red soil. Figure 3 (a) is a Pearson correlation coefficient diagram between the index TBI1 and the organic matter content of red soil. Figure 3 (b) is a Pearson correlation coefficient diagram between the index TBI2 and the organic matter content of red soil. Figure 3 (c) is a Pearson correlation coefficient diagram between the index TBI3 and the organic matter content of red soil. Figure 3 (d) is a Pearson correlation coefficient diagram between the index TBI4 and the organic matter content of red soil. Figure 3 (e) is a Pearson correlation coefficient diagram between the index TBI5 and the organic matter content of red soil. Figure 3 (f) is a Pearson correlation coefficient diagram between the index TBI6 and the organic matter content of red soil;
[0047] Figure 4 This is a spatial distribution map of red soil organic matter content based on the optimized red soil organic matter inversion model of this invention. Detailed Implementation
[0048] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0049] Example 1: As Figure 1 As shown, a method for inverting organic matter in red soil based on UAV hyperspectral imagery and spectral indices is presented. This method involves confirming and debugging the UAV and hyperspectral camera to acquire hyperspectral images of the target area. After preprocessing and spectral data extraction, spectral indices are constructed, and feature variables highly correlated with organic matter content are selected. Next, a support vector regression (SVR) model is selected for red soil organic matter inversion, and the model parameters are optimized using a whale optimization algorithm. Finally, red soil organic matter is inverted under the optimal model parameters, including the following steps:
[0050] Step 1: Acquire UAV hyperspectral images and perform initial preprocessing; wherein, the initial preprocessing includes image stitching, radiometric correction, orthorectification and reflectance data extraction in sequence;
[0051] Optionally, this embodiment uses a compatible Matrice300RTK quadcopter UAV and a HY-9010-L hyperspectral imager to acquire hyperspectral images of the UAV, and performs connection and testing, sets a flight path, and enables the UAV to carry the hyperspectral camera to acquire hyperspectral images of the target area along the prescribed flight path.
[0052] Furthermore, before acquiring hyperspectral imagery from the UAV, pre-flight checks, tests, and parameter settings are performed to reduce risks during data acquisition. Combining the UAV with a hyperspectral camera enables automated acquisition of hyperspectral imagery, improving both the efficiency and security of data collection.
[0053] Specifically, in this embodiment, while acquiring hyperspectral images from the UAV, 69 red soil samples were collected in the target area and their corresponding geographic coordinates were recorded. After processing the samples, the soil organic matter content of the samples was determined using the potassium bibutate-sulfuric acid solution oxidation method, which was used for subsequent model construction and performance verification.
[0054] Optionally, the image stitching specifically involves:
[0055] The hyperspectral imager acquires multiple individual images. The MAXSPEX STUDIO image processing software that comes with the hyperspectral imager is used to stitch all the acquired individual images together into a complete image by flight strip.
[0056] Optionally, the radiation correction specifically includes:
[0057] To obtain accurate reflectance and eliminate radiation errors caused by differences in atmospheric environment, atmospheric scattering and absorption, the spectral image after image stitching is a radiance value. Radiometric correction is performed on the image to convert the radiance value into reflectance.
[0058] Optionally, the orthorectification specifically includes:
[0059] To eliminate image distortion caused by topographic relief, elevation and geographic coordinate information were added to the image, ultimately resulting in a hyperspectral image of the target area.
[0060] Optionally, the reflectance data extraction involves extracting the corresponding spectral information and geographic coordinate information pixel by pixel from the orthorectified UAV hyperspectral image, reading the reflectance values of each pixel in the image at different bands, and establishing the transformation relationship between pixel row and column coordinates and geographic coordinates based on the spatial reference information of the image, thereby obtaining the multi-band reflectance data and geographic coordinate information corresponding to each pixel, specifically:
[0061] Acquire multi-band reflectance data and corresponding spatial reference information of orthorectified UAV hyperspectral images;
[0062] Based on the spatial reference information, establish the conversion relationship between image pixel coordinates and geographic coordinates;
[0063] By traversing the pixel positions in the orthorectified UAV hyperspectral image, the reflectance values of the corresponding pixels in each band are extracted to form the spectral reflectance vector of each pixel.
[0064] Based on the row and column positions of the pixels in the orthorectified UAV hyperspectral image, the corresponding geographic coordinates (longitude and latitude) are calculated using the spatial reference information.
[0065] The geographic coordinates and corresponding multi-band reflectance data of each pixel are recorded and stored to complete the reflectance data extraction.
[0066] It is understandable that, through the above-described method for extracting reflectance data, this embodiment can achieve automatic extraction of the spectral reflectance information and geographic coordinate information of each pixel in a hyperspectral image.
[0067] Step 2: Perform secondary preprocessing on the UAV hyperspectral image after the initial preprocessing; wherein, the secondary preprocessing includes first-order differentiation, second-order differentiation, standard normal variable transformation and multivariate scattering correction;
[0068] It should be understood that the first-order derivative specifically involves performing a first-order derivative operation on the hyperspectral reflectance data along the wavelength direction to enhance the variation characteristics of the spectral curve, reduce background influence, and highlight sensitive band information related to the organic matter content of red soil; the second-order derivative specifically involves further performing a second-order derivative processing on the hyperspectral data based on the first-order derivative to enhance the detailed features of the spectral curve and improve the ability to resolve weak absorption features; the standard normal variable transformation specifically involves centering the mean and normalizing the standard deviation of each spectrum to reduce the influence caused by differences in illumination conditions and sample scattering characteristics; the multivariate scattering correction specifically involves correcting the hyperspectral reflectance data to eliminate scattering effects caused by factors such as particle size differences and surface roughness, thereby improving the consistency and comparability of the spectral data.
[0069] It is understood that in this embodiment, the secondary preprocessing reduces the impact of noise interference and non-target factors on hyperspectral images, and enhances the response relationship between spectral features and red soil organic matter content.
[0070] Step 3: Obtain the spectral indices of the UAV hyperspectral image after secondary preprocessing; wherein, the spectral indices include two-band spectral indices and three-band spectral indices;
[0071] Optionally, the two-band spectral indices include a difference index, a ratio index, a sum index, and a normalized index, specifically:
[0072]
[0073]
[0074]
[0075]
[0076] in, It is the difference index. It is the ratio index. For sum and index, The normalized index, , The first one in the wavelength range Band, First Reflectivity of the band.
[0077] Optionally, the three-band spectral indices include six types of three-band spectral indices, specifically:
[0078]
[0079]
[0080]
[0081]
[0082]
[0083]
[0084] in, , , , , and These represent the three-band spectral indices with coded serial numbers 1, 2, 3, 4, 5, and 6, respectively. , , The first one in the wavelength range Band, First Band, First Reflectivity of the band.
[0085] Understandably, traditional soil organic matter retrieval studies typically utilize only single-band or simple spectral indices for modeling, making it difficult to fully extract the potential spectral information from hyperspectral data. Therefore, this embodiment constructs a two-dimensional and three-dimensional spectral index system based on UAV hyperspectral imagery. By combining two-dimensional and three-dimensional spectral indices, and using a multi-band combination approach, the sensitivity of the spectrum to changes in soil organic matter is enhanced, thereby improving the ability to express feature information.
[0086] It is understandable that while UAV hyperspectral imagery contains a large amount of continuous band information, the raw spectral data often suffers from noise, redundant information, and strong band correlations, making it difficult to fully utilize the effective spectral information and thus affecting the accuracy of soil organic matter inversion models. Therefore, this embodiment enhances the sensitivity of the spectrum to changes in soil organic matter and improves the ability to express effective feature information through the spectral preprocessing and spectral index construction methods in steps 1-3.
[0087] Optionally, in this embodiment, based on the UAV hyperspectral image after secondary preprocessing, two-band spectral indices and three-band spectral indices are calculated respectively to form a spectral feature set containing multiple spectral indices, which serves as input data for subsequent optimization of spectral feature variables and construction of the red soil organic matter inversion model.
[0088] Step 4: Obtain the Pearson correlation coefficient between each spectral index and the organic matter content of red soil in the UAV hyperspectral image after secondary preprocessing. Sort the spectral indices according to the absolute value of the Pearson correlation coefficient and select the spectral indices with high correlation as spectral feature variables. The spectral indices with high correlation are those whose absolute value of the Pearson correlation coefficient is greater than a preset threshold.
[0089] Specifically, in this embodiment, the Pearson correlation coefficient is used to quantitatively evaluate the linear correlation between each spectral index and the organic matter content of red soil. The expression for the Pearson correlation coefficient is as follows:
[0090]
[0091] in, The Pearson correlation coefficient represents the relationship between spectral indices and organic matter content in red soil. Indicates the first The spectral index value corresponding to each sample Indicates the first The measured organic matter content of red soil corresponding to each sample This represents the average value of the spectral index. This represents the average organic matter content of red soil. Indicates the number of samples.
[0092] Furthermore, the Pearson correlation coefficients of 10 spectral indices (including two-band and three-band indices) with the organic matter content of red soil under different band combinations were calculated. The results were then visualized using two-dimensional and three-dimensional Pearson correlation coefficient distribution maps, such as... Figure 2 As shown, the distribution of Pearson correlation coefficients between different two-band spectral indices and organic matter content in red soil under various band combinations is illustrated. The color changes in the figure reflect the strength of the correlation between the corresponding indices and organic matter content under different band combinations. Figure 3 The distribution of Pearson correlation coefficients for different three-band spectral indices in the three-dimensional band combination space is shown, which is used to characterize the sensitivity of the three-band combination to the organic matter content of red soil.
[0093] Through the Figure 2 and Figure 3 A comprehensive analysis of the Pearson correlation coefficient distribution results shows that different spectral indices respond significantly differently to the organic matter content of red soil under different band combinations. Some band combinations show high correlations, while the other combinations show weak correlations.
[0094] Furthermore, in this embodiment, all spectral index variables are sorted according to the principle of Pearson correlation coefficient absolute value from high to low, and finally the 10 spectral feature variables with the largest correlation coefficient absolute value are selected as input features of the red soil organic matter inversion model.
[0095] It is understandable that this embodiment, through the feature selection process, not only ensures that the input variables have a strong characterization ability for the organic matter content of red soil, but also effectively reduces feature redundancy, providing a reliable data foundation for the subsequent construction of the inversion model and parameter optimization.
[0096] Step 5: Construct a red soil organic matter inversion model and perform parameter optimization to obtain an optimized red soil organic matter inversion model;
[0097] Optionally, the red soil organic matter inversion model is a support vector regression (SVR) model;
[0098] Optionally, the parameter optimization employs a whale optimization algorithm, which is a swarm intelligence optimization algorithm simulating the hunting behavior of humpback whale groups. It achieves the search for the global optimum through prey-encircling behavior, spiral hunting behavior, and random search behavior, specifically:
[0099] Each individual whale is represented as a set of spectral characteristic variables. ,in, As a penalty factor, For insensitive loss function parameters, The parameters are set as kernel function parameters, and the spectral feature variables are randomly initialized within a preset parameter range;
[0100] The mean square error of the five-fold cross-validation of the support vector regression model corresponding to the spectral feature variables on the training sample set was used as the fitness evaluation index to calculate the fitness value of each individual whale.
[0101] In each iteration, the position of the whale is updated based on the distance between the individual whale and the current global best individual, and the dynamic optimization of the support vector regression model parameters is achieved by gradually narrowing the search space.
[0102] When the maximum number of iterations is reached or the optimal fitness value no longer changes during multiple consecutive iterations, the corresponding global optimal parameter combination is output as the optimal parameters of the support vector regression model, thereby completing the parameter optimization; wherein, the multiple times refers to a pre-set number of times.
[0103] Understandably, traditional SVR model parameters are often determined based on experience or simple search methods, which can easily lead to insufficient model generalization ability. Therefore, this embodiment introduces the whale optimization algorithm to perform a global optimization search for the model parameters, enabling the model parameters to be determined within a more optimal range, thereby improving the model's prediction accuracy and stability.
[0104] Step 6: Input the spectral feature variables into the optimized red soil organic matter inversion model to obtain the red soil organic matter inversion results.
[0105] Specifically, the SVR model is used to establish a nonlinear regression relationship, and the parameters of the SVR model are optimized by the Whale Optimization Algorithm (WOA) to improve the model's prediction accuracy and stability, resulting in the support vector regression model (WOA_SVR) optimized by the Whale Optimization Algorithm.
[0106] Based on the specific implementation details, the effectiveness of the technical solution of the present invention will be demonstrated through experiments.
[0107] Specifically, this experiment uses the least squares regression (PLSR) model as a comparison model to verify the predictive performance of different modeling methods in the inversion of organic matter in red soil, thereby evaluating the effectiveness of the present invention.
[0108] Furthermore, in this experiment, the population size of the whale optimization algorithm was set to 20, the maximum number of iterations was set to 30, and the range of model parameters was: , , When the maximum number of iterations is reached or the optimal fitness value no longer changes during multiple consecutive iterations, the corresponding globally optimal parameter combination is output as the optimal parameters for the support vector regression model. To verify the inversion effect of this invention, the coefficient of determination (C) is selected in this experiment. 2 ), root mean square error ( RMSE ) and residual prediction bias ratio ( RPD As a performance evaluation metric for the model, the inversion results of the training set and the test set are comprehensively evaluated. The experimental results are shown in Table 1.
[0109] Table 1: Experimental results with different pretreatments and models
[0110]
[0111] As shown in Table 1, under different spectral preprocessing conditions, the support vector regression model WOA_SVR optimized by the whale optimization algorithm exhibits better overall inversion performance. Especially under second-order differential preprocessing, the WOA_SVR model achieves a determination coefficient of 0.721 on the test set, reduces the root mean square error to 7.606, and improves the residual prediction bias ratio to 1.936, demonstrating better prediction accuracy and stability. Therefore, this invention, by introducing the whale optimization algorithm for global optimization of model parameters, can effectively improve the prediction performance of the red soil organic matter inversion model, enhance the model's generalization ability and robustness, and verify the effectiveness and practicality of the method proposed in this invention.
[0112] Furthermore, the optimized red soil organic matter inversion model is used to predict the red soil organic matter in the target area. Coordinate correspondence is performed based on coordinate system information, and the organic matter content data for each element in the target area is displayed in chart form. For example... Figure 4 As shown, the red soil organic matter retrieved by the method of the present invention effectively reflects the spatial distribution of organic matter.
[0113] In summary, addressing the issues of existing technologies that often focus on single spectral indices or single models, high dimensionality and redundancy in hyperspectral data, the inability of single spectral indices to reflect changes in soil organic matter, and the instability of traditional model parameter selection, this invention integrates UAV hyperspectral data acquisition, spectral preprocessing, spectral index construction, feature selection, machine learning modeling, and optimization algorithms to form a complete method for red soil organic matter inversion. Through the synergistic effect of each step, it can fully mine the spectral information in hyperspectral images, improve the accuracy and stability of soil organic matter inversion, reduce the cost of manual sampling, and is suitable for soil quality monitoring in complex terrain areas such as plateaus and mountains.
[0114] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. A method for inverting organic matter in red soil based on UAV hyperspectral imagery and spectral indices, characterized in that, The method includes the following steps: Step 1: Acquire UAV hyperspectral images and perform initial preprocessing; wherein, the initial preprocessing includes image stitching, radiometric correction, orthorectification and reflectance data extraction in sequence; Step 2: Perform secondary preprocessing on the UAV hyperspectral image after the initial preprocessing; wherein, the secondary preprocessing includes first-order differentiation, second-order differentiation, standard normal variable transformation and multivariate scattering correction; Step 3: Obtain the spectral indices of the UAV hyperspectral image after secondary preprocessing; wherein, the spectral indices include two-band spectral indices and three-band spectral indices; Step 4: Obtain the Pearson correlation coefficient between each spectral index and the organic matter content of red soil in the UAV hyperspectral image after secondary preprocessing. Sort the spectral indices according to the absolute value of the Pearson correlation coefficient and select the spectral indices with high correlation as spectral feature variables. The spectral indices with high correlation are those whose absolute value of the Pearson correlation coefficient is greater than a preset threshold. Step 5: Construct a red soil organic matter inversion model and perform parameter optimization to obtain an optimized red soil organic matter inversion model; Step 6: Input the spectral feature variables into the optimized red soil organic matter inversion model to obtain the red soil organic matter inversion results; The red soil organic matter inversion model is a support vector regression model; The parameter optimization employs the whale optimization algorithm, specifically: Each individual whale is represented as a set of spectral characteristic variables. ,in, As a penalty factor, For insensitive loss function parameters, The parameters are set as kernel function parameters, and the spectral feature variables are randomly initialized within a preset parameter range; The mean square error of the five-fold cross-validation of the support vector regression model corresponding to the spectral feature variables on the training sample set was used as the fitness evaluation index to calculate the fitness value of each individual whale. In each iteration, the position of the whale is updated based on the distance between the individual whale and the current global best individual, and the dynamic optimization of the support vector regression model parameters is achieved by gradually narrowing the search space. When the maximum number of iterations is reached or the optimal fitness value no longer changes during multiple consecutive iterations, the corresponding global optimal parameter combination is output as the optimal parameters of the support vector regression model, thereby completing the parameter optimization; wherein, the multiple times refers to a pre-set number of times.
2. The method for inverting organic matter in red soil based on UAV hyperspectral imagery and spectral indices according to claim 1, characterized in that, The specific process for extracting the reflectance data is as follows: Acquire multi-band reflectance data and corresponding spatial reference information of orthorectified UAV hyperspectral images; Based on the spatial reference information, establish the conversion relationship between image pixel coordinates and geographic coordinates; By traversing the pixel positions in the orthorectified UAV hyperspectral image, the reflectance values of the corresponding pixels in each band are extracted to form the spectral reflectance vector of each pixel. Based on the row and column positions of the pixels in the orthorectified UAV hyperspectral image, the corresponding geographic coordinates are calculated using the spatial reference information. The geographic coordinates and corresponding multi-band reflectance data of each pixel are recorded and stored to complete the reflectance data extraction.
3. The method for inverting organic matter in red soil based on UAV hyperspectral imagery and spectral indices according to claim 1, characterized in that, The two-band spectral indices include the difference index, ratio index, sum index, and normalized index, specifically: ； ； ； ； in, It is the difference index. It is the ratio index. For sum and index, The normalized index, , The first one in the wavelength range Band, First Reflectivity of the band.
4. The method for inverting organic matter in red soil based on UAV hyperspectral imagery and spectral indices according to claim 1, characterized in that, The three-band spectral indices include six types, specifically: ； ； ； ； ； ； in, , , , , and These represent the three-band spectral indices with coded serial numbers 1, 2, 3, 4, 5, and 6, respectively. , , The first one in the wavelength range Band, First Band, First Reflectivity of the band.