Screening Method and System for Remote Sensing Monitoring Indexes of Crop Growth Parameters

Abstract

The present invention belongs to the technical field of agricultural engineering, and particularly relates to a method and system for screening remote sensing monitoring indicators of crop growth parameters, including: obtaining unmanned aerial vehicle (UAV) remote sensing images of a research area by using a UAV multispectral sensor, and preprocessing the UAV remote sensing images to obtain orthoimage data, including multi-band reflectance; extracting multi-category vegetation indices based on the orthoimage by using a grid calculation method; calculating the texture features of multiple bands based on the orthoimage by using the gray-level co-occurrence matrix method; measuring the plant nitrogen content by using the Kjeldahl method to obtain the measured value of the plant nitrogen content; screening sensitive feature variables from the multi-band reflectance, vegetation indices and texture features by using a sensitive feature screening method for plant nitrogen content based on variable shrinkage search; and constructing a plant nitrogen content prediction model by using the Adaboost method based on the screened sensitive feature variables. The present invention simplifies variables, compresses the model, and improves the operation efficiency of the model.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural engineering, and particularly relates to a method and system for screening remote sensing monitoring indicators of crop growth parameters, which are used for constructing and estimating a model of winter wheat plant nitrogen content. Background Technique

[0002] Nitrogen is a key indicative parameter for the growth of crops, plays an important regulatory role in the physiological and biochemical processes of crops, and affects the final yield and quality of grains. Reasonable nitrogen supply can promote crop growth, while excessive nitrogen application will cause problems such as soil and water pollution and excessive emission of greenhouse gases (Zhang et al., 2015; Yu et al., 2019; Guo et al., 2022). Therefore, rapid and non-destructive monitoring of crop nitrogen content is crucial for precise farmland management and ecological environment safety. Currently, research on crop nitrogen monitoring based on unmanned aerial vehicle (UAV) remote sensing mainly uses index data such as spectral reflectance, vegetation indices, and texture features to establish prediction models for nitrogen prediction or inversion (Jia and Chen, 2020; Ling et al., 2023). These studies have well promoted the application of UAV remote sensing technology in crop nitrogen content monitoring. However, spectral information, especially vegetation indices such as NDVI, has a saturation problem when the ground coverage is large, which affects the accuracy of the model. In recent years, more and more researchers have begun to pay attention to the rich texture information in UAV ultra-high-resolution images (Liu et al., 2018; Guo et al., 2022; Usha and Vasuki, 2022). As an inherent property of images, texture features are not easily affected by the outside world, reflect the gray properties and spatial relationships of images, expand the spatial information recognition of the brightness of the original image, and can, to a certain extent, solve the saturation problem existing in spectral information inversion and improve the inversion accuracy of parameters. For example, Chen and Liang (2019) showed that after adding texture information, the R of the cotton plant nitrogen content monitoring model of the model 2 increased from the original 0.33 to 0.57. Therefore, making full use of the texture information of UAV remote sensing images can improve the monitoring accuracy.

[0003] To improve the monitoring accuracy, since researchers revealed the physical mechanism between spectral signals and leaf chemical components in the 1960s, more and more researchers have been committed to interpreting the relationship between spectral information and plant nitrogen content, attempting to estimate plant nitrogen content more accurately. From the initial methods such as multiple linear regression and partial least squares regression to the current machine learning methods such as neural networks and random forests (Walsh et al., 2018; Li Jinmin et al., 2021; Qiu et al., 2021), these studies have played an important role in improving the prediction and inversion accuracy of physical and chemical parameters such as plant nitrogen content. Moreover, machine learning methods can improve the accuracy by increasing the number of independent variables in the model and are increasingly used in the prediction and inversion of crop growth parameters (Chlingaryan et al., 2018; Fan Yiguang et al., 2023). However, due to different principles, there are differences in the learning efficiency, prediction, and inversion ability of the models established by different methods (Li Jinmin et al., 2021; Yuan Ying et al., 2023). In the case of multiple variables with severe collinearity among variables, although machine learning methods can improve the accuracy of the model, they also bring the disadvantages of complex models and excessive computing power consumption. Therefore, while ensuring the accuracy, there is an urgent need to innovate methods to eliminate redundant information of model variables and improve the computing efficiency to make the model simple and easy to use. Summary of the Invention

[0004] The purpose of the present invention is to solve the problems of redundant variables and excessive computing power consumption in the machine learning model of crop growth parameters, and to propose a method and system for screening remote sensing monitoring indicators of crop growth parameters. By extracting the most contributing sensitive feature variables to construct a plant nitrogen content prediction model, the variables are simplified, the model is compressed, and the computing efficiency of the model is improved.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] A method for screening remote sensing monitoring indicators of crop growth parameters, comprising the following steps:

[0007] Step 1: Use an unmanned aerial vehicle (UAV) multispectral sensor to obtain UAV remote sensing images of the research area, and preprocess the UAV remote sensing images to obtain orthoimage data, including multi-band reflectance;

[0008] Step 2: Based on the orthoimage in Step 1, use a grid calculation method to extract multi-category vegetation indices;

[0009] Step 3: Based on the orthoimage in Step 1, use the gray-level co-occurrence matrix method to calculate the texture features of multiple bands;

[0010] Step 4: Use the Kjeldahl method to determine the plant nitrogen content and obtain the measured value of the plant nitrogen content;

[0011] Step 5. The method for screening sensitive characteristic variables of plant nitrogen content based on variable shrinkage search screens out sensitive characteristic variables from multi-band reflectance, vegetation indices, and texture features;

[0012] Step 6. Based on the screened sensitive characteristic variables, an Adaboost method is used to construct a prediction model for plant nitrogen content.

[0013] According to the method for screening remote sensing monitoring indicators of crop growth parameters of the present invention, further, in Step 1, the orthophoto image data format is raster TIF, and the preprocessing process of the UAV remote sensing image includes image mosaicking, orthorectification, and radiometric calibration.

[0014] According to the method for screening remote sensing monitoring indicators of crop growth parameters of the present invention, further, the image mosaicking includes: using Pix4D software to perform image mosaicking based on the 3D model mode to obtain the study area image; the orthorectification includes: accurately correcting the study area image obtained by image mosaicking based on the selected ground control points; the radiometric calibration includes: based on the gray value of the radiometric calibration panel, establishing a linear conversion formula in combination with the reflection characteristics of the radiometric calibration panel, and converting the gray value of the orthorectified image into reflectance, so as to obtain the orthophoto image of the accurately corrected study area.

[0015] According to the method for screening remote sensing monitoring indicators of crop growth parameters of the present invention, further, in Step 2, the data format of the vegetation index is raster TIF; the vegetation indices include green band normalized difference vegetation index, green band optimized soil adjusted vegetation index, normalized difference vegetation index, modified simple ratio vegetation index, red edge optimized soil adjusted vegetation index, re-normalized difference vegetation index, difference vegetation index, red edge re-normalized difference vegetation index, chlorophyll absorption ratio index, optimized soil adjusted vegetation index, normalized blue-green difference index, enhanced vegetation index, triangular vegetation index, atmospheric impedance vegetation index, over-green index, ratio vegetation index, modified triangular vegetation index, soil adjusted vegetation index, normalized blue-green band difference vegetation index, and optimized vegetation index.

[0016] According to the method for screening remote sensing monitoring indicators of crop growth parameters of the present invention, further, in Step 3, the texture features include 8 indicators: contrast, correlation, dissimilarity, entropy, homogeneity, mean, angular second moment, and variance.

[0017] According to the method for screening remote sensing monitoring indicators of crop growth parameters of the present invention, further, the method for screening sensitive characteristic variables of plant nitrogen content based on variable shrinkage search in Step 5 includes:

[0018] Step 5.1. VSS method

[0019] Based on the ordinary least squares method, an L1 regularization term is introduced to construct a minimized objective function, and the objective function is:

[0020] L(ξ) = min ||Y - Xξ||^2 + λ||ξ||1

[0021] Where Y is the vector of measured values of plant nitrogen content, X is the feature matrix, ξ is the vector of regression coefficients to be estimated, and λ is the hyperparameter that controls the regularization strength;

[0022] The method of taking partial derivatives is used to find the minimum value. Since the L1 norm is not differentiable, when ξ i = 0, the following formula is used to solve for the elements ξ in each regression coefficient matrix i :

[0023]

[0024] Where X j is the j-th column of matrix X, and sgn is the sign function;

[0025] When ξ i ≠ 0, the following formula is used to solve for the elements ξ in each regression coefficient matrix i :

[0026]

[0027] Where abs is the absolute value;

[0028] When and only when the sum of the absolute values of the regression coefficients is less than a constant constraint, the mean squared error MSE of the residual sum of squares is minimized, and the regression coefficients of some feature variables shrink to 0;

[0029] Step 5.2, Solve for λ

[0030] When the mean squared error MSE is minimized, select the maximum value within one standard error of the minimum mean squared error as the constraint condition to determine the final λ value;

[0031] Step 5.3, Determine sensitive feature variables

[0032] First, use correlation analysis to determine several feature variables whose relationship with plant nitrogen content reaches a significant level of 0.01. Then, use the VSS method to screen these several feature variables. As the value of the parameter λ increases, the penalty strength increases continuously, and the regression coefficients of these several feature variables gradually compress to 0 until all are 0; when the λ value selected in Step 5.2 is reached, there are some feature variables in the model whose regression coefficients are not 0, then these sensitive feature variables are retained.

[0033] According to the method for screening remote sensing monitoring indicators of crop growth parameters of the present invention, further, after step 6, it further includes a comparative analysis of the training efficiency between the plant nitrogen content prediction model constructed based on the feature variables not screened in step 5 and the plant nitrogen content prediction model constructed based on the feature variables screened in step 5.

[0034] A screening system for remote sensing monitoring indicators of crop growth parameters, which is used to implement the method for screening remote sensing monitoring indicators of crop growth parameters as described above, and includes:

[0035] A remote sensing image acquisition module, which is used to acquire unmanned aerial vehicle (UAV) remote sensing images of the research area by using a UAV multispectral sensor, and preprocess the UAV remote sensing images to obtain orthoimage data, including multi-band reflectance;

[0036] A vegetation index calculation module, which is used to extract multi-category vegetation indexes by using a raster calculation method based on the orthoimage;

[0037] A texture feature calculation module, which is used to calculate the texture features of multiple bands by using the gray-level co-occurrence matrix method based on the orthoimage;

[0038] A measured plant nitrogen content module, which is used to measure the plant nitrogen content by using the Kjeldahl method to obtain the measured value of the plant nitrogen content;

[0039] A sensitive feature variable screening module, which is used to screen sensitive feature variables from multi-band reflectance, vegetation indexes and texture features by using the method for screening sensitive features of plant nitrogen content based on variable shrinkage search;

[0040] A nitrogen content prediction model construction module, which is used to construct a plant nitrogen content prediction model by using the Adaboost method based on the screened sensitive feature variables.

[0041] According to the screening system for remote sensing monitoring indicators of crop growth parameters of the present invention, further, it further includes a comparative analysis module, which is used to conduct a comparative analysis of the training efficiency between the plant nitrogen content prediction model constructed based on the unscreened feature variables and the plant nitrogen content prediction model constructed based on the screened feature variables.

[0042] Compared with the prior art, the present invention has the following advantages:

[0043] The present invention proposes a method for screening sensitive characteristics of plant nitrogen content based on variable shrinkage search (VSS). By screening the multi-band reflectance, vegetation indices, and texture features extracted from UAV remote sensing images, the sensitive characteristic variables that contribute the most to the plant nitrogen content prediction model are extracted. The Adaboost method is used to construct a prediction model for winter wheat plant nitrogen content, simplifying the variables, compressing the model, and improving the model operation efficiency. Compared with the model without variable screening, the present invention has the characteristics of high efficiency and high precision, realizing the rapid and accurate monitoring of wheat plant nitrogen content. The present invention provides a new way for simplifying the crop growth parameter prediction model and provides technical support for the precise management of nitrogen fertilizer in smart agriculture. Description of the Drawings

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0045] Figure 1 is a schematic flow chart of the method for screening remote sensing monitoring indicators of crop growth parameters in the embodiment of the present invention;

[0046] Figure 2 is a relationship diagram of λ and MSE in the embodiment of the present invention;

[0047] Figure 3 is a trend diagram of the regression coefficient changing with λ during the screening process of sensitive characteristic variables in the embodiment of the present invention;

[0048] Figure 4 is a relationship diagram of the measured value and the predicted value of winter wheat plant nitrogen content in the embodiment of the present invention. Detailed Embodiments

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0050] As Figure 1 shown, the method for screening remote sensing monitoring indicators of crop growth parameters in this embodiment includes the following steps:

[0051] Step S101: Use a drone multispectral sensor to obtain drone remote sensing images of the research area, and preprocess the drone remote sensing images to obtain orthoimage data, including multi-band reflectance.

[0052] Specifically, during the critical growth stages of winter wheat from 2020 to 2023, a DJ M600 Pro six-rotor drone remote sensing platform equipped with a MicaSense-RedEdge-M multispectral imager was used to obtain canopy multispectral images of winter wheat at the jointing stage, booting stage, flowering stage, and filling stage. This multispectral imager sensor includes 5 bands: blue light (B), green light band (G), red light band (R), red edge band (Rededge), and near-infrared band (Nir). The data format obtained is TIF and is saved on an SD card. When acquiring data, ground control points (GCPs) were set at the 4 vertices and the middle position of the research area for precise correction of flight position and altitude, reducing position errors.

[0053] As a further solution, the preprocessing process of the drone remote sensing images includes:

[0054] Step a: Image mosaicking

[0055] First, copy the remote sensing image data stored on the SD card to the computer. Use Pix4D software to perform image mosaicking based on the 3Dmodel mode to obtain the research area image.

[0056] Step b: Orthorectification

[0057] Based on the ground control points (GCPs), precisely correct the research area image obtained in step a to eliminate position deviations caused by fast flight speed, unstable attitude, and image distortion. This step can make the positions of the drone remote sensing images of the research area obtained at different times match precisely.

[0058] Step c: Radiometric calibration

[0059] Based on the gray values of the radiometric calibration panel and combined with the reflection characteristics of the radiometric calibration panel, establish a linear conversion formula to convert the gray values of the orthorectified image into reflectance to eliminate the influence of factors such as different times, solar altitude angles, and weather, thereby obtaining a precisely corrected orthoimage of the research area. The reflectances of the 5 bands are respectively denoted as R b 、R g 、R r 、R rededge and R nir .

[0060] Step S102: Based on the orthoimage in step c, use the raster calculation method to extract multi-category vegetation indices.

[0061] In this embodiment, a grid calculation method is used to extract vegetation indices of 20 categories, and the data format is grid TIF. See Table 1 for the vegetation indices and calculation formulas.

[0062] Table 1 Vegetation Indices and Calculation Formulas

[0063]

[0064]

[0065] R in Table 1 b , R g , R r , R nir , R rededge respectively represent the spectral reflectances of the blue light band, green light band, red light band, near-infrared light band, and red edge band.

[0066] Step S103: Based on the orthoimage in step c, the gray-level co-occurrence matrix method is used to calculate the texture features of multiple bands.

[0067] In this embodiment, the extracted texture feature information of 5 bands includes 8 indices: contrast (con), correlation (cor), dissimilarity (dis), entropy (ent), homogeneity (hom), mean (mean), sum of squared differences (sm), and variance (var). The specific calculation formulas are as follows:

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076] In the formula, i and j represent the row and column corresponding to the pixel of the UAV remote sensing image, and p(i, j) is the frequency of adjacent pixels.

[0077] Step S104: The Kjeldahl method is used to measure the nitrogen content of the plants, and the measured value of the nitrogen content of the plants is obtained.

[0078] Specifically, the collection of ground data was carried out simultaneously with the collection of UAV remote sensing images. In the study area, a region with uniform growth was selected. By fixing the total number of stems in 2 rows (0.2 m) × 1 m, 20 single-stem samples were taken and put into sealed bags. They were blanched at 105 °C in the laboratory and dried to a constant weight at 80 °C. After the plants were crushed, the Kjeldahl method was used to determine the nitrogen content, and a total of 360 measured values of the nitrogen content of winter wheat plants were obtained.

[0079] Step S105: Based on the 5-band reflectance, vegetation indices, and texture features obtained in steps S101, S102, and S103, a total of 65 feature variables, and the measured values of the plant nitrogen content obtained in step S104, a method for screening sensitive feature variables of plant nitrogen content based on variable shrinkage search (VSS) was constructed to screen out sensitive feature variables from the 65 feature variables.

[0080] Step A: VSS method

[0081] On the basis of ordinary least squares, an L1 regularization term (λ||ξ||1) was introduced to construct a minimization objective function to achieve the screening of feature variables and the sparsity of regression coefficients. The objective function is:

[0082] L(ξ) = min ||Y - Xξ||^2 + λ||ξ||1 (9)

[0083] In the formula, Y is the vector of measured values of plant nitrogen content, X is the feature matrix, ξ is the vector of regression coefficients to be estimated, and λ is the hyperparameter that controls the regularization strength.

[0084] The method of taking partial derivatives was used to find the minimum value. Since the L1 norm is not differentiable, when ξ i = 0, formula (10) was used to solve for the elements ξ in each regression coefficient matrix i :

[0085]

[0086] In the formula, X j is the j-th column of matrix X, and sgn is the sign function.

[0087] When ξ i ≠ 0, formula (11) was used to solve for the elements ξ in each regression coefficient matrix i :

[0088]

[0089] In the formula, abs is the absolute value;

[0090] Under the constraint that the sum of the absolute values of the regression coefficients is less than a constant, the mean square error (MSE) of the residual sum of squares is minimized, and the regression coefficients of some feature variables shrink to 0.

[0091] The calculation formula of MSE is:

[0092]

[0093] In the formula, n is the number of plant nitrogen content samples, and y i represents the measured value of the plant nitrogen content, represents the predicted value corresponding to the measured value of the plant nitrogen content.

[0094] Step B: Solve for λ

[0095] In this embodiment, the relationship between λ and MSE is shown in Figure 2 . When MSE is minimized, λ min = 0.003. Select the maximum value within one standard error of the minimum mean square error as the constraint condition, and λ 1se = 0.08. Therefore, the final selected value of λ is 0.08 to make the model performance excellent and the number of independent variables the least.

[0096] Step C: Determine sensitive feature variables

[0097] First, use correlation analysis to determine 51 feature variables whose relationship with the plant nitrogen content reaches a significant level of 0.01. Then, use the VSS method to screen these 51 feature variables. As the value of the parameter λ increases, the penalty intensity increases continuously, and the regression coefficients of the 51 feature variables in the model gradually shrink to 0 until all are 0. The relationship between λ and the model regression coefficients is shown in Figure 3 . When λ takes the value of 0.08, the regression coefficients of 17 feature variables in the model are not 0, indicating that 17 sensitive feature variables are retained after VSS screening, which are R b 、R nir 、RERDVI、NGBDI、VIopt、cor_G、cor_Rededge、cor_Nir、ent_G、ent_R、hom_R、hom_Nir、mean_R、mean_Nir、sm_B、sm_R、var_Nir.

[0098] Step S106: Based on the screened sensitive feature variables, use the Adaboost method to construct a plant nitrogen content prediction model.

[0099] Based on the 17 sensitive feature variables screened in Step C, use the Adaboost method to construct a prediction model for the plant nitrogen content (g / m 2 ). The relationship between the measured value and the predicted value is shown in Figure 4。Model R 2 is 0.81, and the root mean square error (RMSE) is 3.82 g / m 2 。

[0100] Step S107, use the same method to construct a plant nitrogen content prediction model for 51 variables that have not been screened by VSS. The model training takes 3.906 s, and the model training for the 17 variables after screening takes 1.992 s, with an efficiency improvement of 49.00%.

[0101] Corresponding to the above screening method for remote sensing monitoring indicators of crop growth parameters, this embodiment also proposes a screening system for remote sensing monitoring indicators of crop growth parameters, including:

[0102] A remote sensing image acquisition module, which is used to acquire unmanned aerial vehicle (UAV) remote sensing images of the research area by using a UAV multispectral sensor, and preprocess the UAV remote sensing images to obtain orthoimage data containing multi-band reflectance.

[0103] A vegetation index calculation module, which is used to extract multi-category vegetation indices based on the orthoimage by using a raster calculation method.

[0104] A texture feature calculation module, which is used to calculate the texture features of multiple bands based on the orthoimage by using the gray level co-occurrence matrix method.

[0105] A measured plant nitrogen content module, which is used to measure the plant nitrogen content by using the Kjeldahl method to obtain the measured value of the plant nitrogen content.

[0106] A sensitive feature variable screening module, which is used to screen sensitive feature variables from multi-band reflectance, vegetation indices, and texture features based on the method for screening sensitive features of plant nitrogen content by variable shrinkage search (VSS).

[0107] A nitrogen content prediction model construction module, which is used to construct a plant nitrogen content prediction model by using the Adaboost method based on the screened sensitive feature variables.

[0108] A comparative analysis module, which is used to conduct a comparative analysis of the training efficiency between the plant nitrogen content prediction model constructed based on the unscreened feature variables and the plant nitrogen content prediction model constructed based on the screened feature variables.

[0109] Based on the multi-band reflectance, vegetation indices, and texture features extracted from UAV remote sensing images, the present invention extracts the sensitive feature variables that contribute the most to the plant nitrogen content prediction model through the method for screening sensitive features of plant nitrogen content by variable shrinkage search (VSS), and constructs a winter wheat plant nitrogen content prediction model by using the Adaboost method, simplifies the variables, compresses the model, reduces the model scale, and thus improves the operation efficiency of the model.

[0110] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code. The solutions in the embodiments of the present invention can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript, etc.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 process or multiple processes and / or boxes Figure 1 or multiple boxes.

[0112] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0113] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims

1. A method for screening remote sensing monitoring indicators of crop growth parameters, characterized in that, It includes the following steps: Step 1: Use a drone multispectral sensor to obtain drone remote sensing images of the research area, and preprocess the drone remote sensing images to obtain orthophoto image data, including multi-band reflectance; Step 2: Based on the orthophoto image in Step 1, use a raster calculation method to extract multi-category vegetation indices; Step 3: Based on the orthophoto image in Step 1, use the gray-level co-occurrence matrix method to calculate the texture features of multiple bands; Step 4: Use the Kjeldahl method to determine the plant nitrogen content and obtain the measured value of the plant nitrogen content; Step 5: Use a sensitive feature screening method for plant nitrogen content based on variable shrinkage search to screen out sensitive feature variables from multi-band reflectance, vegetation indices, and texture features, specifically including: Step 5.1: VSS method On the basis of ordinary least squares, introduce the L1 regularization term to construct a minimization objective function. The objective function is: L(ξ)=min||Y-Xξ||^2+λ||ξ||1 In the formula, Y is the vector of measured plant nitrogen content values, X is the feature matrix, ξ is the vector of regression coefficients to be estimated, and λ is a hyperparameter that controls the regularization strength; The method of finding the minimum value by taking partial derivatives is adopted. Since the L1 norm is not differentiable, when ξ i = 0, the following formula is used to solve for the elements in each regression coefficient matrix ξ i : where X j is the j-th column of matrix X, and sgn is the sign function; When ξ i ≠ 0, the following formula is used to solve for the elements in each regression coefficient matrix ξ i : In the formula, abs is the absolute value; When and only when the sum of the absolute values of the regression coefficients is less than a constant constraint condition, the mean square error MSE of the residual sum of squares is minimized, and the regression coefficients of some feature variables shrink to 0; Step 5.2: Solve for λ When the mean square error MSE is the smallest, select the maximum value within one standard error of the minimum mean square error as the constraint condition to determine the final λ value; Step 5.3: Determine sensitive feature variables First, use correlation analysis to determine several feature variables whose relationship with the plant nitrogen content reaches a significant level of 0.

01. Then, use the VSS method to screen these several feature variables. As the value of the parameter λ increases, the penalty intensity continuously increases, and the regression coefficients of these several feature variables gradually compress to 0 until all are 0; when reaching the λ value selected in Step 5.2, the regression coefficients of some feature variables in the model are not 0, so these sensitive feature variables are retained; Step 6: Based on the screened sensitive feature variables, use the Adaboost method to construct a plant nitrogen content prediction model.

2. The screening method of remote sensing monitoring indexes for crop growth parameters according to claim 1, characterized in that In Step 1, the orthophoto image data format is raster TIF, and the preprocessing process of the drone remote sensing images includes image mosaicking, orthorectification, and radiometric calibration.

3. The screening method of remote sensing monitoring indexes for crop growth parameters according to claim 2, characterized in that The image mosaicking includes: using Pix4D software to perform image mosaicking based on the 3Dmodel mode to obtain the research area image; the orthorectification includes: based on the selected ground control points, accurately correcting the research area image obtained by image mosaicking; the radiometric calibration includes: based on the gray value of the radiometric calibration panel, combined with the reflection characteristics of the radiometric calibration panel, establishing a linear conversion formula to convert the gray value of the orthorectified image into reflectance, so as to obtain the accurately rectified orthophoto image of the research area.

4. The screening method for remote sensing monitoring indicators of crop growth parameters according to claim 1, characterized in that In step 2, the data format of the vegetation index is raster TIF; the vegetation index includes green band normalized difference vegetation index, green band optimized soil adjusted vegetation index, normalized difference vegetation index, modified simple ratio vegetation index, red edge optimized soil adjusted vegetation index, re-normalized difference vegetation index, difference vegetation index, red edge re-normalized difference vegetation index, chlorophyll absorption ratio index, optimized soil adjusted vegetation index, normalized blue-green difference index, enhanced vegetation index, triangular vegetation index, atmospheric impedance vegetation index, excessive green index, ratio vegetation index, modified triangular vegetation index, soil adjusted vegetation index, normalized blue-green band difference vegetation index, and optimized vegetation index.

5. The screening method of remote sensing monitoring indicators for crop growth parameters according to claim 1, characterized in that, In step 3, the texture features include 8 indexes: contrast, correlation, dissimilarity, entropy, homogeneity, mean value, angular second moment, and variance.

6. The screening method of remote sensing monitoring indexes for crop growth parameters according to claim 1, characterized in that After step 6, it further includes a comparative analysis of the training efficiency between the plant nitrogen content prediction model constructed based on the feature variables not screened in step 5 and the plant nitrogen content prediction model constructed based on the feature variables screened in step 5.

7. A screening system for remote sensing monitoring indicators of crop growth parameters, characterized in that, A method for screening remote sensing monitoring indicators of crop growth parameters as described in any one of claims 1-6 includes: A remote sensing image acquisition module, which is used to obtain the UAV remote sensing image of the research area by using a UAV multispectral sensor, and preprocess the UAV remote sensing image to obtain orthoimage data, including multi-band reflectance; A vegetation index calculation module, which is used to extract multi-category vegetation indexes based on the orthoimage by using a raster calculation method; A texture feature calculation module, which is used to calculate the multi-band texture features based on the orthoimage by using the gray-level co-occurrence matrix method; A measured plant nitrogen content module, which is used to measure the plant nitrogen content by using the Kjeldahl method to obtain the measured value of the plant nitrogen content; A sensitive feature variable screening module, which is used to screen out sensitive feature variables from the multi-band reflectance, vegetation index, and texture features by using a method for screening sensitive features of plant nitrogen content based on variable shrinkage search; A nitrogen content prediction model construction module, which is used to construct a plant nitrogen content prediction model by using the Adaboost method based on the screened sensitive feature variables.

8. The screening system for remote sensing monitoring indicators of crop growth parameters according to claim 7, characterized in that, It further includes a comparative analysis module, which is used to conduct a comparative analysis of the training efficiency between the plant nitrogen content prediction model constructed based on the unscreened feature variables and the plant nitrogen content prediction model constructed based on the screened feature variables.

Images