Method and device for estimating flue-cured tobacco prosperity index based on unmanned aerial vehicle canopy spectrum and texture

Abstract

The invention relates to the technical field of intelligent agriculture, in particular to a flue-cured tobacco prosperity index estimation method and device based on unmanned aerial vehicle canopy spectrum and texture. The method comprises the following steps: S1, measuring an SPAD value and a business index value; s2, unmanned aerial vehicle multispectral and RGB image acquisition; s3, image processing and data extraction; s4, inputting the model for calculation; s5, standardization and feature engineering; s6, estimating a flue-cured tobacco prosperity index; and S7, determining the precision of the model. According to the method, a set of flue-cured tobacco canopy texture indexes are deeply excavated to serve as a feature set based on unmanned aerial vehicle aerial photography and a multispectral image processing technology, a tobacco leaf prosperity index model is constructed by adopting an advanced machine learning method XGBoost, wide-area, rapid, timely and accurate monitoring is carried out on tobacco leaf growth vigor, and a basis is provided for objectively and truly mastering flue-cured tobacco production conditions in a large area. And a scientific basis is provided for timely, efficient and accurate management of tobacco production, early formulation of a tobacco allocation plan and the like.

Description

Technical Field

[0001] The present invention relates to the field of smart agricultural technology, and in particular to a method and device for estimating a flue-cured tobacco prosperity index based on canopy spectrum and texture of an unmanned aerial vehicle. Background Art

[0002] As an important indicator to measure the growth of tobacco leaves, the prosperity index accurately and timely grasps the tobacco prosperity index information within a certain range of flue-cured tobacco planting areas, which is of great significance for formulating flue-cured tobacco production management plans, allocating tobacco leaves in the later stage, and protecting the income of tobacco farmers. As the forefront of earth information science, UAV remote sensing has the characteristics of macro, fast, accurate, objective, dynamic, multi-scale, large coverage area, short monitoring cycle, and low cost. It has unique advantages in large-scale agricultural monitoring and is an important means of monitoring the growth of flue-cured tobacco. At present, some scholars have used UAV hyperspectral and multi-spectral technology to successfully monitor and invert the growth of wheat, corn, forage, etc., such as leaf area index, and even predict yield. However, there is no report in China on large-scale monitoring of tobacco leaf growth through the inversion of tobacco leaf prosperity index. As shown in Chinese patent CN116778319A, an innovative tobacco prosperity index estimation method is proposed, which cleverly combines UAV remote sensing technology with hyperspectral acquisition technology. However, the former has problems such as the prediction accuracy not meeting the requirements and the image data being easily affected by the weather; the drone hyperspectral has problems such as expensive equipment and high operating costs, and cannot be promoted and used in production. Summary of the invention

[0003] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method and device for estimating the flue-cured tobacco prosperity index based on the spectrum and texture of the canopy of an unmanned aerial vehicle.

[0004] The technical solution adopted by the present invention is as follows:

[0005] A flue-cured tobacco prosperity index estimation method based on drone canopy spectrum and texture includes the following steps:

[0006] S1. Measurement of SPAD value and prosperity index value: Select representative sampling plots according to topography and landforms; Use a sliding ruler to measure the height and width of the flue-cured tobacco canopy and determine the SPAD value; Calculate LNI according to the prosperity index model, and finally calculate the measured prosperity index value and take the average value;

[0007] S2. UAV multispectral and RGB image acquisition: Use a UAV to fly and photograph the sampling plots to obtain multispectral images and RGB images of the flue-cured tobacco canopy;

[0008] S3, image processing and data extraction: process the captured multispectral images and RGB images, and extract the spectral data and texture of the flue-cured tobacco canopy;

[0009] S4. Inputting the model for calculation: inputting the spectral data and the texture of the flue-cured tobacco canopy into the model for calculation, and calculating the spectral index and texture data of the flue-cured tobacco canopy;

[0010] S5. Standardization and feature engineering: Standardize the spectral index and texture data of the flue-cured tobacco canopy to ensure uniform dimensions and ranges; select key feature data sets through feature engineering to construct input data for the estimation model;

[0011] S6. Estimation of flue-cured tobacco prosperity index: using XGBoost integrated algorithm, the flue-cured tobacco prosperity index is estimated based on the screened feature data set;

[0012] S7, determine the accuracy of the model: compare the calculation results of step S1 and step S6 to obtain the estimation error, thereby determining the accuracy of the estimation model.

[0013] This technical solution selects representative sampling plots through topographic analysis, uses physical measurement tools to obtain specific parameters of the flue-cured tobacco canopy, and calculates a preliminary prosperity index value in combination with the prosperity index model; uses drones to collect multispectral and RGB images, so that a wide range of flue-cured tobacco canopy information can be quickly and accurately acquired; by processing multispectral images and RGB images, not only the spectral data of the flue-cured tobacco canopy is extracted, but also its texture features are captured, which are directly related to the growth status and prosperity level of the flue-cured tobacco; the extracted spectral data and texture features are input into a specific model for calculation, and more detailed spectral index and texture data are obtained through calculation; by standardizing the spectral index and texture data The dimensional differences between different features are eliminated to make the data more comparable. The key feature data sets are screened out through feature engineering to build a more streamlined and effective model input data. The XGBoost integrated algorithm is used to estimate the flue-cured tobacco prosperity index for the screened feature data sets. As an efficient machine learning algorithm, XGBoost can make full use of the information in the input data through complex calculation and optimization processes. By comparing the measured prosperity index value in step S1 with the estimated prosperity index value in step S6, the estimation error is calculated to determine the accuracy of the estimation model, which not only verifies the accuracy and reliability of the model, but also provides important feedback information for subsequent model optimization and adjustment.

[0014] According to an embodiment of the present invention, in step S1, the measurement of the SPAD value and the prosperity index value includes the following specific steps:

[0015] S11. Determine the statistical area: select a certain range of base units and determine their statistical area;

[0016] S12. According to the topography, altitude, field type, and tobacco growth in previous years, determine 20-50 field survey plots, each as one point;

[0017] S13. Use a sliding ruler to measure the height and width of the flue-cured tobacco canopy and the SPAD value; determine and calculate the LNI according to the prosperity index model, measure each point 10 times, and finally calculate the prosperity index value and take the average value, where:

[0018] Prosperity index = CW2×CH×SPa / 106×LNI

[0019] Where: CW is the canopy width of the plant; CH is the canopy height of the plant; SPa is the chlorophyll SPAD value;

[0020] LNI is the leaf number index, and the calculation formula is as follows:

[0021]

[0022] Wherein: L is the maximum leaf length of the leaf; W is the maximum leaf width of the leaf, k=1 to n, is the leaf number; C is the coefficient.

[0023] This technical solution determines the statistical area, selects representative field survey plots, uses a sliding ruler to accurately measure the height, width and SPAD value of the flue-cured tobacco canopy, and calculates the LNI based on the prosperity index model, ultimately obtaining an accurate and average flue-cured tobacco prosperity index value to improve the reliability and accuracy of the estimate.

[0024] According to an embodiment of the present invention, in step S13, the steps of measuring the height and width of the flue-cured tobacco canopy with a sliding ruler are as follows:

[0025] S131, unfold the sliding ruler along the middle rotating part, which is divided into a horizontal ruler and a vertical ruler arranged vertically, wherein:

[0026] The horizontal ruler is 2m long, the scale unit is cm, and it is arranged from left to right;

[0027] The vertical ruler is 2m long, the scale unit is cm, and it is arranged from bottom to top;

[0028] S132. Slide left and right along the horizontal ruler to record the maximum canopy width of the flue-cured tobacco plant, accurate to cm;

[0029] S133. Slide up and down along the vertical ruler and record the vertical distance from the ground to the highest leaf, accurate to cm;

[0030] S134. The sliding ruler is folded along the middle rotating part to facilitate carrying.

[0031] This technical solution achieves accurate measurement of geometric features of flue-cured tobacco canopies through the operation steps of the sliding ruler, including unfolding the sliding ruler, recording canopy width and height with accuracy to centimeters, and providing accurate data for subsequent calculations.

[0032] According to one embodiment of the present invention, in step S2, the drone multispectral and RGB image acquisition includes the following specific steps:

[0033] S21. Select a clear day and take a drone photo at noon, and set the drone's photo mode, route generation mode, flight speed, flight altitude, repetition rate on the main route, repetition rate between main routes, repetition rate between main routes, gimbal pitch angle, fixed-point mode, camera parameters, shooting time, and focal length to obtain multispectral and RGB color images of the flue-cured tobacco canopy;

[0034] S22, simultaneously taking a flying shot of the calibration plate to obtain an image;

[0035] S23. Check the image quality on site. If it is unqualified, retake the image in time.

[0036] This technical solution sets the UAV flight parameters and camera mode to collect high-quality multispectral and RGB images under suitable weather conditions, ensuring that clear and accurate images of the flue-cured tobacco canopy are obtained, laying the foundation for subsequent image processing and analysis.

[0037] According to an embodiment of the present invention, in step S3, image processing and data extraction include the following specific steps:

[0038] S31. Use the built-in algorithm of the drone to perform image correction and stitching on multispectral images and RGB images;

[0039] S32. In the spliced ​​image, use image segmentation to identify and extract the flue-cured tobacco canopy area, and extract spectral data and tobacco canopy texture in the segmented flue-cured tobacco canopy area.

[0040] This technical solution uses the built-in algorithm of the drone to correct and stitch images, and uses image segmentation technology to extract the tobacco canopy area and its spectral data and texture features, providing key input data for subsequent model calculations.

[0041] According to an embodiment of the present invention, in step S4, inputting the model for calculation includes the following specific steps:

[0042] S41. The spectral data and the texture of the flue-cured tobacco canopy are input into the model to calculate the vegetation index of the flue-cured tobacco canopy; the vegetation index includes the following characteristic variable sets:

[0043] MeanB, MeanG, MeanR, MeanRE, and MeanNIR are the mean texture eigenvalues ​​of the Blue, Green, Red, Red Edge, and Near Infrared bands calculated based on the gray-level co-occurrence matrix GLCM, respectively;

[0044] HomoB and HomoR are the synergistic texture eigenvalues ​​of the Blue and Red bands calculated based on the gray-level co-occurrence matrix GLCM;

[0045] VarB, VarG, VarRE, and VarNIR are the variance texture eigenvalues ​​of the Blue, Green, RedEdge, and Near Infrared bands calculated based on the gray-level co-occurrence matrix GLCM;

[0046] CorrB and CorrR are the correlation texture eigenvalues ​​of the Blue and Red bands calculated based on the gray-level co-occurrence matrix GLCM;

[0047] Blue, Green, Red, and NIR are the spectral reflectances of the blue, green, red, and near-infrared bands of the flue-cured tobacco canopy, respectively.

[0048] S42. Calculate the spectral index and texture data of the flue-cured tobacco canopy, and evaluate their weights.

[0049] This technical solution inputs spectral data and tobacco canopy texture into the model, calculates vegetation index and texture eigenvalues, and evaluates their weights to comprehensively reflect the growth status and characteristics of the tobacco canopy, providing a scientific basis for the estimation of the prosperity index.

[0050] According to an embodiment of the present invention, in step S5, the standardization and feature engineering includes the following specific steps:

[0051] S51. Data standardization execution: Apply the standardization method to standardize the spectral index and texture data of the flue-cured tobacco canopy one by one; after completing the standardization, immediately perform a data distribution check to ensure that no new abnormal values ​​are introduced during the standardization process, and verify whether each characteristic value has been adjusted to the expected value range;

[0052] S52. Feature engineering selection and optimization: Use the Pearson correlation coefficient and mutual information index to calculate the correlation strength between each feature and the estimated target, and preliminarily identify the feature set that is highly correlated with the target; perform in-depth screening and optimization of the feature set to remove redundant and noise features; perform feature transformation operations on the screened key features, including square root transformation, feature crossover and aggregation.

[0053] This technical solution standardizes spectral index and texture data and uses statistical methods to screen and optimize feature sets to eliminate data differences, reduce noise and redundancy, and improve the efficiency and accuracy of model estimation.

[0054] According to an embodiment of the present invention, in step S6, estimating the flue-cured tobacco prosperity index includes the following specific steps:

[0055] S61, XGBoost model parameter configuration: configure the parameters of the XGBoost model, including learning rate, maximum depth, number of trees, and subsample ratio; use cross-validation method, GridSearchCV or RandomizedSearchCV, combined with the spectral index and texture data of the flue-cured tobacco canopy, to optimize and determine the best parameter combination suitable for the current feature set;

[0056] S62. Training of XGBoost model: Preprocess and fuse the spectral index and texture data of the flue-cured tobacco canopy through feature concatenation and feature conversion; train the XGBoost model based on the training set data containing these features.

[0057] This technical solution configures and optimizes the XGBoost model parameters and uses the training set data to train the model to build a machine learning model that can accurately estimate the flue-cured tobacco prosperity index.

[0058] According to an embodiment of the present invention, in step S7, determining the accuracy of the model includes the following specific steps:

[0059] S71. Use the trained XGBoost model to estimate the test data and obtain the estimated prosperity index. At the same time, perform the same preprocessing on the actual production data to verify the generalization ability and actual application effect of the model.

[0060] S72. Evaluate the accuracy of the model estimation results through mean square error, root mean square error, and mean absolute error.

[0061] This technical solution compares the model estimation results with the actual production data and uses a variety of error evaluation indicators to verify the accuracy and generalization ability of the model, ensuring that the model can reliably estimate the flue-cured tobacco prosperity index in practical applications.

[0062] To achieve the above-mentioned purpose, the present invention also provides a flue-cured tobacco prosperity index estimation device based on drone canopy spectrum and texture.

[0063] A flue-cured tobacco prosperity index estimation device based on drone canopy spectrum and texture, comprising:

[0064] The prosperity index value measurement module is used to measure the SPAD value of flue-cured tobacco on the ground and preliminarily calculate the prosperity index value;

[0065] UAV flight shooting module, used to obtain multispectral images and RGB images of the flue-cured tobacco canopy;

[0066] The image processing and data extraction module is used to process the captured multispectral images and RGB images and extract the spectral data and texture of the flue-cured tobacco canopy;

[0067] Data processing and analysis module, used to standardize and feature engineer the extracted spectral and texture data, and construct input data for the estimation model;

[0068] The flue-cured tobacco prosperity index estimation model uses the XGBoost integrated algorithm to estimate the flue-cured tobacco prosperity index based on the processed feature data set;

[0069] The model accuracy assessment module is used to compare the measured prosperity index value with the estimated prosperity index value to determine the accuracy of the estimation model.

[0070] This technical solution first obtains the basic SPAD value and preliminary prosperity index value through the ground measurement module (prosperity index value measurement module); the SPAD value, as an indirect indicator of chlorophyll content, reflects the growth status of flue-cured tobacco; the UAV flight shooting module uses remote sensing technology to obtain large-scale, high-resolution multi-spectral and RGB images of flue-cured tobacco canopies from the air, making up for the shortcomings of ground measurement in spatial coverage; the image processing and data extraction module uses advanced image processing algorithms to correct, splice, segment and process the multi-spectral and RGB images taken by drones to extract the spectral data and texture characteristics of the flue-cured tobacco canopy; the spectral data reflects the reflection or emission characteristics of the canopy in different bands, while the texture characteristics describe the spatial structure information of the canopy; these characteristics together constitute the basic data for subsequent analysis; the data processing The processing and analysis module standardizes the extracted spectral data and texture data to ensure the comparability between different features; through feature engineering methods such as feature selection and feature transformation, the feature subsets that have an important impact on the estimation of the flue-cured tobacco prosperity index are screened out, and the input data suitable for the estimation model is constructed; the flue-cured tobacco prosperity index estimation model adopts the XGBoost integration algorithm. The advantage of the XGBoost algorithm is that it can handle large-scale data sets, automatically select features and adjust weights, and improve the overall performance by integrating multiple weak learners; the model accuracy evaluation module calculates the estimation error by comparing the measured prosperity index value and the estimated prosperity index value, thereby evaluating the accuracy of the model; if the estimation error is large, the performance of the model can be improved by adjusting model parameters, improving feature engineering, and other methods.

[0071] Compared with the prior art, the present invention has the following beneficial effects:

[0072] The present invention relies on UAV aerial photography and multispectral image processing technology to deeply mine a set of tobacco canopy texture indicators as the feature set, and adopts the advanced machine learning method XGBoost to construct a tobacco leaf prosperity index model to conduct wide-area, rapid, timely and accurate monitoring of tobacco leaf growth, so as to provide a scientific basis for objectively and truly grasping the status of flue-cured tobacco production over a large area, realizing timely, efficient and accurate management of tobacco leaf production, and formulating tobacco leaf allocation plans as early as possible.

[0073] (1) The present invention utilizes drone technology to monitor tobacco leaf prosperity index quickly, accurately, timely, and over a wide area, greatly reducing the workload of field measurements, reducing costs, and improving the scientific and technological content and applicability of smart agriculture;

[0074] (2) The present invention takes the tobacco canopy texture index as the core, and the characteristic indicators are strictly and scientifically screened and repeatedly verified, so that the constructed model has high prediction accuracy; both in terms of indicator screening and modeling methods, it has significantly improved the existing crop modeling technology; the above indicators also provide valuable scientific references for the growth monitoring of other crops;

[0075] (3) The present invention adopts a sliding ruler for flue-cured tobacco field survey, which can be easily operated in the field, is durable, has high accuracy, and saves a lot of manpower and material resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a flow diagram of the method of the present invention.

[0077] Figure 2 It is a cross sliding scale diagram of the flue-cured tobacco field survey in step S1.

[0078] Figure 3 It is a measurement diagram of the height and width of the flue-cured tobacco canopy in step S1.

[0079] Figure 4 It is the setting diagram of the drone image acquisition parameters in step S2.

[0080] Figure 5 It is the measurement image of the flue-cured tobacco canopy texture in step S3.

[0081] Figure 6 It is a diagram for setting the core parameters of the estimation model in step S4.

[0082] Figure 7 It is the distribution diagram of the standardization and feature engineering in step S5.

[0083] Figure 8 It is a comparison chart of the actual measured value and the estimated value of the flue-cured tobacco prosperity index in step S7. DETAILED DESCRIPTION

[0084] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0085] Example 1

[0086] like Figure 1 As shown, this embodiment provides a method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture, including the following steps:

[0087] S1. Measurement of SPAD value and prosperity index value: Select representative sampling plots according to topography and landforms; Use a sliding ruler to measure the height and width of the flue-cured tobacco canopy and determine the SPAD value; Calculate LNI according to the prosperity index model, and finally calculate the measured prosperity index value and take the average value;

[0088] S2. UAV multispectral and RGB image acquisition: Use a UAV to fly and photograph the sampling plots to obtain multispectral images and RGB images of the flue-cured tobacco canopy;

[0089] S3, image processing and data extraction: process the captured multispectral images and RGB images, and extract the spectral data and texture of the flue-cured tobacco canopy;

[0090] S4. Inputting the model for calculation: inputting the spectral data and the texture of the flue-cured tobacco canopy into the model for calculation, and calculating the spectral index and texture data of the flue-cured tobacco canopy;

[0091] S5. Standardization and feature engineering: Standardize the spectral index and texture data of the flue-cured tobacco canopy to ensure uniform dimensions and ranges; select key feature data sets through feature engineering to construct input data for the estimation model;

[0092] S6. Estimation of flue-cured tobacco prosperity index: using XGBoost integrated algorithm, the flue-cured tobacco prosperity index is estimated based on the screened feature data set;

[0093] S7, determine the accuracy of the model: compare the calculation results of step S1 and step S6 to obtain the estimation error, thereby determining the accuracy of the estimation model.

[0094] This technical solution selects representative sampling plots through topographic analysis, uses physical measurement tools to obtain specific parameters of the flue-cured tobacco canopy, and calculates a preliminary prosperity index value in combination with the prosperity index model; uses drones to collect multispectral and RGB images, so that a wide range of flue-cured tobacco canopy information can be quickly and accurately acquired; by processing multispectral images and RGB images, not only the spectral data of the flue-cured tobacco canopy is extracted, but also its texture features are captured, which are directly related to the growth status and prosperity level of the flue-cured tobacco; the extracted spectral data and texture features are input into a specific model for calculation, and more detailed spectral index and texture data are obtained through calculation; by standardizing the spectral index and texture data The dimensional differences between different features are eliminated to make the data more comparable. The key feature data sets are screened out through feature engineering to build a more streamlined and effective model input data. The XGBoost integrated algorithm is used to estimate the flue-cured tobacco prosperity index for the screened feature data sets. As an efficient machine learning algorithm, XGBoost can make full use of the information in the input data through complex calculation and optimization processes. By comparing the measured prosperity index value in step S1 with the estimated prosperity index value in step S6, the estimation error is calculated to determine the accuracy of the estimation model, which not only verifies the accuracy and reliability of the model, but also provides important feedback information for subsequent model optimization and adjustment.

[0095] In addition, the method for estimating the flue-cured tobacco prosperity index based on the spectrum and texture of the canopy of drones proposed in the present invention also has the following additional technical features:

[0096] According to an embodiment of the present invention, in step S1, the measurement of the SPAD value and the prosperity index value includes the following specific steps:

[0097] S11. Determine the statistical area: select a certain range of base units and determine their statistical area;

[0098] S12. According to the topography, altitude, field type, and tobacco growth in previous years, determine 20-50 field survey plots, each as one point;

[0099] S13. Use a sliding ruler to measure the height and width of the flue-cured tobacco canopy and the SPAD value; determine and calculate the LNI according to the prosperity index model, measure each point 10 times, and finally calculate the prosperity index value and take the average value, where:

[0100] Prosperity index = CW2×CH×SPa / 106×LNI

[0101] Where: CW is the canopy width of the plant; CH is the canopy height of the plant; SPa is the chlorophyll SPAD value;

[0102] LNI is the leaf number index, and the calculation formula is as follows:

[0103]

[0104] Wherein: L is the maximum leaf length of the leaf; W is the maximum leaf width of the leaf, k=1 to n, is the leaf number; C is the coefficient.

[0105] This technical solution determines the statistical area, selects representative field survey plots, uses a sliding ruler to accurately measure the height, width and SPAD value of the flue-cured tobacco canopy, and calculates the LNI based on the prosperity index model, ultimately obtaining an accurate and average flue-cured tobacco prosperity index value to improve the reliability and accuracy of the estimate.

[0106] According to one embodiment of the present invention, in step S13, Figure 2 and Figure 3 As shown in the figure, the steps for measuring the height and width of the flue-cured tobacco canopy with a sliding ruler are as follows:

[0107] S131, unfold the sliding ruler along the middle rotating part, which is divided into a horizontal ruler and a vertical ruler arranged vertically, wherein:

[0108] The horizontal ruler is 2m long, the scale unit is cm, and it is arranged from left to right;

[0109] The vertical ruler is 2m long, the scale unit is cm, and it is arranged from bottom to top;

[0110] S132. Slide left and right along the horizontal ruler to record the maximum canopy width of the flue-cured tobacco plant, accurate to cm;

[0111] S133. Slide up and down along the vertical ruler and record the vertical distance from the ground to the highest leaf, accurate to cm;

[0112] S134. The sliding ruler is folded along the middle rotating part to facilitate carrying.

[0113] This technical solution achieves accurate measurement of geometric features of flue-cured tobacco canopies through the operation steps of the sliding ruler, including unfolding the sliding ruler, recording canopy width and height with accuracy to centimeters, and providing accurate data for subsequent calculations.

[0114] According to one embodiment of the present invention, in step S2, the drone multispectral and RGB image acquisition includes the following specific steps:

[0115] S21, such as Figure 4 As shown, a clear day was selected to carry out drone photography at noon, and the multi-spectral and RGB color images of the flue-cured tobacco canopy were obtained by setting the drone's photography mode, route generation mode, flight speed, flight altitude, repetition rate on the main route, repetition rate between main routes, repetition rate between main routes, gimbal pitch angle, fixed-point mode, camera parameters, shooting time, and focal length;

[0116] S22, simultaneously taking a flying shot of the calibration plate to obtain an image;

[0117] S23. Check the image quality on site. If it is unqualified, retake the image in time.

[0118] This technical solution sets the UAV flight parameters and camera mode to collect high-quality multispectral and RGB images under suitable weather conditions, ensuring that clear and accurate images of the flue-cured tobacco canopy are obtained, laying the foundation for subsequent image processing and analysis.

[0119] According to an embodiment of the present invention, in step S3, image processing and data extraction include the following specific steps:

[0120] S31. Use the built-in algorithm of the drone to perform image correction and stitching on multispectral images and RGB images;

[0121] S32, in the spliced ​​image, using image segmentation to identify and extract the flue-cured tobacco canopy area, and extracting spectral data and flue-cured tobacco canopy texture in the segmented flue-cured tobacco canopy area, such as Figure 5 shown.

[0122] This technical solution uses the built-in algorithm of the drone to correct and stitch images, and uses image segmentation technology to extract the tobacco canopy area and its spectral data and texture features, providing key input data for subsequent model calculations.

[0123] According to an embodiment of the present invention, in step S4, inputting the model for calculation includes the following specific steps:

[0124] S41. The spectral data and the texture of the flue-cured tobacco canopy are input into the model to calculate the vegetation index of the flue-cured tobacco canopy; the vegetation index includes the following characteristic variable sets:

[0125] MeanB, MeanG, MeanR, MeanRE, and MeanNIR are the mean texture eigenvalues ​​of the Blue, Green, Red, Red Edge, and Near Infrared bands calculated based on the gray-level co-occurrence matrix GLCM, respectively;

[0126] HomoB and HomoR are the synergistic texture eigenvalues ​​of the Blue and Red bands calculated based on the gray-level co-occurrence matrix GLCM;

[0127] VarB, VarG, VarRE, and VarNIR are the variance texture eigenvalues ​​of the Blue, Green, RedEdge, and Near Infrared bands calculated based on the gray-level co-occurrence matrix GLCM;

[0128] CorrB and CorrR are the correlation texture eigenvalues ​​of the Blue and Red bands calculated based on the gray-level co-occurrence matrix GLCM;

[0129] Blue, Green, Red, and NIR are the spectral reflectances of the blue, green, red, and near-infrared bands of the flue-cured tobacco canopy, respectively.

[0130] S42, calculate the spectral index and texture data of the flue-cured tobacco canopy, such as Figure 6 as shown, and evaluate their weights.

[0131] This technical solution inputs spectral data and tobacco canopy texture into the model, calculates vegetation index and texture eigenvalues, and evaluates their weights to comprehensively reflect the growth status and characteristics of the tobacco canopy, providing a scientific basis for the estimation of the prosperity index.

[0132] Table 1 Characteristic variable set of flue-cured tobacco canopy texture

[0133]

[0134] According to an embodiment of the present invention, in step S5, the standardization and feature engineering includes the following specific steps:

[0135] S51. Data standardization execution: Apply standardization methods, such as Figure 7 As shown in the figure, the spectral index and texture data of the flue-cured tobacco canopy are standardized one by one; after the standardization is completed, the data distribution is checked immediately to ensure that no new abnormal values ​​are introduced during the standardization process, and to verify whether each characteristic value has been adjusted to the expected value range;

[0136] S52. Feature engineering selection and optimization: Use the Pearson correlation coefficient and mutual information index to calculate the correlation strength between each feature and the estimated target, and preliminarily identify the feature set that is highly correlated with the target; perform in-depth screening and optimization of the feature set to remove redundant and noise features; perform feature transformation operations on the screened key features, including square root transformation, feature crossover and aggregation.

[0137] This technical solution standardizes spectral index and texture data and uses statistical methods to screen and optimize feature sets to eliminate data differences, reduce noise and redundancy, and improve the efficiency and accuracy of model estimation.

[0138] According to an embodiment of the present invention, in step S6, estimating the flue-cured tobacco prosperity index includes the following specific steps:

[0139] S61, XGBoost model parameter configuration: configure the parameters of the XGBoost model, including learning rate, maximum depth, number of trees, and subsample ratio; use cross-validation method, GridSearchCV or RandomizedSearchCV, combined with the spectral index and texture data of the flue-cured tobacco canopy, to optimize and determine the best parameter combination suitable for the current feature set;

[0140] S62. Training of XGBoost model: Preprocess and fuse the spectral index and texture data of the flue-cured tobacco canopy through feature concatenation and feature conversion; train the XGBoost model based on the training set data containing these features.

[0141] This technical solution configures and optimizes the XGBoost model parameters and uses the training set data to train the model to build a machine learning model that can accurately estimate the flue-cured tobacco prosperity index.

[0142] According to one embodiment of the present invention, in step S7, Figure 8 As shown, determining the accuracy of the model includes the following specific steps:

[0143] S71. Use the trained XGBoost model to estimate the test data and obtain the estimated prosperity index. At the same time, perform the same preprocessing on the actual production data to verify the generalization ability and actual application effect of the model.

[0144] S72. Evaluate the accuracy of the model estimation results through mean square error, root mean square error, and mean absolute error.

[0145] This technical solution compares the model estimation results with the actual production data and uses a variety of error evaluation indicators to verify the accuracy and generalization ability of the model, ensuring that the model can reliably estimate the flue-cured tobacco prosperity index in practical applications.

[0146] Example 2

[0147] Based on Example 1, Figure 1 As shown, this embodiment provides a flue-cured tobacco prosperity index estimation device based on drone canopy spectrum and texture, comprising:

[0148] The prosperity index value measurement module is used to measure the SPAD value of flue-cured tobacco on the ground and preliminarily calculate the prosperity index value;

[0149] UAV flight shooting module, used to obtain multispectral images and RGB images of the flue-cured tobacco canopy;

[0150] The image processing and data extraction module is used to process the captured multispectral images and RGB images and extract the spectral data and texture of the flue-cured tobacco canopy;

[0151] Data processing and analysis module, used to standardize and feature engineer the extracted spectral and texture data, and construct input data for the estimation model;

[0152] The flue-cured tobacco prosperity index estimation model uses the XGBoost integrated algorithm to estimate the flue-cured tobacco prosperity index based on the processed feature data set;

[0153] The model accuracy assessment module is used to compare the measured prosperity index value with the estimated prosperity index value to determine the accuracy of the estimation model.

[0154] This technical solution first obtains the basic SPAD value and preliminary prosperity index value through the ground measurement module (prosperity index value measurement module); the SPAD value, as an indirect indicator of chlorophyll content, reflects the growth status of flue-cured tobacco; the UAV flight shooting module uses remote sensing technology to obtain large-scale, high-resolution multi-spectral and RGB images of flue-cured tobacco canopies from the air, making up for the shortcomings of ground measurement in spatial coverage; the image processing and data extraction module uses advanced image processing algorithms to correct, splice, segment and process the multi-spectral and RGB images taken by drones to extract the spectral data and texture characteristics of the flue-cured tobacco canopy; the spectral data reflects the reflection or emission characteristics of the canopy in different bands, while the texture characteristics describe the spatial structure information of the canopy; these characteristics together constitute the basic data for subsequent analysis; the data processing The processing and analysis module standardizes the extracted spectral data and texture data to ensure the comparability between different features; through feature engineering methods such as feature selection and feature transformation, the feature subsets that have an important impact on the estimation of the flue-cured tobacco prosperity index are screened out, and the input data suitable for the estimation model is constructed; the flue-cured tobacco prosperity index estimation model adopts the XGBoost integration algorithm. The advantage of the XGBoost algorithm is that it can handle large-scale data sets, automatically select features and adjust weights, and improve the overall performance by integrating multiple weak learners; the model accuracy evaluation module calculates the estimation error by comparing the measured prosperity index value and the estimated prosperity index value, thereby evaluating the accuracy of the model; if the estimation error is large, the performance of the model can be improved by adjusting model parameters, improving feature engineering, and other methods.

[0155] Although the present invention is described in detail by referring to the accompanying drawings and in combination with the preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, a person of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and these modifications or substitutions shall be within the scope of the present invention. Any person of ordinary skill in the art may easily think of changes or substitutions within the technical scope disclosed by the present invention, and these shall be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A flue-cured tobacco prosperity index estimation method based on drone canopy spectrum and texture, characterized in that: The steps include: S1. Measurement of SPAD value and prosperity index value: Select representative sampling plots according to topography and landforms; Use a sliding ruler to measure the height and width of the flue-cured tobacco canopy and determine the SPAD value; Calculate LNI according to the prosperity index model, and finally calculate the measured prosperity index value and take the average value; S2. UAV multispectral and RGB image acquisition: Use a UAV to fly and photograph the sampling plots to obtain multispectral images and RGB images of the flue-cured tobacco canopy; S3, image processing and data extraction: process the captured multispectral images and RGB images, and extract the spectral data and texture of the flue-cured tobacco canopy; S4. Inputting the model for calculation: inputting the spectral data and the texture of the flue-cured tobacco canopy into the model for calculation, and calculating the spectral index and texture data of the flue-cured tobacco canopy; S5. Standardization and feature engineering: Standardize the spectral index and texture data of the flue-cured tobacco canopy to ensure uniform dimensions and ranges; select key feature data sets through feature engineering to construct input data for the estimation model; S6. Estimation of flue-cured tobacco prosperity index: using XGBoost integrated algorithm, the flue-cured tobacco prosperity index is estimated based on the screened feature data set; S7, determine the accuracy of the model: compare the calculation results of step S1 and step S6 to obtain the estimation error, thereby determining the accuracy of the estimation model.

2. The method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture as claimed in claim 1, characterized in that: In step S1, the measurement of the SPAD value and the prosperity index value includes the following specific steps: S11. Determine the statistical area: select a certain range of base units and determine their statistical area; S12. According to the topography, altitude, field type, and tobacco growth in previous years, determine 20-50 field survey plots, each as one point; S13. Use a sliding ruler to measure the height and width of the flue-cured tobacco canopy and the SPAD value; determine and calculate the LNI according to the prosperity index model, measure each point 10 times, and finally calculate the prosperity index value and take the average value, where: Economic sentiment index = CW 2 ×CH× SPa / 10 6 ×LNI Where: CW is the canopy width of the plant; CH is the canopy height of the plant; SPa is the chlorophyll SPAD value; LNI is the leaf number index, and the calculation formula is as follows: Wherein: L is the maximum leaf length of the leaf; W is the maximum leaf width of the leaf, k=1 to n, is the leaf number; C is the coefficient.

3. The method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture as claimed in claim 2, characterized in that: In step S13, the steps of measuring the height and width of the flue-cured tobacco canopy with a sliding ruler are as follows: S131, unfold the sliding ruler along the middle rotating part, which is divided into a horizontal ruler and a vertical ruler arranged vertically, wherein: The horizontal ruler is 2m long, the scale unit is cm, and it is arranged from left to right; The vertical ruler is 2m long, the scale unit is cm, and it is arranged from bottom to top; S132. Slide left and right along the horizontal ruler to record the maximum canopy width of the flue-cured tobacco plant, accurate to cm; S133. Slide up and down along the vertical ruler and record the vertical distance from the ground to the highest leaf, accurate to cm; S134. The sliding ruler is folded along the middle rotating part to facilitate carrying.

4. The method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture as claimed in claim 1, characterized in that: In step S2, the drone multispectral and RGB image acquisition includes the following specific steps: S21. Select a clear day and take a drone photo at noon, and set the drone's photo mode, route generation mode, flight speed, flight altitude, repetition rate on the main route, repetition rate between main routes, repetition rate between main routes, gimbal pitch angle, fixed-point mode, camera parameters, shooting time, and focal length to obtain multispectral and RGB color images of the flue-cured tobacco canopy; S22, simultaneously taking a flying shot of the calibration plate to obtain an image; S23. Check the image quality on site. If it is unqualified, retake the image in time.

5. The method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture as claimed in claim 1, characterized in that: In step S3, image processing and data extraction include the following specific steps: S31. Use the built-in algorithm of the drone to perform image correction and stitching on multispectral images and RGB images; S32. In the spliced ​​image, use image segmentation to identify and extract the flue-cured tobacco canopy area, and extract spectral data and tobacco canopy texture in the segmented flue-cured tobacco canopy area.

6. The method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture as claimed in claim 1, characterized in that: In step S4, inputting the model for calculation includes the following specific steps: S41. The spectral data and the texture of the flue-cured tobacco canopy are input into the model to calculate the vegetation index of the flue-cured tobacco canopy; the vegetation index includes the following characteristic variable sets: MeanB, MeanG, MeanR, MeanRE, and MeanNIR are the mean texture eigenvalues ​​of the Blue, Green, Red, Red Edge, and Near Infrared bands calculated based on the gray-level co-occurrence matrix GLCM, respectively; HomoB and HomoR are the synergistic texture eigenvalues ​​of the Blue and Red bands calculated based on the gray-level co-occurrence matrix GLCM; VarB, VarG, VarRE, and VarNIR are the variance texture eigenvalues ​​of the Blue, Green, Red Edge, and Near Infrared bands calculated based on the gray-level co-occurrence matrix GLCM, respectively; CorrB and CorrR are the correlation texture eigenvalues ​​of the Blue and Red bands calculated based on the gray-level co-occurrence matrix GLCM; Blue, Green, Red, and NIR are the spectral reflectances of the blue, green, red, and near-infrared bands of the flue-cured tobacco canopy, respectively. S42. Calculate the spectral index and texture data of the flue-cured tobacco canopy, and evaluate their weights.

7. The method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture as claimed in claim 1, characterized in that: In step S5, standardization and feature engineering include the following specific steps: S51. Data standardization execution: Apply the standardization method to standardize the spectral index and texture data of the flue-cured tobacco canopy one by one; after completing the standardization, immediately perform a data distribution check to ensure that no new abnormal values ​​are introduced during the standardization process, and verify whether each characteristic value has been adjusted to the expected value range; S52, Feature Engineering Selection and Optimization: Use the Pearson correlation coefficient and mutual information index to calculate the correlation strength between each feature and the estimated target, and preliminarily identify the feature set that is highly correlated with the target; Perform deep screening and optimization on the feature set to remove redundant and noisy features; perform feature transformation operations on the screened key features, including square root transformation, feature crossover and aggregation.

8. The method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture as claimed in claim 1, characterized in that: In step S6, estimating the flue-cured tobacco prosperity index includes the following specific steps: S61, XGBoost model parameter configuration: configure the parameters of the XGBoost model, including learning rate, maximum depth, number of trees, and subsample ratio; use cross-validation method, GridSearchCV or RandomizedSearchCV, combined with the spectral index and texture data of the flue-cured tobacco canopy, to optimize and determine the best parameter combination suitable for the current feature set; S62. Training of XGBoost model: Preprocess and fuse the spectral index and texture data of the flue-cured tobacco canopy through feature concatenation and feature conversion; train the XGBoost model based on the training set data containing these features.

9. The method for estimating flue-cured tobacco prosperity index based on drone canopy spectrum and texture as claimed in claim 1, characterized in that: In step S7, determining the accuracy of the model includes the following specific steps: S71. Use the trained XGBoost model to estimate the test data and obtain the estimated prosperity index. At the same time, perform the same preprocessing on the actual production data to verify the generalization ability and actual application effect of the model. S72. Evaluate the accuracy of the model estimation results through mean square error, root mean square error, and mean absolute error.

10. A flue-cured tobacco prosperity index estimation device based on drone canopy spectrum and texture, using the flue-cured tobacco prosperity index estimation method based on drone canopy spectrum and texture as described in any one of claims 1 to 9, characterized in that: include: The prosperity index value measurement module is used to measure the SPAD value of flue-cured tobacco on the ground and preliminarily calculate the prosperity index value; UAV flight shooting module, used to obtain multispectral images and RGB images of the flue-cured tobacco canopy; The image processing and data extraction module is used to process the captured multispectral images and RGB images and extract the spectral data and texture of the flue-cured tobacco canopy; Data processing and analysis module, used to standardize and feature engineer the extracted spectral and texture data, and construct input data for the estimation model; The flue-cured tobacco prosperity index estimation model uses the XGBoost integrated algorithm to estimate the flue-cured tobacco prosperity index based on the processed feature data set; The model accuracy assessment module is used to compare the measured prosperity index value with the estimated prosperity index value to determine the accuracy of the estimation model.

Images