Thyme biomass prediction method based on multispectral data and machine learning

Abstract

The present application relates to a kind of thyme biomass prediction method based on multispectral data and machine learning, comprising: obtaining the multispectral image data of research area and pre-processing;Carry out single plant scale segmentation, and obtain the target extraction result of single plant scale;Calculate vegetation index;Using machine learning model, vegetation index is modeled, obtain biomass prediction model and training, carry out the automatic prediction of thyme biomass, obtain two prediction targets of dry weight and fresh weight.The present application utilizes unmanned aerial vehicle low-altitude multispectral image data to combine multiple machine learning models, establishes high-precision biomass prediction model, and real-time and accurately obtains the spatial distribution characteristics of large-area thyme biomass, provides basis for grassland ecosystem management and resource assessment;It improves the prediction accuracy, overcomes the limitation of single model through the prediction of 11 kinds of machine learning models, enhances the stability and reliability of prediction result.

Description

Technical Field

[0001] This invention relates to the field of ecological monitoring technology, and in particular to a method for predicting thyme biomass based on multispectral data and machine learning. Background Technology

[0002] Thyme, an aromatic plant belonging to the genus *Thymus* in the family Lamiaceae, is widely distributed in northern my country and possesses significant ecological and economic value. In recent years, researchers have discovered that thyme essential oil contains various alkenes and terpenes, which have positive effects on the treatment of diseases such as cancer and obesity. Therefore, thyme is increasingly being developed and applied in medical and other fields. Furthermore, thyme exhibits strong environmental adaptability, surviving in a variety of habitats. Its well-developed adventitious roots and strong extensibility enable it to play an important role in soil and water conservation and ecological restoration. Therefore, accurate estimation of thyme biomass provides an important reference for assessing its medicinal value and ensuring sustainable resource utilization.

[0003] Traditional biomass measurement methods mainly rely on manual field sampling and weighing. This method has many drawbacks, such as being time-consuming, labor-intensive, inefficient, and destructive; the limited number of sampling points makes it difficult to obtain spatial distribution information over large areas. Furthermore, manual measurements are susceptible to subjective factors, resulting in poor consistency and repeatability of the results. With the development of remote sensing technology, vegetation biomass estimation methods based on remote sensing data are gradually being applied.

[0004] Currently, scholars both domestically and internationally have conducted extensive research on remote sensing estimation of vegetation biomass. Studies have shown that the combination of UAV remote sensing and machine learning provides a reliable technical approach for vegetation biomass estimation. However, due to the small size of individual thyme plants, the spatial resolution of traditional satellite remote sensing data is insufficient for precise monitoring. Furthermore, most existing vegetation index methods are based on simple band combinations, failing to fully utilize the rich information contained in multispectral data. In terms of biomass estimation models, traditional statistical regression methods rely too heavily on single empirical models, making it difficult to accurately describe the complex nonlinear relationship between vegetation indices and biomass.

[0005] Therefore, how to utilize low-cost, high-efficiency UAV multispectral remote sensing technology, combined with advanced machine learning algorithms, to establish an accurate and reliable thyme biomass prediction model is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] To address the problems of low accuracy and efficiency in thyme biomass estimation in existing technologies, and the serious destructive nature of manual sampling, the present invention aims to provide a thyme biomass prediction method based on multispectral data and machine learning. This method utilizes low-altitude multispectral imagery data from unmanned aerial vehicles (UAVs) combined with various machine learning models to accurately acquire the spatial distribution characteristics of thyme biomass over large areas in real time, thereby improving prediction accuracy and enhancing the stability and reliability of prediction results.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting thyme biomass based on multispectral data and machine learning, the method comprising the following sequential steps:

[0008] (1) Acquire multispectral image data of the study area, including red band, near-infrared band and green band;

[0009] (2) Preprocess the acquired multispectral images;

[0010] (3) Perform single-tree scale segmentation on the preprocessed image, and use a combination of mask and vector file to extract the target region to obtain the single-tree scale target extraction result;

[0011] (4) Calculate vegetation index based on the target extraction results at the single-plant scale;

[0012] (5) The vegetation index is modeled using a machine learning model to obtain a biomass prediction model and train it. The multispectral image to be predicted is input into the trained biomass prediction model to automatically predict the biomass of thyme and obtain two prediction targets: dry weight and fresh weight.

[0013] Step (1) specifically refers to: using a UAV equipped with a multispectral camera, at a flight altitude of 15m, to acquire multispectral image data of the study area according to parameters of 80% forward overlap and 70% lateral overlap.

[0014] Step (2) includes the following steps in sequence:

[0015] (2a) Perform radiometric calibration and atmospheric correction on the acquired multispectral images;

[0016] (2b) Set the NDVI threshold range to 0 to 1, and automatically segment the soil background by adjusting the threshold;

[0017] (2c) Generate a binary mask image based on the set NDVI threshold, and mark the region with an NDVI value greater than the NDVI threshold as 1, i.e., the vegetation region, and the region with an NDVI value less than the threshold as 0, i.e., the background region.

[0018] (2d) Output the processed result after removing the background area, which is the preprocessed image data.

[0019] Step (3) includes the following steps in sequence:

[0020] (3a) Import the preprocessed image data;

[0021] (3b) Import the vector boundary file containing the target region boundary information;

[0022] (3c) Extract vegetation areas;

[0023] (3d) The extracted vegetation area is cropped according to the vector boundary file to obtain the target extraction result at the single-plant scale.

[0024] In step (4), the formula for calculating the vegetation index includes:

[0025] NDVI = (NIR - Red) / (NIR + Red);

[0026] GNDVI=(NIR-Green) / (NIR+Green);

[0027] RVI = NIR / Red;

[0028] SAVI=((NIR-Red)×1.5) / (NIR+Red+0.5);

[0029] OSAVI=((NIR-Red)×1.16) / (NIR+Red+0.16));

[0030] In the formula, NDVI is the normalized vegetation index, GNDVI is the normalized vegetation index in the green band, SAVI is the soil-regulated vegetation index, OSAVI is the optimized soil-regulated vegetation index; NIR represents the reflectance value in the near-infrared band, Red represents the reflectance value in the red band, Green represents the reflectance value in the green band, and RVI is the ratio vegetation index.

[0031] Step (5) specifically refers to: standardizing all vegetation index features using a standardization scaler; constructing a biomass prediction model containing the following 11 machine learning models, namely, Lasso regression, ridge regression, improved partial least squares regression, linear regression, improved KNN regression, improved decision tree regression, improved neural network regression, improved random forest regression, improved gradient boosting regression, improved support vector regression, and improved XGBoost regression; evaluating the performance of each machine learning model using cross-validation; and recording the prediction results and evaluation metrics of each machine learning model.

[0032] The improved random forest regression refers to hyperparameter optimization through randomized search cross-validation: the number of trees is 50, the maximum depth is 5, the minimum number of split samples is 2, and the minimum number of leaf node samples is 1; the improved gradient boosting regression refers to hyperparameter optimization through RandomizedSearchCV: the number of trees is 50, the learning rate is 0.01, the maximum depth is 3, the minimum number of split samples is 2, and the minimum number of leaf node samples is 2.

[0033] The improved partial least squares regression refers to fixing the number of principal components to 2; the improved neural network regression refers to optimizing hyperparameters through grid search cross-validation: the hidden layer structure is (50,50), the activation function is set to ReLU, the solver is set to Adam, and the learning rate is 0.001.

[0034] The improved support vector regression refers to hyperparameter optimization through grid search cross-validation: the kernel function type is set to linear, the regularization parameter is 1, and the gamma parameter is set to scale; the improved XGBoost regression refers to setting the objective function to mean squared error.

[0035] The improved KNN regression refers to hyperparameter optimization through grid search cross-validation: the number of neighbors is 5, and the weight scheme is set to distance; the improved decision tree regression refers to hyperparameter optimization through grid search cross-validation: the maximum depth is 10, the minimum number of split samples is 5, and the minimum number of leaf node samples is 2.

[0036] As can be seen from the above technical solution, the beneficial effects of the present invention are as follows: First, compared with the prior art, the present invention utilizes UAV low-altitude multispectral image data combined with multiple machine learning models to establish a high-precision biomass prediction model, accurately and in real time obtains the spatial distribution characteristics of thyme biomass over a large area, providing a basis for grassland ecosystem management and resource assessment; Second, the present invention applies multiple vegetation indices (NDVI, GNDVI, RVI, SAVI, OSAVI) to thyme biomass prediction, and through standardization processing and range constraint mechanisms, highlights the key factors of biomass estimation, improves prediction accuracy, and overcomes the limitations of a single model through the prediction of 11 machine learning models, enhancing the stability and reliability of the prediction results. Attached Figure Description

[0037] Figure 1 This is a flowchart of the method of the present invention;

[0038] Figure 2 This is a schematic diagram of the preprocessing interface;

[0039] Figure 3 This is a schematic diagram of the segmentation interface of a single plant.

[0040] Figure 4 This is a schematic diagram of the biomass prediction interface. Detailed Implementation

[0041] like Figure 1 As shown, a method for predicting thyme biomass based on multispectral data and machine learning is described, which includes the following sequential steps:

[0042] (1) Acquire multispectral image data of the study area, including red band, near-infrared band and green band;

[0043] (2) Preprocess the acquired multispectral images;

[0044] (3) Perform single-tree scale segmentation on the preprocessed image, and use a combination of mask and vector file to extract the target region to obtain the single-tree scale target extraction result;

[0045] (4) Calculate vegetation indices based on the target extraction results at the single-plant scale. The calculated vegetation indices form a dataset, which is then divided into a training set and a test set.

[0046] (5) A machine learning model is used to model the vegetation index, and a biomass prediction model is obtained. The model is trained using a training set. The multispectral image to be predicted is input into the trained biomass prediction model to automatically predict the biomass of thyme and obtain two prediction targets: dry weight and fresh weight.

[0047] Step (1) specifically refers to: using a UAV equipped with a multispectral camera, at a flight altitude of 15m, to acquire multispectral image data of the study area according to parameters of 80% forward overlap and 70% lateral overlap.

[0048] Step (2) includes the following steps in sequence:

[0049] (2a) Perform radiometric calibration and atmospheric correction on the acquired multispectral images;

[0050] (2b) Set the NDVI threshold range to 0 to 1, and automatically segment the soil background by adjusting the threshold;

[0051] (2c) Generate a binary mask image based on the set NDVI threshold, and mark the region with an NDVI value greater than the NDVI threshold as 1, i.e., the vegetation region, and the region with an NDVI value less than the threshold as 0, i.e., the background region.

[0052] (2d) Output the processed result after removing the background area, which is the preprocessed image data.

[0053] Step (3) includes the following steps in sequence:

[0054] (3a) Import the preprocessed image data;

[0055] (3b) Import the vector boundary file containing the target region boundary information;

[0056] (3c) The preprocessed image data is superimposed on the binarized mask image to extract the vegetation area;

[0057] (3d) The extracted vegetation area is cropped according to the vector boundary file to obtain the target extraction result at the single-plant scale.

[0058] In step (4), the formula for calculating the vegetation index includes:

[0059] NDVI = (NIR - Red) / (NIR + Red);

[0060] GNDVI=(NIR-Green) / (NIR+Green);

[0061] RVI = NIR / Red;

[0062] SAVI=((NIR-Red)×1.5) / (NIR+Red+0.5);

[0063] OSAVI=((NIR-Red)×1.16) / (NIR+Red+0.16));

[0064] In the formula, NDVI is the normalized vegetation index, GNDVI is the normalized vegetation index in the green band, SAVI is the soil-regulated vegetation index, OSAVI is the optimized soil-regulated vegetation index; NIR represents the reflectance value in the near-infrared band, Red represents the reflectance value in the red band, Green represents the reflectance value in the green band, and RVI is the ratio vegetation index.

[0065] Step (5) specifically refers to: standardizing all vegetation index features using a standardization scaler; constructing a biomass prediction model containing the following 11 machine learning models, namely, Lasso regression, ridge regression, improved partial least squares regression, linear regression, improved KNN regression, improved decision tree regression, improved neural network regression, improved random forest regression, improved gradient boosting regression, improved support vector regression, and improved XGBoost regression; evaluating the performance of each machine learning model using cross-validation; and recording the prediction results and evaluation metrics of each machine learning model.

[0066] Lasso regression is used to handle multicollinearity among features; Ridge regression is used to prevent overfitting through L2 regularization; Partial Least Squares regression is suitable when the number of features exceeds the sample size; Linear regression is used to establish a linear relationship between features and the target variable; KNN regression makes predictions based on nearest neighbor samples; Decision tree regression uses a tree structure for feature partitioning and prediction; Neural network regression uses a multilayer perceptron for nonlinear modeling; Random forest regression integrates the prediction results of multiple decision trees; Gradient boosting regression improves prediction performance through iteration; Support vector regression uses kernel functions to handle nonlinear problems; XGBoost regression uses a boosting algorithm optimized by second derivatives.

[0067] The improved random forest regression refers to hyperparameter optimization using randomized search cross-validation (Randomized SearchCV): the number of trees is 50, the maximum depth is 5, the minimum number of split samples is 2, and the minimum number of leaf node samples is 1. The improved gradient boosting regression refers to hyperparameter optimization using randomized search cross-validation: the number of trees is 50, the learning rate is 0.01, the maximum depth is 3, the minimum number of split samples is 2, and the minimum number of leaf node samples is 2.

[0068] The improved partial least squares regression refers to fixing the number of principal components to 2; the improved neural network regression refers to optimizing hyperparameters through grid search cross-validation (GridSearchCV): the hidden layer structure is (50,50), the activation function is set to ReLU, the solver is set to Adam, and the learning rate is 0.001.

[0069] The improved support vector regression refers to hyperparameter optimization through grid search cross-validation: the kernel function type is set to linear, the regularization parameter is 1, and the gamma parameter is set to scale; the improved XGBoost regression refers to setting the objective function to mean squared error.

[0070] The improved KNN regression refers to hyperparameter optimization through grid search cross-validation: the number of neighbors is 5, and the weight scheme is set to distance; the improved decision tree regression refers to hyperparameter optimization through grid search cross-validation: the maximum depth is 10, the minimum number of split samples is 5, and the minimum number of leaf node samples is 2. Figure 2 The NDVI-based soil segmentation interface is shown, including an input parameter setting area, a threshold adjustment area, a result display area, and an operation button area.

[0071] like Figure 3 As shown, the single-plant segmentation interface provides an operation process of "image upload → mask upload → SHP import → single-plant extraction", which allows users to easily complete the extraction process from the original image to the single-plant target.

[0072] Using NDVI segmentation results as a mask, the original multispectral image is masked and extracted to obtain pure vegetation area images; a pre-made vector boundary file (shp format) is imported, ensuring that the WGS1984 coordinate system is used, and the target image at the single tree scale is obtained by cropping based on the vector boundary.

[0073] The dataset was divided according to the following rules: biomass (fresh weight and dry weight) was measured for each test sample as label data, and vegetation index features at the corresponding locations were extracted as input features. Then, the dataset was randomly divided into training set and test set in a 7:3 ratio.

[0074] Table 1 Comparison of Fresh Weight Prediction Accuracy of Various Models

[0075] Model Training set R² Test set R² Training set RMSE Test set RMSE Linear Regression 0，72 0，73 29，57 24，15 Ridge Return 0，72 0，72 29，66 24，50 Lasso returns 0，72 0，71 29，73 24，83 Random Forest 0，85 0，67 21，95 26，72 Gradient boosting 0，96 0，48 10，90 33，44 Partial Least Squares 0，72 0，72 29，81 24，47 Neural Networks 0，73 0，69 29，12 25，75 Support Vectors 0，72 0，72 29，75 24，51 XGBoost 1，00 0，38 0，01 36，39 KNN 0，67 0，63 32，09 28，21 Decision Tree 0，80 0，63 25，33 27，99

[0076] Table 2 Comparison of Dry Weight Prediction Accuracy of Each Model

[0077] Model Training set R² Test set R² Training set RMSE Test set RMSE Linear Regression 0，91 0，86 8，62 9，24 Ridge Return 0，90 0，85 8，96 9，43 Lasso returns 0，89 0，84 9，26 9，83 Random Forest 0，96 0，80 5，75 11，02 Gradient boosting 0，99 0，74 2，81 12，63 Partial Least Squares 0，88 0，87 9，71 8，78 Neural Networks 0，91 0，82 8，62 10，49 Support Vectors 0，90 0，85 9，12 9，43 XGBoost 1，00 0，73 0，01 12，74 KNN 1，00 0，83 0，00 10，19 Decision Tree 0，91 0，64 8，62 14，73

[0078] Tables 1 and 2 show the prediction accuracy of different models on the training and test sets for fresh weight and dry weight.

[0079] like Figure 4 As shown, this invention provides an intuitive interface for displaying prediction results, including the following:

[0080] The band data input area is used to select image files in the red, near-infrared, and green bands;

[0081] The vegetation index display area calculates and displays the values ​​of various vegetation indices.

[0082] The prediction results display area shows the prediction results of each model.

[0083] This invention utilizes multispectral remote sensing data combined with various machine learning models to predict thyme biomass. By calculating multiple vegetation indices (NDVI, GNDVI, RVI, SAVI, and OSAVI) as feature inputs, 11 machine learning models were constructed, including linear regression, ridge regression, Lasso regression, random forest, gradient boosting, partial least squares regression, neural network regression, support vector regression, XGBoost regression, K-nearest neighbor regression, and decision tree regression. This invention provides an intuitive interface for displaying prediction results; users only need to input reflectance data for red, near-infrared, and green light bands to obtain biomass predictions. Experimental verification shows that this invention can accurately predict thyme biomass, providing important technical support for grassland ecosystem assessment and management.

Claims

1. A method for predicting thyme biomass based on multispectral data and machine learning, characterized in that: The method includes the following steps in sequence: (1) Acquire multispectral image data of the study area, including red band, near-infrared band and green band; (2) Preprocess the acquired multispectral images; (3) Perform single-tree scale segmentation on the preprocessed image, and use a combination of mask and vector file to extract the target region to obtain the single-tree scale target extraction result; (4) Calculate vegetation index based on the target extraction results at the single-plant scale; (5) The vegetation index is modeled using a machine learning model to obtain a biomass prediction model and train it. The multispectral image to be predicted is input into the trained biomass prediction model to automatically predict the biomass of thyme and obtain two prediction targets: dry weight and fresh weight. Step (3) includes the following steps in sequence: (3a) Import the preprocessed image data; (3b) Import the vector boundary file containing the target region boundary information; (3c) Extract vegetation areas; (3d) The extracted vegetation area is cropped according to the vector boundary file to obtain the target extraction result at the single-tree scale; In step (4), the formula for calculating the vegetation index includes: NDVI = (NIR - Red) / (NIR + Red); GNDVI=(NIR-Green) / (NIR+Green); RVI = NIR / Red; SAVI=((NIR-Red)×1.5) / (NIR+Red+0.5); OSAVI=((NIR-Red)×1.16) / (NIR+Red+0.16)); In the formula, NDVI is the normalized vegetation index, GNDVI is the normalized vegetation index in the green band, SAVI is the soil-regulated vegetation index, OSAVI is the optimized soil-regulated vegetation index, NIR represents the reflectance value in the near-infrared band, Red represents the reflectance value in the red band, Green represents the reflectance value in the green band, and RVI is the ratio vegetation index.

2. The method for predicting thyme biomass based on multispectral data and machine learning according to claim 1, characterized in that: Step (1) specifically refers to: using a UAV equipped with a multispectral camera, at a flight altitude of 15m, to acquire multispectral image data of the study area according to parameters of 80% forward overlap and 70% lateral overlap.

3. The method for predicting thyme biomass based on multispectral data and machine learning according to claim 1, characterized in that: Step (2) includes the following steps in sequence: (2a) Perform radiometric calibration and atmospheric correction on the acquired multispectral images; (2b) Set the NDVI threshold range to 0 to 1, and automatically segment the soil background by adjusting the threshold; (2c) Generate a binary mask image based on the set NDVI threshold, and mark the region with an NDVI value greater than the NDVI threshold as 1, i.e., the vegetation region, and the region with an NDVI value less than the threshold as 0, i.e., the background region. (2d) Output the processed result after removing the background area, which is the preprocessed image data.

4. The method for predicting thyme biomass based on multispectral data and machine learning according to claim 1, characterized in that: Step (5) specifically refers to: standardizing all vegetation index features using a standardization scaler; constructing a biomass prediction model containing the following 11 machine learning models, namely, Lasso regression, ridge regression, improved partial least squares regression, linear regression, improved KNN regression, improved decision tree regression, improved neural network regression, improved random forest regression, improved gradient boosting regression, improved support vector regression, and improved XGBoost regression; evaluating the performance of each machine learning model using cross-validation; and recording the prediction results and evaluation metrics of each machine learning model.

5. The thyme biomass prediction method based on multispectral data and machine learning according to claim 4, characterized in that: The improved random forest regression refers to hyperparameter optimization through randomized search cross-validation: the number of trees is 50, the maximum depth is 5, the minimum number of split samples is 2, and the minimum number of leaf node samples is 1; the improved gradient boosting regression refers to hyperparameter optimization through RandomizedSearchCV: the number of trees is 50, the learning rate is 0.01, the maximum depth is 3, the minimum number of split samples is 2, and the minimum number of leaf node samples is 2.

6. The method for predicting thyme biomass based on multispectral data and machine learning according to claim 4, characterized in that: The improved partial least squares regression refers to fixing the number of principal components to 2; the improved neural network regression refers to optimizing hyperparameters through grid search cross-validation: the hidden layer structure is (50,50), the activation function is set to ReLU, the solver is set to Adam, and the learning rate is 0.

001.

7. The thyme biomass prediction method based on multispectral data and machine learning according to claim 4, characterized in that: The improved support vector regression refers to hyperparameter optimization through grid search cross-validation: the kernel function type is set to linear, the regularization parameter is 1, and the gamma parameter is set to scale; the improved XGBoost regression refers to setting the objective function to mean squared error.

8. The method for predicting thyme biomass based on multispectral data and machine learning according to claim 4, characterized in that: The improved KNN regression refers to hyperparameter optimization through grid search cross-validation: the number of neighbors is 5, and the weight scheme is set to distance; the improved decision tree regression refers to hyperparameter optimization through grid search cross-validation: the maximum depth is 10, the minimum number of split samples is 5, and the minimum number of leaf node samples is 2.

Images