Pennisetum alopecuroides cold resistance evaluation method based on unmanned aerial vehicle multispectral image

Abstract

The invention belongs to the technical field of agricultural remote sensing, and discloses a method for evaluating cold resistance of pennisetum alopecuroides based on unmanned aerial vehicle multispectral images, which comprises the following steps: acquiring agronomic characters related to cold resistance, such as chlorophyll content, overwintering rate and the like of the pennisetum alopecuroides before and after low-temperature stress through manual measurement, and acquiring the multispectral images by using an unmanned aerial vehicle; the image is preprocessed, and various vegetation indexes such as DVI, NDVI and RVI are extracted; on the basis of agronomic character data, a comprehensive cold resistance index CRI is constructed by adopting principal component analysis and a membership function method; using the vegetation index as an independent variable and CRI as a dependent variable, and using a machine learning algorithm to construct a prediction model; and finally, realizing rapid and nondestructive evaluation on the cold resistance of a large-range pennisetum alopecuroides group by utilizing the model, and outputting a five-level cold resistance grading result according to a CRI value. According to the method, the problems of low efficiency and high difficulty of traditional manual evaluation are solved, high-throughput and precise cold resistance identification is realized, and reliable technical support is provided for cold-resistant breeding and cultivation management of pennisetum alopecuroides.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing technology, specifically to a method for evaluating the cold resistance of Napier grass based on UAV multispectral imagery. Background Technology

[0002] Pennisetum alopecuroides, a perennial herbaceous plant belonging to the genus Pennisetum in the subfamily Sorghum of the Poaceae family, is the tallest and highest-yielding forage grass in the Poaceae family, with a yield of 12-15 tons per mu (approximately 0.16 acres). As a high-quality forage, Pennisetum alopecuroides is widely cultivated in tropical and subtropical regions. It not only possesses excellent stress resistance and tolerance to poor soil conditions but also demonstrates significant value in various fields such as soil ecological restoration, landscaping, windbreak and sand fixation, manure disposal, and papermaking. In animal husbandry, the application of Pennisetum alopecuroides can significantly improve livestock productivity and the quality of livestock products, resulting in substantial economic benefits. However, Pennisetum alopecuroides is highly sensitive to low-temperature stress and cannot safely overwinter in the cold mountainous areas of the Yangtze River basin and areas north of the Yangtze River, leading to a shortened growth cycle and reduced yield and quality, greatly limiting its widespread application. Furthermore, due to its excessive height, large-scale field evaluation of its cold resistance is extremely difficult, hindering its large-scale implementation. Moreover, artificial cold resistance assessment can introduce significant errors, posing a major challenge to the screening of cold-resistant germplasm and the breeding of cold-resistant varieties.

[0003] Multispectral sensors can detect the reflectance spectra of visible and near-infrared wavelengths, capturing the differences in reflectance of objects across different wavelengths while maintaining spatial resolution. This allows them to respond to crop growth...

[0004] It has advantages in terms of biochemical indicators and morphological changes. Unmanned aerial vehicles (UAVs) are unmanned aircraft controlled by radio remote control equipment and their own program control devices. They are characterized by their small size, low cost, and simple operation. Using UAVs equipped with high-resolution cameras, multispectral sensors, and other equipment, precise monitoring of crop growth status can be achieved.

[0005] Although this technology has broad application prospects and similar studies have been conducted on other crops, the plant types and leaf spectral reflectance characteristics vary greatly among different plants. Currently, there is a lack of a comprehensive UAV multispectral analysis method for evaluating the cold resistance of *Pennisetum alopecuroides*, particularly a technical system that can organically combine multispectral remote sensing indicators with actual cold resistance phenotypic data to achieve rapid and accurate graded evaluation. Therefore, there is an urgent need to develop a comprehensive evaluation method for the cold resistance of *Pennisetum alopecuroides* based on UAV multispectral imagery to provide reliable technical support for its cold resistance breeding and cultivation promotion. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a method for evaluating the cold resistance of Napier grass based on UAV multispectral imagery. This invention is achieved through the following technical solution.

[0007] A method for evaluating the cold resistance of Napier grass based on UAV multispectral imagery includes the following steps:

[0008] S1. Data Acquisition: Field measurements of agronomic traits and acquisition of UAV multispectral images were carried out simultaneously before and after frost stress on the target Napier grass population.

[0009] S2. Data processing: Preprocess the multispectral image and extract multiple vegetation indices;

[0010] S3. Construction of cold resistance index: Based on the agronomic traits, the cold resistance index (CRI) of individual plants of Napier grass was calculated using principal component analysis and membership function method.

[0011] The cold resistance index (CRI) refers to the cold resistance of Napier grass. The higher the CRI value, the better the cold resistance of Napier grass.

[0012] S4. Model Construction: Using the vegetation index as the independent variable and the cold resistance index CRI as the dependent variable, a cold resistance prediction model is constructed using machine learning algorithms.

[0013] S5. Evaluation and Output: The cold resistance of the Napier grass population is predicted using the trained cold resistance prediction model, and the cold resistance evaluation level is output according to the predetermined CRI grading threshold.

[0014] As a further embodiment of the present invention, in step S1, the agronomic traits include chlorophyll content, leaf width, leaf length, crown width and plant height, and the acquisition bands of the multispectral image include at least the green band, red band, red edge band and near-infrared band.

[0015] As a further aspect of the present invention, in step S1, the acquisition of multispectral images by the UAV follows predetermined flight parameters, including flight altitude, forward overlap rate, and lateral overlap rate, to ensure image stitching accuracy.

[0016] As a further aspect of the present invention, in step S2, the vegetation index includes at least one of the following: Differential Vegetation Index (DVI), Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Optimized Soil-Regulated Vegetation Index (OSAVI), Ratio Vegetation Index (RVI), and Soil-Regulated Vegetation Index (SAVI).

[0017] As a further aspect of the present invention, in step S3, the cold resistance index (CRI) is calculated as follows:

[0018] S31. Perform principal component analysis on multiple agronomic traits and extract principal components whose cumulative contribution rate exceeds a predetermined threshold.

[0019] S32. Calculate the membership function value for each principal component;

[0020] S33. Calculate the weight of each principal component based on its contribution rate;

[0021] S34. The membership function values ​​of each principal component are summed with their weights to obtain the cold resistance index CRI.

[0022] As a further aspect of the present invention, in step S4, the machine learning algorithm is a random forest regression algorithm;

[0023] The model building steps also include: optimizing the hyperparameters of the random forest model through grid search GridSearchCV and cross-validation, wherein the hyperparameters include at least one of the following: number of trees, maximum tree depth, minimum number of leaf node samples, and maximum number of features.

[0024] As a further aspect of the present invention, step S4 includes the following step before constructing the model: performing a correlation analysis between the vegetation index and the cold resistance index CRI, and selecting the vegetation index with the highest correlation to CRI as the input feature for constructing a linear or nonlinear regression model.

[0025] As a further aspect of the present invention, in step S5, the predetermined CRI grading threshold divides cold resistance into the following five levels:

[0026] Sensitivity: CRI ≤ 0.328;

[0027] Non-resistant: 0.328 < CRI ≤ 0.395;

[0028] Antimicrobial efficacy: 0.395 < CRI ≤ 0.421;

[0029] Cold resistance: 0.421 < CRI ≤ 0.446;

[0030] Extremely cold resistant: CRI > 0.446.

[0031] As a further aspect of the present invention, after the model is constructed, the vegetation indices are ranked by feature importance, with the ratio vegetation index (RVI) being the most important.

[0032] As a further aspect of the present invention, in step S5, the output of the cold resistance evaluation level result includes generating and outputting a spatial distribution map of cold resistance grading.

[0033] The beneficial effects of this invention are as follows:

[0034] 1. Achieved high-throughput, rapid, and non-destructive accurate evaluation of the cold resistance of Napier grass: Through UAV multispectral remote sensing technology, image data of large-area Napier grass populations can be quickly acquired, replacing the traditional heavy manual measurement. The entire process, from data acquisition to model evaluation, greatly improves efficiency and does not cause any damage to the plants, achieving non-destructive testing. This solves the problems of difficulty, low efficiency, and difficulty in large-scale application of manual field evaluation due to the tall size of Napier grass plants.

[0035] 2. Improved the objectivity, accuracy, and robustness of cold resistance assessment: Innovatively combining UAV multispectral data (6 vegetation indices) with ground-measured agronomic trait data, a comprehensive cold resistance index (CRI) was constructed using principal component analysis and membership function method, avoiding the limitations of a single indicator. Furthermore, machine learning algorithms (especially the optimized random forest model) were used to build a predictive model, achieving an R² score of over 0.85 on the test set, significantly improving prediction accuracy and model stability. This addresses the problem of insufficient accuracy and reliability in existing technologies due to the lack of effective methods for organically combining multispectral data with cold resistance phenotypic data.

[0036] 3. This invention provides a scalable technical framework, offering direct decision support for cold-resistant breeding and cultivation management: It is not merely a method, but a complete technical solution (including data collection specifications, index construction models, prediction models, and grading standards). The final output is an intuitive cold-resistance grading (5 levels) and spatial distribution map. Breeders can quickly screen for superior cold-resistant germplasm resources, and farmers can accurately grasp the distribution of cold resistance in the field, providing a clear basis for cultivation management measures. This methodology also has good scalability and promotion potential for evaluating the stress resistance of other forage grasses or crops with similar plant types, solving the problem of existing methods lacking a complete and usable technical system to support the cold-resistant breeding and cultivation promotion of Napier grass. Attached Figure Description

[0037] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the specific embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 Flowchart for evaluating the cold resistance of Pennisetum arvense;

[0039] Figure 2 This is a graph showing the cold resistance index prediction based on the RF model.

[0040] Figure 3 This is a ranking plot of vegetation indices based on the RF model.

[0041] Figure 4 This is a grading prediction diagram of cold resistance of Napier grass based on the RF model. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] like Figures 1-4 As shown, the present invention has the following specific embodiments.

[0044] Example

[0045] 1. Measurement of agronomic traits:

[0046] The experiment was conducted at the Sichuan Agricultural University Chongzhou Base (103.64°E, 30.56°N). The day the temperature first dropped to 0°C was designated as the first day of frost. Five agronomic traits of individual *Pennisetum affine* plants were measured, including chlorophyll, leaf width, leaf length, crown width, and plant height, on the day before the frost and the third day after the frost when a distinct phenotype was observed.

[0047] 2. Multispectral image acquisition:

[0048] In December 2024, multispectral image data was acquired from one day before the frost to six days after the frost. The multispectral acquisition equipment used was a DJI Mavic 3M Multispectral UAV, with the following flight path and parameters pre-set: flight altitude 12m, forward overlap 70%, and lateral overlap 80%.

[0049] 3. Multispectral image preprocessing:

[0050] Image correction and stitching software is used to preprocess the UAV multispectral images. The preprocessing includes radiometric correction, smoothing, sharpening and geometric correction of the multispectral images to obtain orthophotos. Then ArcGIS is used to fuse and crop the images.

[0051] 4. Vegetation index extraction:

[0052] ENVI was used to extract six vegetation indices related to cold resistance from the processed Napier grass images, including DVI (Difference Vegetation Index), NDVI (Normalized Difference Vegetation Index), GNDVI (Green Normalized Difference Vegetation Index), OSAVI (Optimized Soil-Regulated Vegetation Index), RVI (Ratio Vegetation Index), and SAVI (Soil-Regulated Vegetation Index).

[0053] 5. Construction of the cold resistance index:

[0054] (1) As shown in Table 1, principal component analysis was performed on five agronomic traits (chlorophyll, leaf width, leaf length, crown width and plant height) using the Social Science Statistical Software Package (SPSS) to confirm the number of four principal components.

[0055] Table 1. Principal Component Analysis Results

[0056]

[0057] (2) The membership function method is used for comprehensive evaluation to construct the cold resistance index. The relevant formula is as follows:

[0058] F(X i ) = a li X li +a 2i X 2i +…+a pi X pi (p=1,2,…,n;i=1,2,…,n)

[0059] μ(X i ) = (X i -X imin ) / (X imax -X imin (i=1,2,…,n)

[0060] In the formula: F(X) i ) represents the comprehensive index value of the i-th principal component; a pi X is the eigenvector corresponding to the eigenvalues ​​of each index of the i-th principal component; pi μ(X) represents the standardized value of the original variable. i ) represents the membership function value of the i-th comprehensive index; X i X is the i-th comprehensive index; imin X is the minimum value of the i-th comprehensive index; imax It represents the maximum value of the i-th comprehensive index.

[0061] The weighting formulas for each comprehensive indicator are as follows:

[0062]

[0063] In the formula: w i p represents the weight of the i-th comprehensive index; i The contribution rate of the i-th comprehensive indicator for variety.

[0064] The formula for calculating the cold resistance index is as follows:

[0065]

[0066] Where D represents the calculated result of the cold resistance index (CRI).

[0067] (3) Evaluation of the cold resistance index effect:

[0068] As shown in Table 2, the average cold resistance index before frost was 0.446, and the average cold resistance index after frost was 0.395. Therefore, the higher the cold resistance index, the better the cold resistance.

[0069] Table 2. Average values ​​of cold resistance index and agronomic traits before and after frost.

[0070]

[0071] 6. Model dataset construction:

[0072] The correlation between vegetation indices and cold hardiness index was analyzed using SPSS. Based on the correlation analysis results, the six vegetation indices were used as independent variables (x) and the cold hardiness index was used as dependent variable (y) through linear regression, four types of nonlinear regression (quadratic function, exponential function, logarithmic function, and power function) and random forest (RF) algorithms.

[0073] 7. Model Training and Optimization:

[0074] (1) Linear model and four nonlinear models

[0075] Based on the correlation results between vegetation indices and cold hardiness indices, the cold hardiness index has the highest correlation coefficient with RVI (ratio vegetation index). Therefore, a model of cold hardiness index and RVI is established. RVI is selected as the independent variable, and linear and four nonlinear regression models are chosen.

[0076] Linear model: y = α + bx

[0077] Exponential model: y = α * e bx

[0078] Logarithmic model: y = α + b * lnx

[0079] Quadratic function model: y = α + bx + cx 2

[0080] Power function model: y = α * x b

[0081] In the formula, x represents the optimal spectral parameters, y represents the predicted values ​​of yield and agronomic parameters, and α, b, and c represent constants. The test set results are shown in Table 3.

[0082] Table 3. Test set model of cold resistance based on RVI and cold resistance index

[0083]

[0084] (2) Random Forest Model

[0085] A prediction model was built using the random forest regression algorithm. The model hyperparameters were optimized using grid search (GridSearchCV) and 5-fold cross-validation. The hyperparameters included the number of decision trees (n_estimators), the maximum depth of the trees (max_depth), the minimum number of leaf node samples (min_samples_leaf), and the maximum number of features (max_features).

[0086] The optimized optimal parameter combination is: 100 decision trees, maximum depth 15, minimum number of leaf node samples 1, and maximum number of features "sqrt".

[0087] The model's best coefficient of determination R² on the test set is 0.8369043678784853, coefficient of determination R² is 0.8515760172235226, root mean square error RMSE is 0.00027380200659701993, and mean absolute error MAE is 0.0126.

[0088] 8. Evaluation and Result Output of Cold Resistance Model:

[0089] Using R 2 RMSE is generally used as a model evaluation metric. 2 The larger the value of , the smaller the value of RMSE, indicating that the established model has better performance and higher prediction accuracy.

[0090] (1) Constructing a cold resistance prediction model: actual value vs. predicted value

[0091] I. Construction of Linear and Nonlinear Models

[0092] Use SPSS to build a regression model and output the R-squared of the model. 2 The RMSE index results are shown in Table 4.

[0093] Table 4. Predictive Models for Cold Resistance Based on RVI and Cold Resistance Index

[0094]

[0095] II. Construction of Random Forest (RF) Model

[0096] like Figure 2 As shown, a cold resistance prediction model was built using Python. The prediction result of the RF model is R. 2= 0.852, RMSE = 0.017, and output a scatter plot of the actual value vs. the predicted value. Figure 2 The horizontal axis represents the measured value of the cold resistance index (the axis where the red dot is located), and the vertical axis represents the predicted value of the cold resistance index (the axis where the blue dot is located).

[0097] Python is a cross-platform, open-source, general-purpose high-level programming language.

[0098] R 2 R0 is the coefficient of determination, used to measure the proportion of variation in the dependent variable (cold hardiness index CRI). Its value ranges from [0,1]. The closer the value is to 1, the better the model fit and the stronger the linear relationship between the predicted and measured values. 2 =0.852, indicating that the model can explain 85.2% of the CRI variation, with only 14.8% of the variation caused by other factors not included in the model.

[0099] RMSE is the root mean square error of a model, which measures the root mean square of the sum of squares of the errors between predicted and measured values. The smaller the value, the smaller the prediction error and the higher the model accuracy. RMSE=0.017 means that the CRI value predicted by the model differs from the measured value by an average of 0.017 units.

[0100] Figure 2 In the diagram, red dots represent measured values ​​of the cold resistance index, and blue dots represent predicted values. The distribution of red and blue dots in the diagram is close to the diagonal, indicating that the error between the measured and predicted values ​​is small.

[0101] In summary, among linear function models, quadratic function models, exponential function models, logarithmic function models, power function models, and RF models, the RF model R... 2 The largest RMSE and the smallest prediction result are obtained from the linear function model and the exponential function model R. 2 The results are the same, but the exponential function model has the largest RMSE, so the exponential function model has the lowest prediction accuracy.

[0102] (2) Ranking the vegetation index features by importance

[0103] like Figure 3 As shown, a ranking map of the importance of vegetation index features based on the Random Forest (RF) model was generated using Python. Figure 3 The horizontal axis represents the vegetation index category, and the vertical axis represents the importance weight of the index. Vegetation indices that contribute significantly to cold resistance prediction are selected.

[0104] Contribution value ranking: RVI (Ratio Vegetation Index) > NDVI (Normalized Difference Vegetation Index) > OSAVI (Optimized Soil-Regulated Vegetation Index) > SAVI (Soil-Regulated Vegetation Index) > GNDVI (Green Normalized Difference Vegetation Index) > DVI (Difference Vegetation Index).

[0105] 9. Cold resistance grading and visualization:

[0106] like Figure 4 As shown, the predicted results are divided into 5 cold resistance levels based on the cold resistance index: Sensitive: D ≤ 0.328; Not resistant: 0.328 ≤ D < 0.395; Moderately resistant: 0.395 ≤ D < 0.421; Cold resistant: 0.421 ≤ D < 0.446; Extremely cold resistant: 0.446 ≤ D. A scatter plot of the cold resistance classification is then output using graphing software. Figure 4 The horizontal axis represents the measured value of the cold resistance index, and the vertical axis represents the predicted value of the cold resistance index.

[0107] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery, characterized in that, Includes the following steps: S1. Data Acquisition: Field measurements of agronomic traits and acquisition of UAV multispectral images were carried out simultaneously before and after frost stress on the target Napier grass population. S2. Data processing: Preprocess the multispectral image and extract multiple vegetation indices; S3. Construction of cold resistance index: Based on the agronomic traits, the cold resistance index (CRI) of individual plants of Napier grass was calculated using principal component analysis and membership function method. The cold resistance index (CRI) refers to the cold resistance of Napier grass. The higher the CRI value, the better the cold resistance of Napier grass. S4. Model Construction: Using the vegetation index as the independent variable and the cold resistance index CRI as the dependent variable, a cold resistance prediction model is constructed using machine learning algorithms. S5. Evaluation and Output: The cold resistance of the Napier grass population is predicted using the trained cold resistance prediction model, and the cold resistance evaluation level is output according to the predetermined CRI grading threshold.

2. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 1, characterized in that: In step S1, the agronomic traits include chlorophyll content, leaf width, leaf length, crown width, and plant height, and the acquisition bands of the multispectral images include at least the green band, red band, red edge band, and near-infrared band.

3. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 1, characterized in that: In step S1, the acquisition of multispectral images by the UAV follows predetermined flight parameters, including flight altitude, forward overlap rate, and lateral overlap rate, to ensure image stitching accuracy.

4. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 1, characterized in that: In step S2, the vegetation index includes at least one of the following: Differential Vegetation Index (DVI), Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Optimized Soil-Regulated Vegetation Index (OSAVI), Ratio Vegetation Index (RVI), and Soil-Regulated Vegetation Index (SAVI).

5. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 1, characterized in that, In step S3, the cold resistance index (CRI) is calculated as follows: S31. Perform principal component analysis on multiple agronomic traits and extract principal components whose cumulative contribution rate exceeds a predetermined threshold. S32. Calculate the membership function value for each principal component; S33. Calculate the weight of each principal component based on its contribution rate; S34. The membership function values ​​of each principal component are summed with their weights to obtain the cold resistance index CRI.

6. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 1, characterized in that: In step S4, the machine learning algorithm is the random forest regression algorithm; The model building steps also include: optimizing the hyperparameters of the random forest model through grid search GridSearchCV and cross-validation, wherein the hyperparameters include at least one of the following: number of trees, maximum tree depth, minimum number of leaf node samples, and maximum number of features.

7. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 6, characterized in that, In step S4, before constructing the model, the following steps are also included: performing a correlation analysis between the vegetation index and the cold resistance index CRI, and selecting the vegetation index with the highest correlation to CRI as the input feature for constructing a linear or nonlinear regression model.

8. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 1, characterized in that, In step S5, the predetermined CRI grading threshold divides cold resistance into the following five levels: Sensitivity: CRI ≤ 0.328; Non-resistant: 0.328 < CRI ≤ 0.395; Antimicrobial efficacy: 0.395 < CRI ≤ 0.421; Cold resistance: 0.421 < CRI ≤ 0.446; Extremely cold resistant: CRI > 0.

446.

9. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 7, characterized in that: After the model was built, the vegetation indices were ranked by feature importance, with the ratio vegetation index (RVI) being the most important.

10. The method for evaluating the cold resistance of *Phragmites australis* based on UAV multispectral imagery according to claim 1, characterized in that: In step S5, the output of cold resistance evaluation results includes generating and outputting a spatial distribution map of cold resistance grading.

Images