A Method and System for Inverting Nutrient Content in Eucalyptus Canopy Leaves Based on Multi-Feature Fusion

Abstract

This invention discloses a method and system for inverting the nutrient content of eucalyptus canopy leaves based on multi-feature fusion, belonging to the field of UAV remote sensing and forestry information technology. The method includes: acquiring and preprocessing multispectral image data from a UAV; extracting vegetation index features and texture features of the eucalyptus canopy; filtering the fused features based on mutual information to construct a multi-feature fusion dataset; constructing an inversion model for the nitrogen, phosphorus, and potassium content of eucalyptus leaves using a random forest regression algorithm; and verifying and optimizing the accuracy of the inversion model. This invention effectively compensates for the information gaps of single features by fusing spectral and spatial structure information, significantly improving the inversion accuracy of eucalyptus canopy leaf nutrient content, and providing reliable technical support for precise nutrient management of eucalyptus plantations.

Description

Technical Field

[0001] This invention belongs to the field of UAV remote sensing and forestry information technology, specifically relating to a method and system for inverting the nutrient content of eucalyptus canopy leaves based on multi-feature fusion. Background Technology

[0002] The nitrogen, phosphorus, and potassium content of eucalyptus canopy leaves is a core indicator reflecting the growth status of eucalyptus and the balance of nutrient supply and demand. Accurate and rapid monitoring of leaf nutrition status is a key prerequisite for achieving scientific fertilization of eucalyptus plantations, improving nutrient utilization efficiency, and reducing forestry production costs.

[0003] Traditional methods for monitoring the nutrient content of eucalyptus leaves rely on manual field sampling and laboratory chemical analysis. This method requires setting up a large number of sampling points in the study area, collecting leaf samples, and then determining the nutrient content through chemical means such as digestion and titration. This method suffers from problems such as long monitoring cycles, high manpower and material costs, limited representativeness of sampling points, and inability to achieve large-scale continuous monitoring. It is difficult to meet the requirements of modern precision nutrient management of eucalyptus plantations for monitoring timeliness, spatial accuracy, and comprehensive coverage.

[0004] In recent years, UAV remote sensing technology has been rapidly applied in the field of vegetation nutrient status monitoring due to its advantages of high spatiotemporal resolution, flexible operation, low cost, and non-contact monitoring. By equipping UAV platforms with multispectral sensors, spectral reflectance information of eucalyptus canopy can be quickly acquired, and leaf nutrient content can be retrieved by combining it with vegetation indices. However, most existing studies on UAV remote sensing for retrieving eucalyptus leaf nutrient content use only single vegetation index features or single texture features for modeling, failing to fully explore the complementarity between spectral and spatial structural information: vegetation indices can only reflect the physiological and biochemical characteristics of leaves and cannot reflect structural information such as the spatial distribution and growth uniformity of the canopy; while single texture features are insufficient to reflect the nutrient and biochemical responses inside the leaves, resulting in limited accuracy and stability of the retrieval model, making it difficult to meet the actual needs of precision nutrient management.

[0005] Meanwhile, existing inversion models often employ simple correlation analysis for feature selection, which easily retains redundant features, leading to multicollinearity and reduced model computational efficiency. Furthermore, some models use traditional linear regression algorithms, which are insufficient for fitting the nonlinear relationship between spectral and nutrient content, further impacting inversion accuracy. Therefore, developing an inversion method for nutrient content in eucalyptus canopy leaves based on the fusion of multiple feature types, optimized feature selection methods, and improved model fitting capabilities has become an urgent need in the field of precision management of eucalyptus plantations. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for inverting the nutrient content of eucalyptus canopy leaves based on multi-feature fusion. By fusing vegetation index features and texture features, the inversion accuracy of nitrogen, phosphorus and potassium content in eucalyptus leaves can be effectively improved.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: The method for inverting nutrient content in eucalyptus canopy leaves based on multi-feature fusion includes the following steps: S1: Acquire and preprocess multispectral image data from UAVs. A UAV platform equipped with a multispectral sensor was used to acquire multispectral image data of the eucalyptus canopy in the study area. The multispectral sensor must cover at least five core bands: blue (440-460nm), green (540-560nm), red (650-670nm), red edge (710-730nm), and near-infrared (830-850nm). During flight, a forward overlap of 70%-80% and a lateral overlap of 60%-70% were set to ensure the integrity of the image stitching.

[0008] The acquired raw multispectral images underwent standardized preprocessing, including radiometric calibration, geometric correction, image stitching, and canopy extraction. Radiometric calibration converted the raw DN values ​​of the images into surface reflectance, eliminating the sensor's own radiometric errors. Geometric correction, based on ground control points (GCPs) deployed in the field and the digital elevation model (DEM) of the study area, employed a polynomial correction algorithm to eliminate image geometric distortions caused by terrain undulations and UAV flight attitude. Image stitching used a seamless stitching algorithm to stitch together multiple images of the study area, generating a global orthophoto image. Canopy extraction employed the NDVI thresholding method, setting the threshold to 0.25-0.35 to separate the eucalyptus canopy area from the background areas such as soil and weeds, outputting a standardized image containing only the canopy.

[0009] S2: Extract vegetation index and texture features from the eucalyptus canopy. From the preprocessed canopy images, vegetation index features are calculated based on reflectance at each band. These include at least one of the following: Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), Soil-Adjusted Vegetation Index (SAVI), Normalized Difference Red Edge Index (NDRE), Enhanced Vegetation Index (EVI), and Normalized Greenness Vegetation Index (GNDVI). Each vegetation index is calculated using established band calculation formulas and reflects the physiological and biochemical characteristics of eucalyptus canopy leaves, such as chlorophyll content and biochemical composition.

[0010] Texture features are extracted from canopy images based on gray-level co-occurrence matrix (GLCM). The extracted texture features include at least one of contrast, entropy, dissimilarity, variance, mean, coherence, second moment of angle, and correlation. The texture features reflect spatial structural information such as spatial distribution, texture roughness, and growth uniformity of eucalyptus canopy.

[0011] S3: Based on mutual information, the fusion features are filtered to construct a multi-feature fusion dataset. The vegetation index features and texture features extracted in step S2 are concatenated dimensionally to form an initial fusion feature set, achieving the preliminary fusion of spectral biochemical information and spatial structure information.

[0012] The mutual information algorithm based on information theory calculates the mutual information value between each feature in the initial fusion feature set and the measured nutrient content of nitrogen, phosphorus, and potassium in eucalyptus leaves. The magnitude of the mutual information value quantifies the degree of correlation between the feature and the target nutrient content; the larger the value, the stronger the explanatory power of the feature for the nutrient content.

[0013] Feature selection is performed based on mutual information values, retaining the top 70%-80% of highly correlated features and eliminating redundant features with low mutual information values. This constructs an optimized multi-feature fusion dataset, effectively eliminating multicollinearity among features and improving the computational efficiency and fitting accuracy of subsequent models.

[0014] S4: Construct an inversion model for the nutrient content of eucalyptus leaves using the random forest regression algorithm. Match the multi-feature fusion dataset with the corresponding measured nutrient content data of nitrogen, phosphorus, and potassium in eucalyptus leaves to form a modeling sample set. Randomly divide the sample set into a training set and an independent test set in a 7:3 ratio.

[0015] Using multi-feature fusion data from the training set as input and measured nutrient content of nitrogen, phosphorus, and potassium in leaves as output, a random forest regression inversion model is constructed. A grid search combined with 5-fold cross-validation is used to optimize the model parameters. The parameters to be optimized include: number of decision trees (100-500), maximum depth (10-50), minimum number of split samples (2-10), and maximum number of features (sqrt / log2). The optimal values ​​of each parameter are determined by the accuracy index of cross-validation to obtain the initial inversion model.

[0016] S5: Verify and optimize the accuracy of the inversion model to obtain the optimal inversion model. Input the independent test set into the initial inversion model, and use the coefficient of determination R², root mean square error RMSE, and normalized root mean square error nRMSE as the model accuracy evaluation indicators. The closer R² is to 1, the better the model fit. The smaller the RMSE and nRMSE, the smaller the model prediction error.

[0017] Perform residual analysis and extreme value validation on the model: plot the residual distribution between the model's predicted values ​​and the measured values, and remove outlier samples with excessively large residuals; validate extreme value samples of nutrient content separately to ensure the model's ability to predict extreme values. Fine-tune the model parameters based on the analysis results, retrain the model after removing outlier samples, obtain the optimal inversion model, and save the optimal parameters of the model.

[0018] S6: The optimal inversion model is used to invert the nutrient content of eucalyptus canopy leaves in the test area. The UAV multispectral image of the test area is preprocessed, feature extracted and filtered according to the methods in steps S1-S3 to obtain multi-feature fusion data consistent with the modeling set format.

[0019] The fused feature data of the area to be tested is input pixel by pixel into the optimal inversion model to perform nutrient content prediction calculations. Kriging interpolation or inverse distance weighted interpolation algorithms are used to perform spatial interpolation processing on the prediction results to generate a raster spatial distribution map of nitrogen, phosphorus and potassium content in the eucalyptus canopy. The spatial resolution is consistent with the UAV multispectral image, realizing a full-domain visualization of eucalyptus nutrient content in the area to be tested.

[0020] This invention also provides a eucalyptus canopy leaf nutrient content inversion system based on multi-feature fusion, comprising: Data Acquisition Module: Used to acquire multispectral image data of eucalyptus canopy in the study area via a UAV multispectral platform. It supports batch acquisition, format conversion, and storage of multi-band images including blue, green, red, red-edge, and near-infrared, and is compatible with image formats from mainstream UAV multispectral sensors. Preprocessing Module: Used to sequentially perform radiometric calibration, geometric correction, image stitching, and canopy extraction on the multispectral image data. It includes a built-in radiometric calibration coefficient library, a geometric correction polynomial algorithm, and an NDVI threshold extraction algorithm, outputting standardized eucalyptus canopy image data.

[0021] Feature extraction module: used to automatically extract preset vegetation index features and texture features from preprocessed canopy images. It has built-in calculation formulas for multiple vegetation indices and gray-level co-occurrence matrix texture feature extraction algorithms, and supports batch calculation of features and dataset generation.

[0022] Feature fusion module: Used to fuse and filter vegetation index features and texture features based on mutual information algorithm. It has a built-in mutual information calculation model and feature ranking and filtering algorithm, outputs an optimized multi-feature fusion dataset, and supports custom settings for feature filtering ratio.

[0023] The model building module calls the random forest regression algorithm, using a multi-feature fusion dataset as input and measured leaf nutrient content as output to construct an initial inversion model. It includes built-in grid search and cross-validation algorithms, supporting automatic optimization of model parameters. The model optimization module performs accuracy verification, residual analysis, and extreme value verification on the initial inversion model, using R², RMSE, and nRMSE as evaluation metrics. It supports outlier removal and fine-tuning of model parameters, outputting and saving the optimal inversion model. Ideally, when R² is close to 1 and RMSE and nRMSE are low, it indicates that the model has high prediction accuracy and reliability.

[0024]

[0025] In the formula, Xi, Yi and These represent the measured value, measured mean, estimated value, and estimated mean, respectively; n represents the number of samples in the inversion model or validation model.

[0026] Inversion Application Module: Used to load the optimal inversion model, perform pixel-by-pixel nutrient content inversion of the canopy image of the area to be tested, and has a built-in spatial interpolation algorithm to generate and visualize the spatial distribution map of nitrogen, phosphorus and potassium content in the eucalyptus canopy. It supports the export and overlay analysis of the distribution map.

[0027] In addition, the present invention also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the processor; the memory stores computer program instructions executable by the processor, which, when executed by the processor, implement the above-described method for inverting nutrient content of eucalyptus canopy leaves based on multi-feature fusion.

[0028] The present invention also provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the above-described method for inverting the nutrient content of eucalyptus canopy leaves based on multi-feature fusion.

[0029] Beneficial effects Compared with the prior art, the present invention has the following significant advantages: 1. Multi-feature fusion to fill information gaps: By fusing vegetation index features and texture features, spectral information reflecting the physiological and biochemical characteristics of leaves is organically combined with texture information reflecting the spatial structure of the canopy. This fully taps into the information value of UAV multispectral imagery, effectively fills the information gaps of single features, and significantly enhances the accuracy and stability of the model's prediction of eucalyptus leaf nutrient content.

[0030] 2. Mutual information feature selection improves model efficiency: The mutual information algorithm is used for feature selection, which quantifies the correlation between features and nutrient content. Redundant features are eliminated while retaining key information, which effectively eliminates the multicollinearity problem between features, reduces the feature dimension of the model, and improves the computational efficiency and generalization ability of the model.

[0031] 3. Random Forest Algorithm Enhances Nonlinear Fitting Ability: The random forest regression algorithm is selected to construct the inversion model. This algorithm has strong nonlinear fitting ability and anti-overfitting performance, and can fully fit the complex nonlinear relationship between spectral-texture features and leaf nutrient content. The inversion accuracy is significantly better than that of the traditional linear regression model. At the same time, the accuracy and stability of the model are further improved by using grid search + 5-fold cross-validation for parameter optimization.

[0032] 4. Achieve rapid and non-destructive full-area monitoring: Based on UAV remote sensing technology, this invention enables non-contact, large-area, and rapid monitoring of nutrient content in eucalyptus canopy leaves. Compared with traditional manual sampling and laboratory analysis methods, it significantly shortens the monitoring cycle, reduces monitoring costs, and can generate spatial distribution maps of nutrient content, achieving full-area visualization of nutrient status.

[0033] 5. Systematic design enhances practicality: An inversion system integrating data acquisition, preprocessing, feature extraction, fusion screening, model building, optimization, and inversion application has been constructed. Each module is independent yet collaborative, easy to operate, supports batch processing and visualization output, and can be directly applied to the production practice of eucalyptus plantations, providing reliable technical support and data reference for precision fertilization, nutrient management, and growth regulation. Attached Figure Description

[0034] Figure 1 This is a flowchart of the construction method of the present invention; Figure 2 This is a schematic diagram of the module structure of the inversion system of the present invention; Figure 3 A scatter plot of predicted and measured values ​​from the eucalyptus leaf nitrogen content inversion model; Figure 4 A bar chart comparing the accuracy of inversion models for different feature types; Figure 5 Spatial distribution map of nitrogen, phosphorus, and potassium nutrient content in the canopy of eucalyptus trees; Detailed Implementation The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. These embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0035] This embodiment uses the eucalyptus plantation area in Shangsi County, Fangchenggang City, Guangxi Zhuang Autonomous Region as the study area to carry out the inversion experiment of nitrogen, phosphorus and potassium content in eucalyptus canopy leaves. The eucalyptus variety in the study area is Eucalyptus urophylla, the trees are 3 years old, the planting density is 110 trees / mu, and the study area is about 500 mu.

[0036] Example 1: Nitrogen content inversion of eucalyptus leaves 1. Data Acquisition and Preprocessing: A DJI M350 RTK drone equipped with an AQ600 multispectral sensor acquired multispectral images of the study area. The sensor includes six bands: blue (450nm), green (555nm), red (660nm), red edge (720nm), near-infrared (840nm), and red edge 2 (750nm). The flight altitude was 100m, with a forward overlap of 80%, a lateral overlap of 70%, and a ground resolution of 0.1m. Preprocessing of the raw images included: radiometric calibration to convert DN values ​​to surface reflectance; geometric correction based on 15 field-deployed ground control points and a 30m resolution DEM; seamless image stitching using a seamless stitching algorithm; and extraction of the eucalyptus canopy region using an NDVI threshold of 0.3, resulting in standardized canopy images.

[0037] 2. Feature Extraction: Twenty vegetation indices and eight texture features were extracted from the canopy images. Vegetation indices included NDVI, RVI, SAVI, NDRE, EVI, and GNDVI. Texture features were extracted based on a 5×5 gray-level co-occurrence matrix, including contrast, entropy, dissimilarity, variance, mean, synergy, second moment of angle, and correlation. Pearson correlation analysis was used to initially screen vegetation indices (NDRGI, NGBDI, ExG, ExR, RGRI, MGRVI, etc.) with correlation coefficients greater than 0.7 with nitrogen content, and texture features with high correlation (mean, synergy), forming 15 candidate features.

[0038] 3. Feature Fusion and Selection: The 15 candidate features are concatenated dimensionally to construct an initial fused feature set; the mutual information value between each feature and the leaf nitrogen content is calculated, and the top 80% of the features are retained. Finally, 10 core features are retained: NGRDI, ExG, ExR, RGRI, ExGR, CVI, GLI, Intensity, Mean, and Coordination, to construct a multi-feature fusion dataset.

[0039] 4. Model Construction and Optimization: The fusion dataset was matched with the measured nitrogen content data, and divided into training and test sets in a 7:3 ratio. A random forest regression model was constructed, and the parameters were optimized using grid search and 5-fold cross-validation. The optimal parameters were determined to be: 300 decision trees, a maximum depth of 30, a minimum number of split samples of 5, and a maximum number of features of sqrt. The model was then validated and optimized for accuracy: residual analysis was used to remove 3 outlier samples. After retraining the model, the training set R²=0.84, RMSE=0.28%, and the test set R²=0.78, RMSE=0.32%, nRMSE=1.96%.

[0040] 5. Inversion Application: Ten core features were extracted pixel by pixel from the canopy image of the study area and input into the optimal model for inversion. Kriging interpolation was used to generate a spatial distribution map of nitrogen content, which clearly showed the spatial distribution pattern of eucalyptus nitrogen content in the study area. High content areas were mainly concentrated in the central part of the study area, while low content areas were distributed at the edge of the study area.

[0041] Example 2: Inversion of phosphorus and potassium content in eucalyptus leaves Using the same UAV image acquisition, preprocessing, and feature extraction methods as in Example 1, inversion experiments were conducted on phosphorus and potassium content: 1. Phosphorus content inversion: Through mutual information feature screening, the mean and NGBDI were identified as the core features for phosphorus content inversion. The constructed random forest model test set showed R²=0.77 and RMSE=0.06%, which is 10.1% more accurate and 20% lower RMSE than the single vegetation index model (R²=0.69). 2. Potassium content inversion: Through mutual information feature screening, synergy and CVI were identified as the core features for potassium content inversion. The constructed random forest model test set showed R²=0.80 and RMSE=0.15%, which is 11.3% more accurate and 30% lower RMSE than the single vegetation index model (R²=0.70).

[0042] The results show that the fusion of spectral features and texture features can effectively compensate for the indirectness of the spectral response of phosphorus and potassium elements, and significantly improve the inversion accuracy.

[0043] Example 3: Multi-nutrient synergistic inversion and system application Based on the same set of UAV multispectral imagery and multi-feature fusion dataset, optimal inversion models for the nitrogen, phosphorus, and potassium contents of eucalyptus in random forests were constructed and integrated into the inversion system of this invention, forming a eucalyptus multi-nutrient collaborative inversion system. This system was used to collaboratively invert 500 mu (approximately 33 hectares) of eucalyptus plantation in the study area, simultaneously outputting spatial distribution maps of nitrogen, phosphorus, and potassium contents, achieving a comprehensive diagnosis of the nutrient status of the eucalyptus canopy. Based on the inversion results, targeted fertilization plans were formulated for low-nutrient-content areas at the edge of the study area, with a nitrogen, phosphorus, and potassium application ratio of 3:1:2, providing direct data support for precision fertilization.

[0044] Comparative test To verify the advantages of this invention, a comparative experiment was conducted. A single vegetation index model, a single texture feature model, and the multi-feature fusion model of this invention were constructed respectively. R² and RMSE were used as evaluation indicators to compare the inversion accuracy of the three models for nitrogen, phosphorus, and potassium content. The results are shown in Table 1.

Claims

1. A method for inverting nutrient content in eucalyptus canopy leaves based on multi-feature fusion, characterized in that, Includes the following steps: S1: Acquire UAV multispectral image data and perform preprocessing, including radiometric calibration, geometric correction, image stitching and canopy extraction; S2: Extract vegetation index features and texture features of the eucalyptus canopy from the preprocessed multispectral images; S3: Based on mutual information, vegetation index features and texture features are fused and filtered to construct a multi-feature fusion dataset; S4: Using the random forest regression algorithm, with the multi-feature fusion dataset as input, construct an inversion model for the nitrogen, phosphorus, and potassium content of eucalyptus leaves; S5: Verify the accuracy and optimize the parameters of the constructed inversion model to obtain the optimal inversion model; S6: Use the optimal inversion model to invert the nutrient content of eucalyptus canopy leaves in the test area and generate a spatial distribution map of nutrient content.

2. The method according to claim 1, characterized in that, The UAV multispectral image data mentioned in step S1 is acquired by a UAV platform equipped with a multispectral sensor, which includes blue light, green light, red light, red edge, and near-infrared bands.

3. The method according to claim 1, characterized in that, The specific preprocessing operations described in step S1 are as follows: Radiometric calibration converts the original DN values ​​of the image into surface reflectance, eliminating sensor errors; the core formula for radiometric calibration is as follows: in, Radiance, This is the gain coefficient. The bias coefficient, For surface reflectance, d The distance between the Earth and the Sun. Solar spectral irradiance, This is the solar zenith angle. It can be obtained through calibration using a ground-based standard reflector. and This can be completed by combining the solar angle parameters measured simultaneously. DN Conversion from reflectance to reflectance.

4. Geometric correction, based on ground control points and the Digital Elevation Model (DEM), eliminates image distortion caused by terrain and flight. Seamless image stitching is employed to achieve overall image stitching of the study area. Canopy extraction uses the NDVI thresholding method, with a threshold set to 0.25-0.35, to separate the eucalyptus canopy from the background area. Let the original coordinates of the original image be (x, y), and the original coordinates of the corrected image be (X, Y). The quadratic polynomial correction model can be expressed as: in, and The coefficients are polynomials, obtained by solving using the least squares method with ground control points (GCPs). The ground control points must be accurately identifiable in both the image and the actual geospatial environment. For a quadratic polynomial, at least six evenly distributed control points are required, and the number should satisfy the following: 。 5. The method according to claim 1, characterized in that, The vegetation index features mentioned in step S2 include at least one of the following: Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), Soil-Adjusted Vegetation Index (SAVI), Normalized Difference Red Edge Index (NDRE), Enhanced Vegetation Index (EVI), and Greenness Normalized Difference Vegetation Index (GNDVI); the texture features include at least one of the following extracted based on the gray-level co-occurrence matrix: contrast, entropy, dissimilarity, variance, mean, synergy, second moment of angle, and correlation. The window size of the gray-level co-occurrence matrix is ​​set to 3×3 or 5×5, and the step size is 1.

6. The method according to claim 1, characterized in that, The feature fusion and filtering based on mutual information described in step S3 specifically includes: concatenating the vegetation index features and texture features dimensionally to construct an initial fusion feature set; calculating the mutual information value between each feature and the nitrogen, phosphorus, and potassium nutrient content of the leaves based on information theory to quantify the correlation between the features and the target value; retaining the top 70%-80% of the features with the highest mutual information value, eliminating redundant features with low correlation, constructing an optimized multi-feature fusion dataset, and eliminating the multicollinearity problem.

7. The method according to claim 1, characterized in that, The parameter optimization of the random forest regression algorithm described in step S4 includes: using a grid search combined with 5-fold cross-validation to optimize the number of decision trees, maximum depth, minimum number of split samples, and maximum number of features; wherein the search range for the number of decision trees is 100-500, the search range for the maximum depth is 10-50, the search range for the minimum number of split samples is 2-10, and the maximum number of features is selected using sqrt or log2.

8. The method according to claim 1, characterized in that, The accuracy verification described in step S5 includes: dividing the sample set into a training set and an independent test set in a 7:3 ratio, using the coefficient of determination R², root mean square error RMSE, and normalized root mean square error nRMSE as model evaluation metrics to verify the model's generalization ability; after parameter optimization, performing residual analysis and extreme value verification on the model, removing outlier samples and adjusting model parameters to improve model stability.

9. The method according to claim 1, characterized in that, The specific operation of nutrient content inversion in step S6 is as follows: extract and fuse features pixel by pixel from the preprocessed multispectral image of the area to be tested, input them into the optimal inversion model for calculation, and generate a raster spatial distribution map of nitrogen, phosphorus and potassium content in the eucalyptus canopy through a spatial interpolation algorithm, with the spatial resolution consistent with that of the UAV multispectral image.

10. A eucalyptus canopy leaf nutrient content inversion system based on multi-feature fusion, characterized in that, include: The data acquisition module is used to acquire multispectral image data of eucalyptus canopy in the study area through a UAV multispectral platform, supporting batch acquisition and storage of multi-band images; the preprocessing module is used to perform radiometric calibration, geometric correction, image stitching and canopy extraction on the multispectral image data in sequence, and output standardized canopy image data; the feature extraction module is used to automatically extract preset vegetation index features and texture features from the preprocessed images to generate feature datasets. The feature fusion module is used to fuse and filter vegetation index features and texture features based on the mutual information algorithm, and output an optimized multi-feature fusion dataset. The model building module is used to call the random forest regression algorithm, taking the multi-feature fusion dataset as input and the measured leaf nutrient content as output, to build the initial inversion model. The model optimization module is used to optimize parameters, verify accuracy, and analyze stability of the initial inversion model, output the optimal inversion model, and save the model parameters. The inversion application module is used to load the optimal inversion model, perform pixel-by-pixel inversion of the canopy image of the area to be tested, and generate and visualize the spatial distribution map of nutrient content.

11. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the processor; The memory stores computer program instructions that can be executed by the processor, which, when executed by the processor, implement the method for inverting nutrient content of eucalyptus canopy leaves based on multi-feature fusion as described in any one of claims 1-8.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the method for inverting nutrient content of eucalyptus canopy leaves based on multi-feature fusion as described in any one of claims 1-8.

Images