Method for distinguishing uranium and associated elements from other heavy metal stress in vegetation

Abstract

The invention belongs to the technical field of vegetation stress monitoring, and particularly relates to a method for distinguishing uranium and associated elements from other heavy metal stress in vegetation, which comprises the following steps: step 1, pollution stress experiment and vegetation sample collection; step 2, collecting and preprocessing spectral data; step 3, spectral feature extraction and stress response analysis; 4, constructing a stress distinguishing model; and step 5, pollution type determination and result output. According to the method, accurate judgment of stress types is realized through a machine learning algorithm on the basis of vegetation spectral characteristic differences of uranium and associated elements and other heavy metal stress, and the method has the advantages of accurately distinguishing pollution types, quantitatively evaluating pollution intensity, quickly and efficiently monitoring pollution, intuitively displaying pollution conditions through a spatial distribution diagram, being wide in application scene and the like.

Description

Technical Field

[0001] The invention belongs to the technical field of vegetation stress monitoring, and in particular relates to a method for distinguishing stresses of uranium and associated elements from other heavy metals in vegetation. Background Art

[0002] The mining and utilization of uranium and associated elements in uranium mining areas will cause radioactive elements such as uranium, thorium, and radium to enter the soil and water, causing stress to the surrounding vegetation. At the same time, other heavy metals (such as cadmium, lead, copper, etc.) will also enter the environment through industrial activities, agricultural fertilization, etc., which will have a significant impact on vegetation growth. Although uranium and associated elements and other heavy metals can cause stress to vegetation, their mechanisms of action are different: uranium and associated elements mainly act through radiotoxicity and chemical toxicity, causing oxidative stress, DNA damage and photosynthesis disorders in plant cells. Other heavy metals mainly interfere with plant enzyme activity, metabolic processes and water absorption through chemical toxicity, affecting photosynthesis and chlorophyll content.

[0003] Existing technologies for distinguishing uranium and associated elements from other heavy metals mainly rely on chemical analysis to detect pollutant concentrations, which is complex and time-consuming, and it is difficult to quickly and accurately distinguish the source of stress. Traditional pollution monitoring methods (such as chemical analysis of soil and water samples) have problems such as complex operation, long time consumption, high cost, and it is difficult to quickly and non-destructively determine the type and source of pollution. As a direct reflection of the physiological state of plants, vegetation spectral characteristics will show specific spectral responses when subjected to environmental stress. They are fast, non-destructive, and efficient, but existing technologies have not fully utilized vegetation spectral characteristics to distinguish uranium and associated element stress from other heavy metal stress.

[0004] To this end, the present invention proposes an innovative method to accurately distinguish the sources of stress based on the differences in the spectral characteristics of vegetation. Summary of the invention

[0005] The purpose of the present invention is to provide a method for distinguishing uranium and associated elements from other heavy metal stresses in vegetation. The method is based on the differences in the spectral characteristics of vegetation under uranium and associated elements and other heavy metal stresses, and through a machine learning algorithm, accurate determination of the stress type is achieved. The method has the advantages of accurately distinguishing pollution types, quantitatively evaluating pollution intensity, fast and efficient pollution monitoring, intuitive display of pollution conditions through spatial distribution maps, and a wide range of application scenarios.

[0006] The technical solution to achieve the purpose of the present invention is:

[0007] A method for distinguishing uranium and associated elements from other heavy metal stresses in vegetation, the method comprising:

[0008] Step 1: Pollution stress experiment and vegetation sample collection;

[0009] Step 2: Spectral data collection and preprocessing;

[0010] Step 3: Spectral feature extraction and stress response analysis;

[0011] Step 4: Construction of stress differentiation model;

[0012] Step 5: Pollution type determination and result output.

[0013] The step 1 comprises:

[0014] Step 1.1, Experimental design: Select pollutants, set stress concentration and stress time, and simulate the stress of uranium and associated element pollution and other heavy metal pollution on vegetation through laboratory control experiments or field sampling;

[0015] Step 1.2, vegetation sample selection: select plants that are sensitive to pollution and widely distributed, and collect background healthy vegetation as a control group;

[0016] Step 1.3, experimental environment control: Under laboratory conditions, use culture medium to grow vegetation, control environmental factors, and ensure that pollution stress is the only variable; in field experiments, select uranium mining areas, industrial areas and unpolluted areas for comparative sampling, and record the concentration of pollutants in soil and water.

[0017] The step 2 includes: using a portable hyperspectral instrument or a hyperspectral sensor carried by an unmanned aerial vehicle to collect reflectance spectrum data of vegetation; and preprocessing the spectrum data by denoising, normalizing, and background removal.

[0018] The spectral characteristic parameters extracted in step 3 include: absorption peak characteristics, red edge wavelength position, vegetation index, and feature dimensionality reduction; the stress response analysis includes: uranium and associated element stress and other heavy metal stress.

[0019] The step 4 comprises:

[0020] Step 4.1, data set construction: take the extracted spectral feature parameters and the corresponding pollution stress types as input and labels to construct a training data set;

[0021] Step 4.2: Model training and optimization: Use a variety of machine learning algorithms to train the model, and use optimization strategies to improve the classification performance of the model;

[0022] Step 4.3, model validation: Use multiple indicators to evaluate the classification performance of the model, and use the validation set, test set, and field data to evaluate and validate the model.

[0023] The step 4.2 comprises:

[0024] Step 4.2.1, Model selection: Select the following machine learning algorithms for model training: support vector machine, kernel function, random forest, gradient boosting decision tree;

[0025] Step 4.2.2, model training: Use the training set data to train the model, and use cross-validation to evaluate the generalization ability of the model; tune the key hyperparameters of the model through grid search or Bayesian optimization methods; during the training process, analyze the contribution of each feature to the classification results.

[0026] The classification performance indicators of the evaluation model in step 4.3 include: accuracy, precision and recall, F1 score, and confusion matrix.

[0027] The step 5 comprises:

[0028] Step 5.1, spectral data input: input the pre-processed vegetation spectral data of the target area into the trained stress differentiation model in a specific format;

[0029] Step 5.2, stress type determination: Through the trained stress differentiation model, automatically determine whether the vegetation is stressed by uranium and associated elements or other heavy metals, and further evaluate the pollution intensity;

[0030] Step 5.3, spatial distribution map generation: Combine the pollution determination results and geographic coordinates of each sampling point, and use the spatial interpolation algorithm to generate the pollution type and intensity distribution map of the target area.

[0031] The step 5.2 comprises:

[0032] Step 5.2.1, stress type classification: First determine whether the vegetation is under stress. If the vegetation is under stress, further determine the stress type: uranium and associated element stress or other heavy metal stress; the stress differentiation model outputs the classification result of each sample according to the input spectral characteristics: category label, classification probability;

[0033] Step 5.2.2, pollution intensity labeling: classify the pollution intensity according to the change amplitude of the spectral characteristics and the degree of damage to the fluorescence parameters; calculate the pollution intensity score according to the classification probability and characteristic value change output by the model; automatically label the pollution intensity level of each sampling point according to the calculated pollution intensity score.

[0034] The beneficial technical effects of the present invention are:

[0035] 1. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation provided by the present invention combines multidimensional spectral features (such as red edge position, absorption peak depth, vegetation index) with fluorescence parameters (such as Fv / Fm) to comprehensively characterize the stress effects of uranium and associated elements and other heavy metal pollution on vegetation. The high-dimensional spectral data is optimized by recursive feature elimination (RFE) and principal component analysis (PCA), which significantly improves the ability to distinguish pollution types.

[0036] 2. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation provided by the present invention utilizes hyperspectral remote sensing technology and machine learning algorithms to realize automation and rapidity of pollution monitoring, greatly reducing manpower and time costs, and is more efficient than traditional soil and water sample detection methods; and, a variety of machine learning algorithms such as support vector machine (SVM), random forest (RF) and gradient boosted decision tree (GBDT) are used to classify pollution stress types; and integrated learning strategies (such as weighted voting method) are used to further improve classification accuracy and robustness.

[0037] 3. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation provided by the present invention realizes the scientific classification of pollution intensity through quantitative analysis of the amplitude of change of spectral characteristics, and can divide pollution stress into mild, moderate and severe, providing a basis for priority sorting and formulation of measures for pollution control; and generates a distribution map of pollution stress type and intensity in the target area, which can intuitively reflect the pollution status of the target area, and provide an intuitive and scientific reference for pollution stress monitoring and control.

[0038] 4. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation provided by the present invention combines laboratory simulation experiments (controlled variable method) and field sampling (complex environmental noise) in the process of data set construction to ensure that the model has a high degree of generalization ability and applicability in actual scenarios.

[0039] 5. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation provided by the present invention can efficiently and accurately distinguish uranium and associated element pollution from other heavy metal pollution, solving the problem that the pollution type is difficult to determine in traditional methods, and providing a reliable technical means for mining area environmental monitoring and pollution source identification.

[0040] 6. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation provided by the present invention is suitable for various scenarios such as uranium mining areas, industrial pollution areas, and farmland heavy metal pollution monitoring, and has broad promotion and application value. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below in conjunction with embodiments.

[0042] The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation provided by the present invention specifically comprises the following steps:

[0043] Step 1: Pollution stress experiment and vegetation sample collection

[0044] Step 1.1 Experimental design

[0045] The present invention simulates the stress of uranium and associated element pollution and other heavy metal pollution on vegetation by means of laboratory control tests or field sampling.

[0046] Pollutant selection:

[0047] Uranium and associated element group: uranium (U), thorium (Th), radium (Ra).

[0048] Other heavy metal groups: cadmium (Cd), lead (Pb), copper (Cu).

[0049] Stress concentration setting:

[0050] The pollutant concentrations were set at three gradients: low, medium and high, for example, uranium concentrations were 5 mg / kg, 10 mg / kg and 20 mg / kg, and cadmium concentrations were 1 mg / kg, 5 mg / kg and 10 mg / kg, to simulate environments with different pollution intensities.

[0051] Stress time: The pollution treatment time was 1 week, 2 weeks, and 4 weeks, and the dynamic spectral changes of vegetation were observed.

[0052] Step 1.2: Vegetation sample selection

[0053] Select plants that are sensitive to pollution and widely distributed, such as grasses (such as ryegrass), shrubs (such as vitex negundo), etc. At the same time, collect background healthy vegetation as a control group.

[0054] Step 1.3: Experimental environment control

[0055] Under laboratory conditions, vegetation is grown using culture media, and environmental factors such as temperature, humidity, and light are controlled to ensure that pollution stress is the only variable.

[0056] In the field experiment, uranium mining areas, industrial areas and unpolluted areas were selected for comparative sampling, and the concentrations of pollutants in soil and water were recorded.

[0057] Step 2: Spectral data collection and preprocessing

[0058] Step 2.1: Spectroscopic instrument and acquisition method

[0059] Use a portable hyperspectral instrument (such as ASD FieldSpec 4) or a hyperspectral sensor mounted on a drone to collect the reflectance spectrum of vegetation.

[0060] The spectral range covers 350nm to 2500nm with a resolution better than 10nm, focusing on the following bands:

[0061] Blue light region (450nm): carotenoid absorption;

[0062] Red light region (680nm): chlorophyll absorption;

[0063] Red edge area (680-750nm): chlorophyll content and photosynthesis efficiency;

[0064] Near infrared region (750-800nm): vegetation health status.

[0065] Step 2.2: Collection process

[0066] Laboratory conditions: Place the probe of the spectrometer 10 cm above the top of the vegetation and collect the reflected spectrum vertically to ensure a uniform light source.

[0067] Field conditions: Use a drone equipped with a hyperspectral sensor to fly 30m-50m above the vegetation to collect spectral data of large-area vegetation.

[0068] Step 2.3: Spectral data preprocessing

[0069] Denoising: The Savitzky-Golay smoothing algorithm is used to eliminate random noise in the spectral curve.

[0070] Normalization: Standardize the spectral reflectance to between 0 and 1 to eliminate the impact of ambient lighting differences.

[0071] Background removal: Eliminate the spectral data of non-vegetation pixels such as soil and rocks, and only retain the vegetation spectrum.

[0072] Step 3: Spectral feature extraction and stress response analysis

[0073] Step 3.1: Extraction of spectral characteristic parameters

[0074] Step 3.1.1, absorption peak characteristics:

[0075] The depth and width of the absorption peaks in the blue light region (450nm) and the red light region (680nm) were extracted to reflect the changes in the content of carotenoids and chlorophyll.

[0076] Step 3.1.2, Red Edge Wavelength Position (REIP):

[0077] The position of the red edge wavelength (680-750nm) was calculated. Uranium and associated element stress usually leads to a blue shift of the red edge (shortening of the wavelength), while other heavy metal stresses manifest as an overall decrease in the red edge.

[0078] Step 3.1.3, vegetation index:

[0079] Normalized Difference Vegetation Index (NDVI): reflects the chlorophyll content and health of vegetation.

[0080] NDVI=(R NIR +R Red )(R NIR -R Red )

[0081] Where NDVI is the normalized difference vegetation index; R NIR is the reflectivity in the near-infrared band; R Red is the reflectivity of the red band.

[0082] Photosynthetic reflectance index (PRI): reflects the photosynthetic efficiency of the light system.

[0083] PRI=(R 531 +R 570 )(R 531 -R 570 )

[0084] Where PRI is the photosynthetic reflectance index; R 531 is the reflectivity of the 531st band; R 570 is the reflectivity of the 570th band.

[0085] Fluorescence parameters: Maximum fluorescence efficiency (Fv / Fm): It is a very important parameter in plant photosynthesis research, used to evaluate the potential photochemical efficiency of plant photosystem II (PSII). It reflects the maximum potential of plant photosynthesis under stress-free conditions and is often used as an indicator of plant health. Uranium and associated element stress usually lead to a significant decrease in Fv / Fm. Its formula is:

[0086] Fv / Fm=(F m -F0) / F m

[0087] Where Fv is variable fluorescence, which represents the photochemical effect in photosynthesis system II. The calculation formula is Fv = F m -F0.

[0088] Fm: Maximum Fluorescence, which refers to the maximum fluorescence value measured when the reaction center of photosynthesis system II is completely closed by a strong saturating light pulse in the completely dark-adapted state.

[0089] F0: Minimum Fluorescence, which refers to the initial fluorescence value measured when the reaction center of photosynthesis system II (PSII) is fully opened in the completely dark-adapted state. At this time, no photochemical reaction occurs, only basic fluorescence emission.

[0090] Step 3.1.4: Feature Dimensionality Reduction

[0091] Principal component analysis (PCA) and recursive feature elimination (RFE) methods are used to screen out the most representative features from high-dimensional spectral data, reduce redundant information and improve computational efficiency.

[0092] Step 3.2: Analysis of stress response patterns

[0093] Step 3.2.1: Uranium and associated element stress

[0094] The blue light absorption peak (450nm) decreased significantly;

[0095] The red edge position has a significant blue shift;

[0096] The photosynthetic reflectance index (PRI) and fluorescence parameters (Fv / Fm) were significantly reduced.

[0097] Step 3.2.2. Other heavy metal stress

[0098] The red light absorption peak (680nm) decreased significantly;

[0099] The NDVI decreased significantly, indicating that the chlorophyll content was severely damaged;

[0100] The blue light absorption changes less.

[0101] Step 4: Construction of stress differentiation model

[0102] The construction of the data set is the basis of the stress differentiation model. The present invention obtains a comprehensive and accurate spectral data set by combining laboratory simulation experiments and field sampling. At the same time, steps 2.3 and 3.1 preprocess and extract features of the data, providing high-quality data for subsequent model training.

[0103] Step 4.1: Dataset construction

[0104] The extracted spectral characteristic parameters and the corresponding pollution stress types (uranium and associated element stress or other heavy metal stress) are used as input and labels to construct a training data set.

[0105] The dataset includes laboratory simulation data and field sampling data to ensure the generalization ability of the model.

[0106] The processed data set is divided into 70% training set, 15% validation set, and 15% test set to ensure the independence of model training, optimization, and verification.

[0107] Step 4.2: Model training and optimization

[0108] After the data set is constructed, the present invention uses a variety of machine learning algorithms to train the model and improves the classification performance of the model through optimization strategies.

[0109] Step 4.2.1. Model selection

[0110] The present invention comprehensively considers the high dimensionality and nonlinear characteristics of spectral data and selects the following machine learning algorithms for model training:

[0111] Support Vector Machine (SVM)

[0112] SVM has a strong ability to distinguish the boundaries of different pollution stresses in spectral data.

[0113] The kernel function (such as RBF kernel) is used to capture the nonlinear relationship of spectral features, which is suitable for the classification of small sample high-dimensional data.

[0114] Random Forest (RF)

[0115] Through the integration of multiple decision trees, it has strong noise resistance and feature selection capabilities and is suitable for processing complex spectral data.

[0116] Gradient Boosted Decision Tree (GBDT)

[0117] By gradually optimizing the loss function, the complex interactive relationship between spectral features can be captured, with higher classification accuracy.

[0118] Step 4.2.2: Model training

[0119] (1) Training process

[0120] The model is trained using the training set data with the goal of minimizing the classification error.

[0121] Cross-validation (such as 5-fold cross-validation) is used to evaluate the generalization ability of the model and prevent overfitting.

[0122] (2) Hyperparameter Optimization

[0123] Use grid search or Bayesian optimization to tune the key hyperparameters of the model, for example:

[0124] The kernel function parameters of SVM (such as C and γ).

[0125] The number and maximum depth of decision trees for the random forest.

[0126] Learning rate and number of iterations for GBDT.

[0127] By optimizing hyperparameters, the classification accuracy and stability of the model can be improved.

[0128] (3) Feature Importance Analysis

[0129] During the training process, the contribution of each feature to the classification results was analyzed, and it was found that the red edge position and Fv / Fm made the greatest contribution to the identification of uranium and associated element stress, while NDVI and red light absorption peak were more sensitive to other heavy metal stresses.

[0130] Step 4.3: Model Validation

[0131] Step 4.3.1. Verification indicators

[0132] The following indicators are used to evaluate the classification performance of the model:

[0133] (1) Accuracy

[0134] Reflects the overall accuracy of the model's prediction results.

[0135] Accuracy = number of correctly classified samples / total number of samples

[0136] (2) Precision and Recall

[0137] Precision: Indicates the model’s prediction accuracy for a certain type of pollution.

[0138] Recall rate: indicates the recognition coverage of a certain type of pollution by the model.

[0139] (3) F1-Score:

[0140] Comprehensively consider the balance between precision and recall.

[0141] F1=2 Precision Recall / Precision+Recall

[0142] (4) Confusion Matrix

[0143] It is used to analyze the classification effect of the model on different pollution types and identify easily confused categories.

[0144] Step 4.3.2, Verification method

[0145] (1) Validation set verification

[0146] The validation set data is used to perform preliminary validation of the model and evaluate the performance of the model after hyperparameter optimization.

[0147] (2) Test set verification

[0148] The model is finally validated using independent test set data to evaluate the actual classification ability of the model.

[0149] (3) Field data verification

[0150] The model is verified using spectral data collected in the field to evaluate its applicability in actual application scenarios.

[0151] Step 4.3.3: Result analysis and improvement

[0152] (1) Classification results analysis

[0153] The classification effect of the model on uranium and associated elements and other heavy metal stress was analyzed through confusion matrix and F1 score.

[0154] It was found that some categories (such as uranium and cadmium) might be confused, and the problem was solved by further optimizing feature selection and model parameters.

[0155] (2) Model improvement

[0156] According to the deficiencies found in the validation, adjust the feature selection method or try other machine learning algorithms (such as XGBoost).

[0157] Model or enhance the noise interference of field data to improve the robustness of the model.

[0158] Step 5: Pollution type determination and result output

[0159] Step 5.1. Spectral data input

[0160] The pre-processed vegetation spectral data of the target area are input into the trained stress differentiation model in a specific format (such as matrix form).

[0161] The format of data input must be consistent with the data structure during model training, for example:

[0162] Rows represent samples (each sample corresponds to the spectral data of a sampling point).

[0163] The columns represent features (such as spectral feature parameters and fluorescence parameters).

[0164] Step 5.2: Determine the type of coercion

[0165] Stress type determination is the core link of pollution type determination and result output. The present invention automatically determines whether vegetation is under uranium and associated element stress or other heavy metal stress through the trained stress differentiation model, and further evaluates the pollution intensity.

[0166] Step 5.2.1. Classification of stress types

[0167] (1) Logic of coercion classification

[0168] Secondary classification system:

[0169] First, determine whether the vegetation is stressed (stressed / no stress).

[0170] If the vegetation is stressed, further determine the type of stress:

[0171] Uranium and associated element stress (such as uranium, thorium, radium, etc.).

[0172] Other heavy metal stress (such as cadmium, lead, copper, etc.).

[0173] (2) Model output

[0174] The stress differentiation model outputs the classification result of each sample based on the input spectral features and the classification rules learned during the training process:

[0175] Category label: type of stress (such as "uranium stress" or "cadmium stress").

[0176] Classification probability: The model's confidence in each stress type (e.g. "Uranium stress: 90%").

[0177] (3) Multi-algorithm integration results

[0178] If a multi-model integration strategy is adopted (such as the weighted voting method of random forest, SVM and GBDT), the output is a comprehensive judgment of the classification results of multiple models, which further improves the classification accuracy.

[0179] Step 5.2.2: Pollution intensity labeling

[0180] (1) Pollution intensity classification

[0181] The present invention classifies the pollution intensity according to the variation range of the spectral characteristics and the damage degree of the fluorescence parameters:

[0182] Mild pollution: The spectral characteristics change slightly (such as a 5%-10% decrease in NDVI and a Fv / Fm decrease of no more than 10%).

[0183] Moderate pollution: The spectral characteristics change significantly (such as NDVI decreases by 10%-30%, Fv / Fm decreases by 10%-30%).

[0184] Severe pollution: The spectral characteristics change greatly (such as NDVI decreases by more than 30%, Fv / Fm decreases by more than 30%).

[0185] (2) Pollution intensity calculation method

[0186] The pollution intensity score (Score) is calculated based on the classification probability and feature value changes output by the model:

[0187] Score=w1·ΔNDVI+w2·ΔFv / Fm+w3·ΔREIP

[0188] Where w1, w2, and w3 are feature weights determined in the model training phase; ΔNDVI, ΔFv / Fm, and ΔREIP are the magnitude of changes in feature values ​​relative to uncontaminated samples.

[0189] (3) Pollution intensity output

[0190] Based on the calculated pollution intensity score, the pollution intensity level (mild, moderate, severe) of each sampling point is automatically labeled.

[0191] Step 5.3: Generate spatial distribution map

[0192] Step 5.3.1: Spatial interpolation and visualization

[0193] Combining the pollution determination results and geographic coordinates of each sampling point, a spatial interpolation algorithm (such as Kriging interpolation) is used to generate a pollution type and intensity distribution map of the target area.

[0194] Different colors are used on the map to represent different types of pollution (such as red for uranium and associated element stress, blue for other heavy metal stress), and the depth of color is used to represent the intensity of pollution.

[0195] Step 5.3.2, Dynamic Update

[0196] If new spectral data is collected or the pollution conditions change, the system can update the pollution distribution map in real time to reflect the latest environmental conditions.

[0197] The present invention is described in detail above in conjunction with the embodiments, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of ordinary technicians in the field without departing from the purpose of the present invention. The contents not described in detail in the present invention can adopt the existing technology.

Claims

1. A method for distinguishing uranium and associated elements from other heavy metal stress in vegetation, characterized in that: The method comprises: Step 1: Pollution stress experiment and vegetation sample collection; Step 2: Spectral data collection and preprocessing; Step 3: Spectral feature extraction and stress response analysis; Step 4: Construction of stress differentiation model; Step 5: Determine the pollution type and output the results.

2. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation according to claim 1, characterized in that: The step 1 comprises: Step 1.1, Experimental design: Select pollutants, set stress concentration and stress time, and simulate the stress of uranium and associated element pollution and other heavy metal pollution on vegetation through laboratory control experiments or field sampling; Step 1.2, vegetation sample selection: select plants that are sensitive to pollution and widely distributed, and collect background healthy vegetation as a control group; Step 1.3, experimental environment control: Under laboratory conditions, use culture medium to grow vegetation, control environmental factors, and ensure that pollution stress is the only variable; in field experiments, select uranium mining areas, industrial areas and unpolluted areas for comparative sampling, and record the concentration of pollutants in soil and water.

3. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation according to claim 1, characterized in that: The step 2 includes: using a portable hyperspectral instrument or a hyperspectral sensor carried by an unmanned aerial vehicle to collect reflectance spectrum data of vegetation; and preprocessing the spectrum data by denoising, normalizing, and background removal.

4. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation according to claim 1, characterized in that: The spectral characteristic parameters extracted in step 3 include: absorption peak characteristics, red edge wavelength position, vegetation index, and feature dimensionality reduction; the stress response analysis includes: uranium and associated element stress and other heavy metal stress.

5. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation according to claim 1, characterized in that: The step 4 comprises: Step 4.1, data set construction: take the extracted spectral feature parameters and the corresponding pollution stress types as input and labels to construct a training data set; Step 4.2: Model training and optimization: Use a variety of machine learning algorithms to train the model, and use optimization strategies to improve the classification performance of the model; Step 4.3, model validation: Use multiple indicators to evaluate the classification performance of the model, and use the validation set, test set, and field data to evaluate and validate the model.

6. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation according to claim 5, characterized in that: The step 4.2 comprises: Step 4.2.1, Model selection: Select the following machine learning algorithms for model training: support vector machine, kernel function, random forest, gradient boosting decision tree; Step 4.2.2, model training: Use the training set data to train the model, and use cross-validation to evaluate the generalization ability of the model; tune the key hyperparameters of the model through grid search or Bayesian optimization methods; during the training process, analyze the contribution of each feature to the classification results.

7. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation according to claim 5, characterized in that: The classification performance indicators of the evaluation model in step 4.3 include: accuracy, precision and recall, F1 score, and confusion matrix.

8. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation according to claim 1, characterized in that: The step 5 comprises: Step 5.1, spectral data input: input the pre-processed vegetation spectral data of the target area into the trained stress differentiation model in a specific format; Step 5.2, stress type determination: Through the trained stress differentiation model, automatically determine whether the vegetation is stressed by uranium and associated elements or other heavy metals, and further evaluate the pollution intensity; Step 5.3, spatial distribution map generation: Combine the pollution determination results and geographic coordinates of each sampling point, and use the spatial interpolation algorithm to generate the pollution type and intensity distribution map of the target area.

9. The method for distinguishing uranium and associated elements from other heavy metal stress in vegetation according to claim 8, characterized in that: The step 5.2 comprises: Step 5.2.1, stress type classification: First determine whether the vegetation is under stress. If the vegetation is under stress, further determine the stress type: uranium and associated element stress or other heavy metal stress; the stress differentiation model outputs the classification result of each sample according to the input spectral characteristics: category label, classification probability; Step 5.2.2, pollution intensity labeling: classify the pollution intensity according to the change amplitude of the spectral characteristics and the degree of damage to the fluorescence parameters; calculate the pollution intensity score according to the classification probability and characteristic value change output by the model; automatically label the pollution intensity level of each sampling point according to the calculated pollution intensity score.