Tea tree leaf mineral element deficiency diagnosis method based on hyperspectral imaging and DRIS diagnosis

Abstract

The invention discloses a tea tree leaf mineral element deficiency diagnosis method based on hyperspectral imaging and DRIS diagnosis. The method comprises the following steps: S1, collecting a sample; s2, acquiring and preprocessing hyperspectral data; s3, screening characteristic wave bands; s4, measuring the contents of various mineral elements in the sample in the S1, and establishing a basic database; s5, based on the data in S3 and S4, utilizing a machine learning algorithm to respectively construct inversion models of the multiple mineral elements, performing performance evaluation on the inversion models, and screening to obtain an optimal inversion model of the multiple mineral elements; s6, hyperspectral data of tea tree leaves to be detected are input into the optimal inversion model obtained in the S5, the corresponding mineral element content is predicted, and in combination with the DRIS diagnosis standard, diagnosis on whether the mineral elements of the tea tree leaves are in a lack state or not is completed. The invention provides an innovative method which organically combines hyperspectral nondestructive testing, machine learning intelligent modeling and DRIS comprehensive diagnosis standards, and realizes rapid, nondestructive, accurate and comprehensive diagnosis of the mineral nutrition status of the tea trees.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, and in particular to a method for diagnosing mineral elements in tea leaves based on hyperspectral imaging and DRIS diagnostics. Background Technology

[0002] Tea (Camellia sinensis (L.) O. Ktze) is an important economic crop in my country. In northern tea-producing areas, such as Shandong Province, tea garden soils often suffer from malnutrition symptoms such as yellowing leaves due to factors such as high pH, ​​poor soil compaction, and improper fertilization, which severely restricts the yield and quality of tea.

[0003] Currently, the Diagnostic Fertilizer Integration Method (DRIS) is one of the most widely used methods for diagnosing mineral elements in leaves. This method analyzes the content and ratios of various mineral elements in leaves to comprehensively evaluate the balance between elements, thereby identifying key elements limiting plant growth and their nutrient requirements. However, traditional DRIS diagnosis relies entirely on laboratory chemical analysis (such as ICP-OES), which involves cumbersome procedures, including destructive processes such as blanching, drying, grinding, and digestion. This results in long testing cycles and high costs, failing to meet the demands of modern precision agriculture for large-scale, rapid, and non-destructive diagnostics.

[0004] Hyperspectral imaging, as an emerging non-destructive testing technique, can capture the fine spectral features of plant leaves in the visible to near-infrared band (e.g., 391-1001 nm). These features are closely related to the biochemical components inside the leaves (such as chlorophyll, water, and mineral elements). Machine learning algorithms (such as random forests and support vector machines) can extract hidden feature patterns from hyperspectral data and establish quantitative or qualitative relationship models between spectral information and target components.

[0005] Although research combining hyperspectral imaging and machine learning has been reported, existing technologies, such as Chinese patent CN201010208851.7 (rapid detection method and device for tea tree nutrient information based on hyperspectral imaging technology), can only predict the content of three elements: N, P, and K. Furthermore, it does not incorporate the DRIS diagnostic system and cannot assess the balance between elements. Existing technologies also have the following shortcomings: 1. It is mostly limited to predicting the content of a single or a few elements, and has failed to be systematically integrated with the mature DRIS diagnostic system, thus failing to achieve a comprehensive diagnosis from element content to nutritional status.

[0006] 2. The lack of specialized models for accurately identifying and judging key limiting elements that lead to specific physiological phenomena (such as yellowing) limits their practicality.

[0007] 3. The model lacks universality, and the data processing procedures and diagnostic standards are not standardized, making it difficult to directly apply to tea garden production practices.

[0008] Therefore, there is an urgent need in this field for an innovative method that organically combines hyperspectral nondestructive testing, machine learning intelligent modeling, and the DRIS comprehensive diagnostic standard to achieve rapid, nondestructive, accurate, and comprehensive diagnosis of the mineral nutritional status of tea trees. Summary of the Invention

[0009] This invention aims to solve the above problems and provides a method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnosis, comprising the following steps: S1: Obtain tea leaf samples with different mineral element contents; S2: Perform hyperspectral data acquisition and preprocessing on the sample from S1 respectively; S3: Feature band selection is performed on the hyperspectral data preprocessed in S2 using a feature selection algorithm; S4: Determine the content of various mineral elements in the sample from S1 and establish a basic database; S5: Based on the data from S3 and S4, inversion models for various mineral elements are constructed using machine learning algorithms, and the performance of the inversion models is evaluated to select the optimal inversion model for various mineral elements. S6: Input the hyperspectral data of the tea leaves to be tested into the optimal inversion model obtained in S5 to predict the content of corresponding mineral elements, and combine it with the DRIS diagnostic criteria to complete the diagnosis of whether the tea leaves are deficient in mineral elements.

[0010] Furthermore, in step S1, the first mature leaf below the new shoot of the tea tree is collected as a sample; this application selected five representative two-year-old clonal tea gardens in Shandong Province as sampling points. Sample classification: Based on the field symptoms, the leaves were divided into a normal group (bright green color, no visible lesions) and a yellowing group (obvious yellowing or brown spots). Sample composition: Every five leaves constituted one biological replicate sample, for a total of 151 valid samples collected (71 in the normal group and 80 in the yellowing group).

[0011] Furthermore, in step S2, hyperspectral data acquisition and preprocessing involves using a hyperspectral imaging system to acquire hyperspectral images of the leaf sample from S1, followed by correction and preprocessing to extract the average reflectance spectrum of the region of interest on the leaf.

[0012] Specifically, the hyperspectral imaging system includes: a hyperspectral camera with 1101×960 (spatial × spectral) pixels, used to acquire hyperspectral images in the 391-1001nm band; a light source: a halogen lamp line light source, used to provide stable and uniform illumination and avoid light interference during spectral acquisition; and calibration software: using Spec View software (Dualix Spectral Imaging, China) for lens calibration and reflectivity calibration, and using ENVI 5.3 software to extract the average reflectance spectrum of the region of interest (ROI) of the leaf.

[0013] Furthermore, the preprocessing includes at least one of multivariate scattering correction (MSC), Savitzky-Golay smoothing (SG), and first derivative (1D) processing to eliminate scattering effects, noise, and baseline drift.

[0014] Furthermore, the feature selection algorithm in step S3 includes one or more of the following: Uninformative Variable Elimination (UVE), Continuous Projection (SPA), and Competitive Adaptive Reweighted Sampling (CARS); to select the subset of feature bands most relevant to the mineral element content, so as to eliminate redundant information and improve model efficiency.

[0015] Furthermore, in step S4, after blanching, drying, and pulverizing, the mineral element content is determined by the Kjeldahl method for N and by inductively coupled plasma atomic emission spectrometry for P, K, Ca, Mg, Fe, Cu, Zn, Mn, and Al.

[0016] Furthermore, in step S5, the machine learning algorithm includes one or more of Partial Least Squares Regression (PLS), Support Vector Regression (SVM), and Random Forest (RF). In step S5, independent test sets are used to evaluate the performance of various element content inversion models, and the evaluation metrics include the test set determination coefficient (R²), root mean square error (RMSEP), and relative analysis error (RPD).

[0017] Model construction: Using the selected characteristic band data and measured element content data, a machine learning algorithm is used to construct a model: a content inversion model (regression model) for multiple mineral elements.

[0018] After the model is built, its performance needs to be evaluated using an independent test set (samples are randomly divided from the total sample at a 3:1 ratio, with no overlap with the training set). The coefficient of determination (R²) on the test set reflects the goodness of fit between the model's predicted values ​​and the measured values ​​(the closer R² is to 1, the better the fit), and the root mean square error (RMSEP) reflects the prediction error (the smaller the RMSEP, the higher the accuracy). The model with the highest R² value on the test set is selected as the 'optimal model' to ensure optimal prediction accuracy.

[0019] The optimal inversion model refers to: in model evaluation, it simultaneously satisfies that the coefficient of determination (R²) of the test set ≥ 0.7, the root mean square error of prediction (RMSEP) ≤ 1000 mg / kg, the relative analytical error (RPD) ≥ 2.0, and the model with the highest R² value of the test set.

[0020] Further, in step S6, the DRIS diagnostic criterion is the DRIS index. The DRIS index is the arithmetic mean of the partial functions f(X / A) corresponding to the ratios of multiple mineral elements in the sample to be evaluated, and is used to comprehensively evaluate the balance state between elements. The calculation formula of the DRIS index includes the partial function of the element ratio. The calculation method of the partial function of the element ratio f(X / A) is as follows: When X / A > x / a, f(X / A) = [ (X / A) / (x / a) - 1 ] × 1000 / CV; When X / A < x / a, f(X / A) = [ 1 - (x / a) / (X / A) ] × 1000 / CV; Where, X / A is the nutrient concentration ratio of the sample to be evaluated, x / a is the average value of the nutrient ratios of the leaves in the normal group, and CV is the coefficient of variation of the nutrient ratios of the leaves in the normal group; Based on the mineral element content data of the leaves, calculate the element ratios and the DRIS index, construct a diagnosis criterion for element abundance and deficiency including five levels of "excess, high, balanced, low, deficiency", and determine the key limiting elements causing leaf chlorosis.

[0021] Further, model evaluation and diagnostic application: Use an independent test set to evaluate the model performance. In practical applications, collect the hyperspectral data of the leaves to be measured. After preprocessing and feature band extraction, input it into the optimal content inversion model to predict the element content. Finally, combine the constructed DRIS diagnostic criterion to determine the abundance and deficiency status of each element, and output the diagnostic results and management suggestions.

[0022] The present invention has the following beneficial effects: Compared with the prior art, the method of the present invention has the following beneficial effects: Non-destructive and efficient: Using hyperspectral imaging technology, there is no need to damage the leaf tissue, and single-sample detection can be completed within a few minutes. The diagnostic accuracy rate of the Mn element (example data) is above 85%, and it supports synchronous diagnosis of 10 elements. Compared with the traditional chemical analysis which takes several days, the efficiency is increased by dozens of times, and it is suitable for large-area tea garden census.

[0023] Accuracy and reliability: Through combined spectral preprocessing and multi-algorithm feature band screening, core spectral information related to element content is effectively extracted; combined with advanced machine learning algorithms, the established model has high accuracy in retrieving the content of key elements (such as Mn) and the diagnostic results are reliable.

[0024] Comprehensiveness and practicality: This invention enables simultaneous diagnosis of 10 key mineral elements and integrates the DRIS concept. It can not only reflect the absolute content of a single element, but also assess the balance relationship between elements, accurately locate key limiting factors, and provide more targeted and systematic fertilization recommendations.

[0025] Universality and scalability: The methodology of this invention is complete and the process is standardized. Although the embodiments use young tea trees in Shandong Province as examples, by adjusting the DRIS diagnostic criteria and model parameters, this method can be adapted to tea trees of different varieties, different growth stages, and different regions, and has good prospects for promotion and application. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only one embodiment of the present invention. For those skilled in the art, other embodiments can be derived from the provided drawings without creative effort.

[0027] Figure 1 : A flowchart of the diagnostic method of the present invention; Figure 2 Comparison diagram of the normal group and the etiolated group in this invention; Figure 3 This invention relates to the DRIS index of 10 elements; Figure 4 This invention provides a DRIS grading diagnostic standard for 10 elements. Figure 5 The original spectral curve of this invention; Figure 6 : Spectral preprocessing curve of this invention; Figure 7 The invention provides a heatmap of R² values ​​for predicting the content of ten elements in tea trees using different models. Figure 8 : Comparison chart of measured and predicted values ​​of the optimal model for 10 elements in this invention. Detailed Implementation

[0028] 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 a part of the embodiments of the present invention, not all of them. The following embodiments are only for illustrative purposes and are not intended to limit the scope of the present invention in any way. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Unless otherwise specified, the experimental methods used in the present invention are conventional methods. Unless otherwise specified, the materials and reagents used in the present invention are commercially available. Furthermore, other terms used in the present invention, unless otherwise specified, generally have the meanings commonly understood by those skilled in the art.

[0029] Example 1: Diagnosis of Mineral Element Deficiency in Tea Trees in Shandong Province 1. Sample collection and baseline data determination To verify the method of this invention, five representative two-year-old asexual tea gardens in Shandong Province were selected as sampling points. These five tea gardens were: Jimo Ruicaoyuan Tea Garden, Qingdao Tea Research Institute, Zhucheng Ganquanling Tea Garden, Zhucheng Caijiagou Tea Garden (Caijiagou Village), and Zhucheng Dongsheng Tea Garden (Dongsheng Village). In November 2023 and March 2024, the first mature leaf below the new shoot of each tea tree was collected from each garden.

[0030] Sample classification: Based on the visual symptoms in the field, the leaves were divided into a normal group (dark green color, no visible lesions) and a yellowing group (distinct yellowing or brown spots). For example... Figure 2 As shown, Figure 2 The upper and middle parts show normal tea plant plants and leaves, while the lower part shows yellowed tea plant plants and leaves.

[0031] Sample composition: 5 leaves constituted 1 biological replicate sample, and a total of 151 valid samples were collected (71 from the normal group and 80 from the etiolated group).

[0032] Chemical Analysis: After washing and drying with deionized water, the samples were first used for hyperspectral data acquisition, followed by destructive chemical analysis. Mineral element content was determined by Kjeldahl nitrogen determination after blanching, drying, and pulverizing, and by inductively coupled plasma atomic emission spectrometry (ICP-AES) for P, K, Ca, Mg, Fe, Cu, Zn, Mn, and Al.

[0033] 2. Construction of DRIS diagnostic criteria and identification of key limiting elements Data processing: Calculate the mean, standard deviation, and coefficient of variation of all possible ratios of the 10 elements for the 151 samples, categorized by group.

[0034] DRIS index calculation: In this invention, we combined 10 target elements in pairs to calculate the DRIS index. Specifically, we considered all possible element pairs, that is, the combinations of arbitrarily selecting two different elements from 10 elements, and a total of C(10, 2) = 45 combinations were obtained. The DRIS index of each element was calculated by applying the element ratio deviation function formula and the DRIS index formula. The calculation formula is shown as follows.

[0035] When X / A > x / a, f(X / A) = [ (X / A) / (x / a) - 1 ] × 1000 / CV; When X / A < x / a, f(X / A) = [ 1 - (x / a) / (X / A) ] × 1000 / CV; Where, X / A is the nutrient concentration ratio of the sample to be evaluated, x / a is the average value of the nutrient ratio of the leaves in the normal group, and CV is the coefficient of variation of the nutrient ratio of the leaves in the normal group.

[0036] DRISindex = [f (X / A) + f(X / B) + …… - f(H / X)] / (n - 1) When the observed element is X in X / A, f(X / A) takes a positive value. If the observed element is A in X / A, f(X / A) takes a negative value. n represents the number of elements.

[0037] The balance status of the corresponding nutrient element is reflected by the Nutrient imbalance index (NII). When NII is 0, the nutrient element content is in a balanced state. Positive and negative values of NII indicate relative surplus or relative deficiency of the nutrient element content respectively. The larger the absolute value of NII, the more unbalanced the nutrient element content is.

[0038] NII =

[0039] The results are as Figure 3 shown, and the analysis Figure 3The data shows that the DRIS indices for Mn (-102.78), K (-8.27), N (-7.49), P (-6.93), and Al (-5.26) are negative, indicating a relative deficiency of these elements; while the DRIS indices for Ca (1.48), Mg (17.18), Fe (45.86), Cu (44.60), and Zn (21.10) are positive, indicating a relatively abundant supply of these elements. The Nutrition Imbalance Index (NII) is 26.1 > 0, reflecting a certain degree of imbalance in overall nutrition. In summary, the order of nutrient requirements for tea leaves is Mn > K > N > P > Al > Ca > Mg > Zn > Cu > Fe. Mn has the lowest DRIS index (-102.78) and has been identified as the key limiting element for yellowing tea leaves in the above-mentioned regions.

[0040] DRIS diagnostic criteria were established by using the average elemental content in the normal group as the "balance value". "High / Low" and "Excess / Deficiency" thresholds were defined based on "balance value ± (4 / 3) × standard deviation" and "balance value ± (8 / 3) × standard deviation", respectively, thus establishing DRIS diagnostic criteria for 10 elements. Figure 4 ).

[0041] 3. Hyperspectral Data Acquisition and Preprocessing System configuration: The GaiaField-Pro-V10 hyperspectral imaging system (including a 1101×960 (spatial×spectral) pixel camera + halogen lamp line light source) is used to acquire hyperspectral images of leaves in the 391-1001nm band in a dark chamber environment; the dark chamber temperature is controlled at 25±2℃ and the relative humidity is 50±5% to avoid environmental factors from interfering with spectral stability.

[0042] Image Correction: To address issues such as lens distortion, dark current noise, and uneven light source, this invention employs a reflectance correction method based on a reference standard. Its core principle is to eliminate system noise and light source influences by acquiring images of a known reference object, thereby converting the dimensionless DN value of the original image into a standard, reproducible reflectance or relative reflectance. This correction process is based on the following physical model: Original image signal (S) = target true reflection signal + dark current noise + ambient stray light.

[0043] The whiteboard image signal (W) = total internal reflection signal of an ideal diffuse reflector (reflectivity ≈ 1 or a known constant) + dark current noise + ambient stray light, which characterizes the illumination response of the system at the current location.

[0044] The dark field image signal (D) = pure dark current noise + ambient stray light, which characterizes the system's background noise.

[0045] The formula reflectance = (S - D) / (W - D) is used, where (S - D) represents the true optical response of the target object obtained by subtracting the system background noise from the original image, and (W - D) represents the "pure" signal that should be produced when a light source illuminates an object with 100% reflectance under ideal conditions, i.e., the maximum value of the system response.

[0046] Ratio calculation: Normalize the target's true response to the system's maximum response, and finally obtain a relative value that is independent of system parameters such as light source intensity and sensor gain, namely reflectivity.

[0047] First, lens calibration (to eliminate lens distortion) was performed using Spec View software. Then, white (W) and dark (D) images were acquired, and the reflectance of the original image (S) was calibrated according to the formula. After calibration, the region of interest (ROI) of the leaf was manually delineated using ENVI 5.3 software (avoiding petioles and lesion areas), and the average reflectance spectrum was extracted.

[0048] Spectral preprocessing: Multiplicative Scatter Correction (MSC), Savitzky-Golay filter (SG), and First Derivative (1-D) were applied sequentially. MSC aims to eliminate spectral baseline drift and amplification effects caused by uneven particle distribution, surface scattering, and changes in optical path.

[0049] After preprocessing, the smoothness of the spectral curve was significantly improved, high-frequency noise components were effectively filtered out, and the curve trend became clearer and more stable. Within the same wavelength range (673-811nm), the curve fluctuation amplitude was significantly reduced, and the signal-to-noise ratio was significantly improved. Figure 6 As shown, the preprocessed spectrum exhibits clearly distinguishable characteristic absorption peaks at 535 nm, 673 nm, and 811 nm. Figure 5 As shown, these characteristic peaks are submerged in noise in the original spectrum and are difficult to identify directly.

[0050] 4. Feature band selection To reduce data dimensionality, three algorithms were employed to select characteristic bands from the preprocessed spectrum: Uninformative Variables Elimination (UVE), Successive Projections Algorithm (SPA), and Competitive Adaptive Reweighted Sampling (CARS). UVE algorithm: A total of 597 information-rich bands were selected, with a wide distribution range.

[0051] SPA algorithm: A total of 139 feature bands with low redundancy were selected, concentrated in the chlorophyll absorption region and the near-infrared region.

[0052] CARS algorithm: A total of 189 bands with strong correlation to element content were selected, mainly concentrated in the near-infrared region.

[0053] Table 1 shows the selection results for the SPA, CARS, and UVE algorithms.

[0054]

[0055] 5. Model Building The samples were randomly divided into training and test sets in a 3:1 ratio.

[0056] Mineral element content inversion model: Based on the training set, three algorithms—Partial Least Squares (PLS), Support Vector Regression (SVR), and Random Forest (RF)—were employed, combined with three feature band subsets (UVE, SPA, and CARS), to construct a total of 90 regression models for the 10 elements. Figure 7 As shown, Figure 7 The model highlighted in the red box is the one with the highest R², which is the optimal model.

[0057] 6. Model Evaluation and Diagnostic Applications Content inversion model: Figure 8 The test set R for different models is shown. 2 Numerical results. The model performs well in predicting K, Ca, Fe, Cu, Mn, and N, but poorly in predicting P, Al, Zn, and Mg. Among the predictions for these 10 elements, the model with the highest prediction accuracy is Mn-UVE-RF(R). 2 =0.9283), Ca-UVE-RF(R 2 =0.9071), Cu-SPA-SVR(R 2 =0.8336), Fe-CARS-SVR(R 2 =0.7903), K-SPA-SVR(R 2 =0.7893), N-CARS-RF(R 2 =0.769), Zn-SPA-RF(R 2 =0.7431), Mg-UVE-RF(R 2=0.7169), P-SPA-SVR(R 2 =0.5967), Al-UVE-SVR(R 2 =0.5346). Except for P (R²=0.5967, RPD=1.8) and Al (R²=0.65, RPD=1.7), the optimal model for the other eight elements has R²≥0.7 and RPD≥2.0. An R² greater than 0.7 indicates that the model can explain most of the data variation, falling into the "good" to "excellent" category. An RPD greater than 2 indicates high prediction accuracy, suitable for practical applications.

[0058] According to the DRIS grading standard, the predicted content levels of 10 elements in leaves by the inversion model were compared with the measured content levels of the 10 elements. Table 2 shows that the accuracy of the test set results ranged from 60.53% to 84.21%, with the highest accuracy for Mn. The accuracy of the training set results ranged from 82.30% to 90.27%.

[0059] Table 2: Training and Test Set Data

[0060] Based on the nutrient requirement order (Mn>K>N>P>Al>Ca>Mg>Zn>Cu>Fe) determined in Example 1 and the DRIS diagnostic criteria, this invention provides the following precise fertilization scheme to correct the nutrient imbalance in tea trees: For manganese (Mn), a key limiting element, foliar spraying with easily absorbed manganese fertilizers such as manganese sulfate (MnSO4) is recommended to quickly alleviate chlorosis symptoms. For the relative deficiency of macroelements such as nitrogen (N), phosphorus (P), and potassium (K), nitrogen, phosphorus, and potassium compound fertilizers should be applied to the roots to synergistically improve the overall nutritional level of the tree, enhance root vitality, and thus promote the absorption and translocation of various mineral elements, including manganese, fundamentally restoring the tree's nutritional balance.

[0061] It is understood that those skilled in the art can make equivalent substitutions or modifications to the technical solutions and concepts of this invention, and all such substitutions or modifications should fall within the protection scope of the appended claims.

Claims

1. A method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnostics, characterized in that, Includes the following steps: S1: Obtain tea leaf samples with different mineral element contents; S2: Perform hyperspectral data acquisition and preprocessing on the sample from S1 respectively; S3: Feature band selection is performed on the hyperspectral data preprocessed in S2 using a feature selection algorithm; S4: Determine the content of various mineral elements in the sample from S1 and establish a basic database; S5: Based on the data from S3 and S4, inversion models for various mineral elements are constructed using machine learning algorithms, and the performance of the inversion models is evaluated to select the optimal inversion model for various mineral elements. S6: Input the hyperspectral data of the tea leaves to be tested into the optimal inversion model obtained in S5 to predict the content of corresponding mineral elements, and combine it with the DRIS diagnostic criteria to complete the diagnosis of whether the tea leaves are deficient in mineral elements.

2. The method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnosis according to claim 1, characterized in that, In step S2, hyperspectral data acquisition and preprocessing are performed by using a hyperspectral imaging system to acquire hyperspectral images of the leaf sample from step S1, followed by correction and preprocessing to extract the average reflectance spectrum of the region of interest on the leaf.

3. The method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnostics according to claim 2, characterized in that, The preprocessing includes one or more of multivariate scattering correction (MSC), Savitzky-Golay smoothing (SG), and first derivative (1D) processing to eliminate scattering effects, noise, and baseline drift.

4. The method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnostics according to claim 1, characterized in that, The feature selection algorithm in step S3 includes one or more of the following: Uninformed Variable Elimination (UVE), Continuous Projection Algorithm (SPA), and Competitive Adaptive Reweighted Sampling Algorithm (CARS).

5. The method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnosis according to claim 1, characterized in that, In step S5, the machine learning algorithm includes one or more of partial least squares regression (PLS), support vector regression (SVM), and random forest (RF).

6. The method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnostics according to claim 1, characterized in that, In step S5, the performance of the inversion models for the content of multiple elements is evaluated using independent test sets. The evaluation indicators include the test set determination coefficient (R²), root mean square error (RMSEP), and relative analysis error (RPD).

7. A method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnostics, as described in claims 5 and 6, characterized in that, The optimal inversion model refers to the model that simultaneously satisfies the following conditions in model evaluation: a test set determination coefficient (R²) ≥ 0.7, a root mean square error (RMSEP) ≤ 1000 mg / kg, a relative analysis error (RPD) ≥ 2.0, and the highest R² value on the test set.

8. The method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnostics according to claim 1, characterized in that, In step S6, the DRIS diagnostic criterion is the DRIS index. The DRIS index is the arithmetic mean of the partial functions f(X / A) corresponding to the ratios of multiple mineral elements in the sample to be evaluated, used to comprehensively evaluate the balance state among elements. The calculation formula of the DRIS index includes the partial functions of the element ratios, and the calculation method of the partial function f(X / A) is as follows: When X / A > x / a, f(X / A) = [ (X / A) / (x / a) - 1 ] × 1000 / CV; When X / A < x / a, f(X / A) = [ 1 - (x / a) / (X / A) ] × 1000 / CV; Where X / A is the nutrient concentration ratio of the sample to be evaluated, x / a is the average nutrient ratio of the leaves in the normal group, and CV is the coefficient of variation of the nutrient ratio of the leaves in the normal group. Based on the mineral element content data of the leaves, the element ratios and DRIS index were calculated, DRIS diagnostic criteria were constructed, and key limiting elements causing leaf yellowing were identified.

9. The method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnostics according to claim 1, characterized in that, The various mineral elements in step S4 are N, P, K, Ca, Mg, Fe, Cu, Zn, Mn, and Al.

10. The application of the method for diagnosing mineral element deficiency in tea leaves based on hyperspectral imaging and DRIS diagnosis as described in any one of claims 1-9 in the diagnosis of nutrient deficiency in tea leaves.

Images