Rubber tree leaf nutrient content detection method based on dry leaf hyperspectrum
By screening dried powder samples and characteristic wavelengths, a multi-nutrient synergistic prediction model for rubber tree leaves was established, solving the problems of model complexity and moisture interference in existing technologies, and realizing rapid and accurate detection of nutrients in rubber tree leaves.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-15
- Publication Date
- 2026-04-03
AI Technical Summary
In existing technologies, the nutrient detection models for rubber tree leaves are complex, redundant, and impractical. There is a lack of multi-nutrient synergistic detection schemes for rubber trees, and the detection of fresh leaves is easily affected by moisture.
Using dry powder samples, common characteristic wavelengths were extracted through a characteristic wavelength screening strategy, and a partial least squares regression collaborative prediction model was established to achieve simultaneous detection of nitrogen, phosphorus, potassium, calcium and magnesium.
The model structure has been simplified, the detection accuracy and efficiency have been improved, and moisture interference has been eliminated, making it suitable for rapid and accurate multi-nutrient diagnosis of rubber trees.
Smart Images
Figure CN121783908A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of leaf detection technology, and in particular relates to a method for detecting the nutrient content of rubber tree leaves based on the hyperspectral analysis of dry leaves. Background Technology
[0002] Natural rubber is an important strategic material and industrial raw material, widely used in industry, transportation, national defense, medicine, and other fields. Rubber yield and quality are mainly affected by multiple factors, including variety, tree age, tapping intensity, fertilization, weather, season, and phenology. Fertilization is a crucial human factor influencing rubber yield and quality. Leaf nutrient diagnosis is currently a relatively mature, simple, and feasible method for diagnosing rubber tree nutrition and is the primary basis for rubber tree fertilization. The main nutrient elements in rubber tree leaves are nitrogen, phosphorus, potassium, calcium, and magnesium. Nitrogen is the most important, participating in plant protein synthesis and promoting plant tissue development. Nitrogen is also a key component of photosynthesis, aiding in enzymatic reactions and increasing the quantity and quality of plant dry matter. Phosphorus is a structural component of nucleic acids related to plant genetic information and assists in energy transfer and storage during photosynthesis. Potassium enhances plant disease resistance and regulates water and sugar absorption during photosynthesis by inducing changes in metabolites in plant tissues, while simultaneously reducing water potential in xylem vessels. Magnesium is a component of chlorophyll and is closely related to photosynthesis, significantly impacting rubber tree growth and latex production. Calcium is an essential element for the normal growth of rubber trees, participating in the formation and development of new tissues and maintaining basic cellular functions. Establishing an efficient and accurate model for detecting nutrient content in rubber tree leaves is of great significance for precise fertilization and scientific management of rubber trees.
[0003] Hyperspectral technology, with its advantages of speed and non-destructive operation, has shown potential in the inversion of plant biochemical parameters. Current research on predicting plant leaf nutrients using hyperspectral imaging largely focuses on single nutrient prediction (especially nitrogen) or uses fresh leaves for detection (susceptible to interference from moisture and leaf surface structure). A few multi-nutrient prediction studies typically involve independently selecting characteristic bands for each nutrient and building models, resulting in complex models, wavelength redundancy, and poor practicality. Furthermore, existing technologies are mostly geared towards remote sensing of field crops or forest canopies, lacking a dedicated solution for rubber trees, based on dried powder samples (to eliminate moisture interference) and a simplified multi-nutrient synergistic model. Summary of the Invention
[0004] To address the aforementioned issues, this invention proposes a method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral imaging. This method utilizes dried powder samples to eliminate moisture interference and employs a specific characteristic wavelength screening strategy to extract a set of common characteristic wavelengths that can be used simultaneously for predicting five nutrients: nitrogen, phosphorus, potassium, calcium, and magnesium. A simplified collaborative prediction model is then established to achieve rapid, synchronous, and efficient detection of nutrients in rubber tree leaves.
[0005] To achieve the above objectives, the technical solution adopted by this invention is: a method for detecting the nutrient content of rubber tree leaves based on dry leaf hyperspectral imaging, comprising the following steps: S10, collect rubber tree leaf samples and prepare them into dried leaf powder samples; S20, Obtain the hyperspectral reflectance data of the dried leaf powder sample; S30, determine the content of nitrogen, phosphorus, potassium, calcium and magnesium in the dried leaf powder sample; S40, preprocess the hyperspectral reflectance data; S50, Cooperative Feature Wavelength Extraction: Based on the preprocessed spectral data and the contents of nitrogen, phosphorus, potassium, calcium and magnesium, a competitive adaptive reweighted sampling algorithm is used to screen feature wavelengths, and several common feature wavelengths for cooperative prediction of the five nutrients nitrogen, phosphorus, potassium, calcium and magnesium are extracted from them. The common feature wavelengths are located in the near-infrared spectral region. S60, Establishment of a collaborative prediction model: Using the spectral data corresponding to the common characteristic wavelengths as input and the contents of nitrogen, phosphorus, potassium, calcium and magnesium as output, a multi-nutrient collaborative prediction model is constructed using the partial least squares regression method. S70: Hyperspectral data of the trunk and leaves of the rubber tree to be tested were collected and input into a multi-nutrient synergistic prediction model to obtain data on the content of nitrogen, phosphorus, potassium, calcium and magnesium.
[0006] Furthermore, the step of extracting common feature wavelengths for collaborative prediction in step S50 specifically includes: S51 uses a competitive adaptive reweighted sampling algorithm, taking the preprocessed spectral data as the independent variable and the contents of nitrogen, phosphorus, potassium, calcium and magnesium as the dependent variable, to screen the characteristic wavelengths and obtain the characteristic wavelength subsets corresponding to each nutrient. S52, filter out wavelengths that appear at least twice in all the subsets of characteristic wavelengths to form a set of common characteristic wavelengths; S53, from the set of common characteristic wavelengths, select wavelengths located in the near-infrared spectral region as several common characteristic wavelengths for final collaborative prediction.
[0007] Furthermore, the number of common characteristic wavelengths used for collaborative prediction in step S50 is 56, with a wavelength range between 1195 nm and 2490 nm.
[0008] Furthermore, the preprocessing in step S40 includes: resampling and / or smoothing and denoising the original hyperspectral reflectance data, and performing standard normal variable transformation.
[0009] Furthermore, the resampling interval is 5 nm; the smoothing and denoising process uses the Savitzky-Golay smoothing method.
[0010] Furthermore, in step S10, the dried leaf powder sample is prepared from fresh rubber tree leaves by drying, crushing, and sieving.
[0011] Furthermore, in step S20, the hyperspectral reflectance data is obtained by measuring the diffuse reflectance spectrum of the dried leaf powder sample using a hyperspectral analyzer.
[0012] Furthermore, the nutrient content mentioned in step S30 is determined by a continuous flow analyzer and an atomic absorption spectrophotometer.
[0013] The beneficial effects of adopting this technical solution are: Simultaneous prediction of multiple nutrients and model simplification: An innovative approach was proposed to screen for "common characteristic wavelengths" from the characteristic wavelengths of multiple nutrients, and 56 common near-infrared characteristic wavelengths were successfully extracted. Based on this, a partial least squares regression collaborative prediction model was constructed, which can simultaneously output the content of five nutrients, greatly simplifying the model structure and reducing computational complexity and the requirements for the number of instrument bands.
[0014] Balancing prediction accuracy and practicality: Using dried rubber tree leaf powder as the detection target effectively eliminates interference from moisture, pigments, and surface structures in fresh leaves, improving the stability of the relationship between spectra and nutrients and the model's prediction accuracy (e.g., RPD > 2.0 for nitrogen and calcium). Even after model simplification, it maintains good predictive ability and is highly practical.
[0015] Specificity and potential for widespread application: This invention is specifically designed for the rubber tree, an important economic tree species, addressing the practical need for rapid diagnosis of multiple nutrients. The extracted common characteristic wavelengths are concentrated in the near-infrared region, providing a clear design basis for developing low-cost, portable, dedicated spectroscopic detection equipment for rubber trees, facilitating technology transfer and widespread application.
[0016] Highly efficient and convenient: It enables a rapid process from sample preparation to one-time prediction of the content of five nutrients, significantly improving the efficiency of rubber tree nutrient diagnosis and providing an efficient technical tool for precision fertilization. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the process for detecting the nutrient content of rubber tree leaves based on dry leaf hyperspectral imaging according to the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings.
[0019] In this embodiment, see Figure 1 As shown, this invention proposes a method for detecting the nutrient content of rubber tree leaves based on dry leaf hyperspectral imaging, comprising the following steps: S10, collect rubber tree leaf samples and prepare them into dried leaf powder samples; S20, Obtain the hyperspectral reflectance data of the dried leaf powder sample; S30, determine the content of nitrogen, phosphorus, potassium, calcium and magnesium in the dried leaf powder sample; S40, preprocess the hyperspectral reflectance data; S50, Cooperative Feature Wavelength Extraction: Based on the preprocessed spectral data and the contents of nitrogen, phosphorus, potassium, calcium and magnesium, a competitive adaptive reweighted sampling algorithm is used to screen feature wavelengths, and several common feature wavelengths for cooperative prediction of the five nutrients nitrogen, phosphorus, potassium, calcium and magnesium are extracted from them. The common feature wavelengths are located in the near-infrared spectral region. S60, Establishment of a collaborative prediction model: Using the spectral data corresponding to the common characteristic wavelengths as input and the contents of nitrogen, phosphorus, potassium, calcium and magnesium as output, a multi-nutrient collaborative prediction model is constructed using the partial least squares regression method. S70: Hyperspectral data of the trunk and leaves of the rubber tree to be tested were collected and input into a multi-nutrient synergistic prediction model to obtain data on the content of nitrogen, phosphorus, potassium, calcium and magnesium.
[0020] As an optimized solution of the above embodiment, in step S10, the dried leaf powder sample is prepared from fresh rubber tree leaves after drying, crushing and sieving.
[0021] Rubber tree leaf samples were dried at 60°C and then pulverized through a 40-mesh sieve. The prepared leaf powder samples were used to determine the nutrient content and spectral reflectance.
[0022] As an optimization of the above embodiment, in step S20, the hyperspectral reflectance data is obtained by measuring the diffuse reflectance spectrum of the dried leaf powder sample using a hyperspectral analyzer.
[0023] Leaf spectral reflectance was measured using a FieldSpec4 spectrometer (ASD, USA), covering a wavelength range of 350–2500 nm. The sampling interval for 350–1000 nm was 1.4 nm, with a spectral resolution of 3 nm; the sampling interval for 1001–2500 nm was 2 nm, with a spectral resolution of 8 nm. Output spectral data were spaced 1 nm apart, resulting in a total of 2151 spectral bands for each leaf sample. The entire measurement was conducted in a laboratory equivalent to a darkroom. Leaf powder samples were filled into petri dishes (10 cm radius, 1.5 cm depth), the surface was smoothed, and the dishes were placed on a black rubber pad with approximately zero reflectance. A 50W halogen lamp was used as the light source, with an incident angle of 45° and a distance of 70 cm from the sample surface. The sensor probe had a 25° field of view and was positioned 15 cm vertically above the sample surface. White plate calibration was performed before spectral measurement and every 15 minutes during the measurement period. Ten spectral curves were collected for each sample, and the spectral reflectance data for that leaf sample was obtained by arithmetic averaging.
[0024] As an optimization of the above embodiment, the nutrient content in step S30 is determined by a continuous flow analyzer and an atomic absorption spectrophotometer.
[0025] Weigh 0.1 g of the prepared leaf powder sample into a digestion test tube, add 5 ml of concentrated H2SO4, and digest at 380℃ for 1 h. When the sample digests to a brown transparent liquid, add H2O2 until clear, continue digesting for 5 min, remove and cool, transfer to a 100 mL volumetric flask, add 5 mL of 20 g / L lanthanum solution, and dilute to the mark, mixing well. Measure the nitrogen and phosphorus content of the test solution using a continuous flow analyzer (AA3), and the potassium, calcium, and magnesium content using an atomic absorption spectrophotometer (AA-6300C). Each leaf sample is analyzed twice, and the average value is taken as the leaf nitrogen, phosphorus, potassium, calcium, and magnesium content.
[0026] As an optimization of the above embodiment, the preprocessing in step S40 includes: resampling and / or smoothing and denoising the original hyperspectral reflectance data, and performing standard normal variable transformation.
[0027] To improve model execution speed, the reflectance spectral data was first resampled at 5 nm intervals before subsequent analysis. Due to the high noise in high-frequency reflectance spectra, and to facilitate spectral resampling, wavelengths at both ends were removed, retaining only the reflectance data for the 398–2497 nm range. A moving window averaging method was used to resample the 398–2497 nm reflectance data into 5 nm intervals (400, 405, 410, …, 2495, 2495 nm).
[0028] Using the principle that the absolute value of the studentized residual is greater than 3, 28 samples were removed, leaving 537 samples for modeling research. The Kennard-Stone (KS) algorithm was used to divide the 537 leaf samples into a modeling set, selecting 358 as the calibration set and the remaining 179 as the test set.
[0029] Savitzky-Golay (SG) smoothing was used to further eliminate noise and improve the quality of spectral data. The window size and polynomial degree of SG smoothing were 3 and 2, respectively. Based on this, the spectra were preprocessed using the following four methods: (1) spectral reflectance (R); (2) Standard Normal Transform (SNV); (3) First Derivative (FirstDer); (4) Standard Normal Transform followed by First Derivative (SNV+FirstDer). Through screening, the optimal spectral preprocessing method suitable for the co-prediction of nitrogen, phosphorus, potassium, calcium and magnesium was found to be SNV.
[0030] As an optimization of the above embodiment, the step of extracting common feature wavelengths for collaborative prediction in step S50 specifically includes: S51 uses a competitive adaptive reweighted sampling algorithm, taking the preprocessed spectral data as the independent variable and the contents of nitrogen, phosphorus, potassium, calcium and magnesium as the dependent variable, to screen the characteristic wavelengths and obtain the characteristic wavelength subsets corresponding to each nutrient. S52, filter out wavelengths that appear at least twice in all the subsets of characteristic wavelengths to form a set of common characteristic wavelengths; S53, from the set of common characteristic wavelengths, select wavelengths located in the near-infrared spectral region as several common characteristic wavelengths for final collaborative prediction.
[0031] The 56 common characteristic wavelengths used for collaborative prediction in step S50 range from 1195 nm to 2490 nm.
[0032] Validated on the test set, the determination coefficients (R²) of the synergistic prediction models for nitrogen, phosphorus, potassium, calcium, and magnesium were 0.851, 0.518, 0.755, 0.810, and 0.648, respectively; the root mean square errors (RMSEs) were 1.266 g / kg, 0.502 g / kg, 1.402 g / kg, 1.032 g / kg, and 0.430 g / kg, respectively; and the relative analytical errors (RPDs) were 2.574, 1.438, 1.935, 2.252, and 1.642, respectively. In terms of predictive performance, the model prediction RPDs for each nutrient element were all greater than 1.4. Nitrogen showed excellent estimation ability, while calcium, potassium, magnesium, and phosphorus showed good estimation ability.
[0033] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.
Claims
1. A method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral analysis, characterized in that, Including the following steps: S10, collect rubber tree leaf samples and prepare them into dried leaf powder samples; S20, Obtain the hyperspectral reflectance data of the dried leaf powder sample; S30, determine the content of nitrogen, phosphorus, potassium, calcium and magnesium in the dried leaf powder sample; S40, preprocess the hyperspectral reflectance data; S50, Cooperative Feature Wavelength Extraction: Based on the preprocessed spectral data and the contents of nitrogen, phosphorus, potassium, calcium and magnesium, a competitive adaptive reweighted sampling algorithm is used to screen feature wavelengths, and several common feature wavelengths for cooperative prediction of the five nutrients nitrogen, phosphorus, potassium, calcium and magnesium are extracted from them. The common feature wavelengths are located in the near-infrared spectral region. S60, Establishment of a collaborative prediction model: Using the spectral data corresponding to the common characteristic wavelengths as input and the contents of nitrogen, phosphorus, potassium, calcium and magnesium as output, a multi-nutrient collaborative prediction model is constructed using the partial least squares regression method. S70: Hyperspectral data of the trunk and leaves of the rubber tree to be tested were collected and input into a multi-nutrient synergistic prediction model to obtain data on the content of nitrogen, phosphorus, potassium, calcium and magnesium.
2. The method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral analysis according to claim 1, characterized in that, The step of extracting common feature wavelengths for collaborative prediction in step S50 specifically includes: S51 uses a competitive adaptive reweighted sampling algorithm, taking the preprocessed spectral data as the independent variable and the contents of nitrogen, phosphorus, potassium, calcium and magnesium as the dependent variable, to screen the characteristic wavelengths and obtain the characteristic wavelength subsets corresponding to each nutrient. S52, filter out wavelengths that appear at least twice in all the subsets of characteristic wavelengths to form a set of common characteristic wavelengths; S53, from the set of common characteristic wavelengths, select wavelengths located in the near-infrared spectral region as several common characteristic wavelengths for final collaborative prediction.
3. The method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral analysis according to claim 2, characterized in that, The 56 common characteristic wavelengths used for collaborative prediction in step S50 range from 1195 nm to 2490 nm.
4. The method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral analysis according to claim 1, characterized in that, The preprocessing described in step S40 includes: resampling and / or smoothing and denoising the original hyperspectral reflectance data, and performing standard normal variable transformation.
5. The method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral analysis according to claim 4, characterized in that, The resampling interval is 5 nm; the smoothing and denoising process uses the Savitzky-Golay smoothing method.
6. The method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral analysis according to claim 1, characterized in that, In step S10, the dried leaf powder sample is prepared from fresh rubber tree leaves by drying, crushing, and sieving.
7. The method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral analysis according to claim 1, characterized in that, In step S20, the hyperspectral reflectance data is obtained by measuring the diffuse reflectance spectrum of the dried leaf powder sample using a hyperspectral analyzer.
8. The method for detecting nutrient content in rubber tree leaves based on dry leaf hyperspectral analysis according to claim 1, characterized in that, The nutrient content mentioned in step S30 is determined by a continuous flow analyzer and an atomic absorption spectrophotometer.