Method for detecting nitrogen contents of organs and tissues of different varieties of oilseed rapes based on visible near infrared spectrum

Abstract

The invention discloses a method for detecting the nitrogen content of organ tissues of different varieties of oilseed rapes based on visible near-infrared spectroscopy, which comprises the following steps: firstly, collecting oilseed rape leaf, shell, stalk and root system samples under the conditions of multiple varieties and multiple nitrogen fertilizer levels, and acquiring spectral data within the range of 430-2500nm by using a visible near-infrared spectroscopy; and a Kjeldahl method is synchronously adopted to measure the real nitrogen content as a reference value. Preprocessing the spectral data, including de-noising, standard normal variable transformation, multivariate scatter correction and derivative processing, so as to weaken the influence of scattering and baseline drift; characteristic wavelengths related to the nitrogen content are screened through stepwise regression, variable projection importance, competitive self-adaptive reweighted sampling, a continuous projection algorithm and other methods, and a random forest model, a support vector machine model, a partial least squares discriminant analysis model, a partial least squares regression model, a support vector regression model, an XGBoost model and other models are combined. And respectively constructing a classification identification model and a regression prediction model.

Description

Technical Field

[0001] This invention relates to the field of agricultural information detection and crop nutrition monitoring technology, and more specifically, to a method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy. Background Technology

[0002] Nitrogen is one of the most important nutrients for crop growth, playing a crucial role in photosynthesis, protein synthesis, and yield and quality formation. As a major oilseed crop globally, rapeseed's nitrogen absorption and utilization efficiency directly affects oil yield and seed quality. In actual production, the rationality of nitrogen fertilizer application rate and method not only affects crop yield but also has significant implications for ecological environmental protection. However, problems such as low nitrogen fertilizer utilization and environmental pollution caused by excessive nitrogen application have long existed. Therefore, rapid and accurate detection of nitrogen content in different parts of rapeseed is of great value for efficient nitrogen fertilizer utilization and precision fertilization.

[0003] Traditional methods for detecting crop nitrogen primarily rely on chemical analysis techniques, such as the Kjeldahl method and the Dumas combustion method. While these methods are accurate, they suffer from drawbacks such as cumbersome operation, long processing times, high consumption of chemical reagents, and high testing costs, making it difficult to achieve large-scale real-time monitoring in the field. Especially under the complex conditions of rapeseed with multiple varieties, multiple growth stages, and multiple nitrogen fertilizer levels, traditional methods cannot meet the needs for rapid, high-throughput, and non-destructive testing, which to some extent limits the development of intelligent and refined nitrogen management.

[0004] In recent years, near-infrared spectroscopy (NIRS), as a rapid, green, and non-destructive detection technique, has been widely used in crop quality analysis and nutrient monitoring due to its ability to reflect the vibrational characteristics of molecular groups in plant tissues. Existing research has shown that spectroscopic methods can extract spectral features related to nitrogen content without damaging the sample, and predictive models can be built by combining statistical and machine learning methods. However, most existing studies focus on single organs or single varieties, and the models have limited generalization ability across varieties, organs, and multiple nitrogen fertilizer levels, with insufficient selection of characteristic wavelengths and weak model robustness.

[0005] Therefore, we urgently need to design a method for detecting nitrogen content in the organs and tissues of different rapeseed varieties based on visible and near-infrared spectroscopy to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to solve the technical problems mentioned in the background section, and to provide a method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy, comprising the following steps:

[0007] S1. Multiple rapeseed varieties are planted under the same natural environment, and different nitrogen fertilizer application levels are set. Sampling is carried out in stages from the seedling stage to the initial flowering stage. The sampling objects include leaves, fruit shells, stems and roots. The collected samples are dried, crushed and sieved.

[0008] S2. Use a visible and near-infrared spectrometer to perform spectral scanning on the processed sample. The spectral range is 430–2500 nm. The incident light source is a halogen lamp. The acquisition mode is diffuse reflection mode. Each sample is acquired no less than 10 times and the average spectral curve is taken.

[0009] S3. The nitrogen content of the sample was chemically analyzed according to the standard methods of the agricultural industry. The Kjeldahl method was used to determine the true nitrogen content, which was used as a reference value for model building.

[0010] S4. Preprocess the spectral data, including denoising, scattering correction, mean centering, standard normal variable transformation, multivariate scattering correction, and first or second derivative transformation, to eliminate spectral baseline drift and noise interference.

[0011] S5. Based on the preprocessed spectral data, a feature variable selection algorithm is used to screen the feature wavelength range that is highly correlated with the nitrogen content of the sample.

[0012] S6. Using the characteristic wavelength as input variable, construct a classification model and a regression prediction model, which are used for rapeseed variety and organ identification and quantitative detection of nitrogen content, respectively.

[0013] S7. Perform cross-validation on the model, calculate the accuracy, coefficient of determination, and root mean square error to evaluate the robustness and generalization ability of the model.

[0014] As a preferred embodiment of the present invention, the rapeseed varieties are Qinyou 908, Zheda 619, Zheda 635, Zheda 649, Zheda Pink, and Zheda Orange, totaling six varieties. The nitrogen fertilizer level is N0–N5 in six gradients, and the nitrogen application rate ranges from 0 to 300 kg / hm. 2 A randomized block design was used and repeated 3 times.

[0015] As a preferred technical solution of the present invention, the sample collection method in S1 is as follows: samples are collected once each at the seedling stage, bolting stage and initial flowering stage of rapeseed, and 8 plants are randomly selected from each plot for mixed samples, resulting in a total of 324 samples for spectral analysis and modeling.

[0016] As a preferred technical solution of the present invention, the spectral preprocessing step further includes: removing random noise by smoothing filtering, eliminating particle effects and scattering effects by standard normal variable transformation and multivariate scattering correction, and highlighting spectral detail features by derivative transformation to enhance band information related to nitrogen content.

[0017] As a preferred technical solution of the present invention, the feature wavelength screening method includes stepwise regression, variable projection importance method, competitive adaptive reweighted sampling method and continuous projection algorithm, and determines the optimal wavelength subset based on the cross-validation results.

[0018] As a preferred technical solution of the present invention, the classification model is used to identify rapeseed varieties and different organs, and the classification model includes at least one of random forest classifier, support vector machine classifier and partial least squares discriminant analysis model.

[0019] As a preferred technical solution of the present invention, the regression prediction model is used to quantitatively analyze the nitrogen content of rapeseed samples. The regression prediction model includes at least one of partial least squares regression model, support vector regression model, and regression model based on extreme gradient boosting algorithm.

[0020] As a preferred technical solution of the present invention, the model evaluation indicators include:

[0021] ① Overall accuracy, sensitivity, and specificity of the classification model;

[0022] ② The coefficient of determination R of the regression model 2 This is used to reflect the degree of fit between the predicted value and the measured value;

[0023] ③ Root mean square error (RMSE) is used to characterize prediction bias;

[0024] The robustness of the model was verified using the K-fold cross-validation method.

[0025] As a preferred technical solution of the present invention, by introducing sample data under different nitrogen fertilizer levels during the modeling process, a training set containing multiple nitrogen fertilizer gradient information is constructed, so that the established spectral prediction model can adapt to low to high nitrogen levels at the same time, thereby improving the applicability and robustness of the model to actual field detection.

[0026] As a preferred technical solution of the present invention, the detection model can realize rapid and non-destructive detection of multiple organ samples such as rapeseed leaves, fruit shells, stems and roots, and combine spectral characteristic variables to realize variety identification and quantitative prediction of nitrogen content, thereby achieving high-throughput and real-time nitrogen nutrition diagnosis under field conditions.

[0027] Beneficial effects:

[0028] The proposed method for detecting nitrogen content in rapeseed varieties and organ tissues based on visible and near-infrared spectroscopy enables rapid detection under different growth stages and nitrogen fertilizer levels, avoiding the problems of cumbersome operation, long detection cycle, and significant environmental pollution associated with traditional chemical methods such as Kjeldahl nitrogen determination. This method significantly shortens the detection time and reduces chemical reagent consumption through non-destructive acquisition and preprocessing of spectral data, achieving a green and environmentally friendly detection process.

[0029] This invention effectively extracts key spectral features related to nitrogen content by introducing feature wavelength screening methods such as stepwise regression, variable projection importance (VIP), competitive adaptive reweighted sampling (CARS), and continuous projection algorithm (SPA), thus avoiding the interference of redundant information on modeling accuracy. By combining various machine learning models such as random forest, support vector machine, partial least squares discriminant analysis, partial least squares regression, support vector regression, and XGBoost, the established detection model exhibits high accuracy and stability in both classification and regression prediction, thereby improving the model's robustness across varieties, organs, and nitrogen fertilizer levels.

[0030] The detection method established in this invention can be widely applied in both the field and the laboratory. It can perform rapid quantitative analysis of multiple organs, including leaves, fruit shells, stems, and roots. Furthermore, it can combine the spectral characteristics of different nitrogen fertilizer levels to form a more adaptable nitrogen nutrition diagnostic model. This method not only provides a scientific basis for precision fertilization of rapeseed but also offers new technical means for crop nitrogen management and high-throughput breeding, demonstrating broad prospects for widespread application. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 Flowchart of a method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with embodiments and appendices. Figure 1 The present invention will be further described in detail below. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0034] The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy includes the following steps:

[0035] I. Sample Collection and Basic Data Acquisition;

[0036] 1. Experimental Design and Sample Collection:

[0037] Experimental Design: Multiple rapeseed varieties were selected as research subjects. The selection criteria included parameters such as region, oil content, yield, maturity period, and flower color. The varieties selected in this application are as follows:

[0038] This study investigated six rapeseed varieties (genotypes): Qinyou 908 (QY908), Zheda 619 (ZD619), Zheda 635 (ZD635), Zheda 649 (ZD649), Zheda Pink (ZDFS), and Zheda Orange (ZDJH). Qinyou 908 is a hybrid variety bred by the Shaanxi Hybrid Rapeseed Research Center, with an oil content of 46.24%, suitable for planting in the upper, middle and lower reaches of the Yangtze River; Zheda 619 is a leading rapeseed variety in Zhejiang Province, a semi-winter conventional variety with an oil content of 45.20%, widely planted in Zhejiang Province; Zheda 635 is a mid-maturing semi-winter rapeseed variety with an oil content of 50.64%, suitable for planting in Zhejiang as a double-low oil and high oil content variety; Zheda 649 is a semi-winter rapeseed variety with an oil content of 50.89%, a double-low oil and high oil content variety with a moderate maturity and wide adaptability; Zheda Pink and Zheda Orange are colored rapeseed varieties bred by our laboratory with stable petal color and good yield and quality.

[0039] Field trials were conducted on the selected rapeseed varieties: six nitrogen fertilizer gradients were set up (N0-N5, nitrogen application rate 0-300 kg / hm). 2 The randomized block design was used and repeated 3 times.

[0040] Sample collection: This application mainly collects samples of leaves, pods, stems and roots.

[0041] Leaves: Samples were collected three times from the seedling stage to the initial flowering stage, with 8 plants per plot as mixed samples, for a total of 324 samples.

[0042] Roots, stems, and pods: Rapeseed root, stem, and pod samples were collected at harvest time, with 3 replicates per plot. Ten rapeseed plants were randomly selected from each replicate. After threshing, their roots, stems, and pods were separated and mixed separately as samples for that replicate. A total of 324 root, stem, and pod samples were collected.

[0043] 2. Spectral Data Acquisition: Spectral data was acquired using the OFS2500 ground object spectrometer. The acquisition process is as follows:

[0044] Rapeseed sample pretreatment: For rapeseed leaf samples, blanching was performed at 105℃ for 30 minutes, followed by drying at 60℃ to constant weight. For rapeseed root, stem, and silique samples, residual soil was removed and the samples were cleaned, sun-dried, and then dried at 60℃ to constant weight. The dried plant samples were then ground into powder using a grinder and passed through a 60-mesh sieve for testing.

[0045] Spectral acquisition procedure: Before acquiring spectra, preheat the spectrometer for 30 minutes to reduce baseline drift caused by environmental factors. Before measurement, use a standard reference plate for reflectance correction. Spread the powder sample flat in the sample chamber, flatten it, and then perform spectral acquisition. Each acquisition consists of 10 scans, and the average value is recorded as the measurement value for that sample. Remove high-noise beginning and end spectra, retaining the spectrum in the 430-2500 nm range for subsequent analysis.

[0046] 3. Chemical determination of nitrogen content: In accordance with the industry standard "NY / T 2017-2011" (Determination of nitrogen, phosphorus and potassium in plants), the nitrogen content of the sample was determined by sulfuric acid-hydrogen peroxide digestion and then by a fully automated flow analyzer.

[0047] II. Preprocessing of raw spectral data;

[0048] 1. The original data is denoised using a polynomial smoothing algorithm. The calculation formula and parameter settings are as follows:

[0049]

[0050] Where, x k x is the smoothed spectral reflectance at spectral band k. k+i The spectral reflectance value at the spectral band k+i, (2ω+1) is the window size, H is the normalization coefficient, and h i is the weighted smoothing coefficient for the bands within the window.

[0051] In this invention, the SG smoothing filter window size is set to 11, and the polynomial order is set to 3.

[0052] Following the noise reduction process in step 1, the first derivative is used to further eliminate baseline drift and enhance spectral resolution. The specific calculation method is as follows:

[0053]

[0054] Where g is the window width, x k With x k+g Let be the spectral values ​​at wavelength point k and wavelength point k+g.

[0055] The light scattering effect caused by the inhomogeneity of the physical properties of the sample particles is eliminated by standard normal variable transformation. The calculation method is as follows:

[0056]

[0057] Where, x snv Here is the spectrum after SNV processing, and x is the converted spectrum. The value is the average of the spectra at each wavelength in the spectrum, where m is the number of wavelength points in the spectrum, and x is the average of the values ​​at each wavelength in the spectrum. k denoted as , where is the spectral value at wavelength point k.

[0058] III. Characteristic wavelength screening;

[0059] 1. The continuous projection algorithm is used to screen feature wavelengths by minimizing collinearity in the variable space. The specific steps are as follows:

[0060] Step 1: Let the entire spectral data be M-dimensional, and the extracted spectral data be N-dimensional, with the i-th column of the spectral matrix being xi;

[0061] Step 2: Calculate the projection of xi onto the spectral data of the other columns respectively;

[0062] Step 3: Select the wavelength with the largest projection vector influence as the characteristic wavelength;

[0063] Step 4: Exclude the selected spectral vectors and repeat the above calculation process until the selected spectral data is reduced to N dimensions.

[0064] 2. Using the random frog-jumping algorithm, a subset of variables is randomly generated, and high-frequency wavelengths are filtered using the cross-validation error (RMSEcv). The specific steps are as follows:

[0065] Step 1: Select a subset of variables M0 through random initialization, where the data dimension of this subset of variables is M.

[0066] Step 2: Generate a new variable subset N0 through the positive distribution effect of M0. The data dimension of this variable subset is N. Build regression models or discriminant models based on M0 and N0 respectively, depending on the model task, and obtain the root mean square error of cross-validation (RMSECV). Compare the differences in the RMSECV of different subsets and select the variable subset with the smaller RMSECV. Repeat this process K times to obtain K different variable subsets.

[0067] Step 3: Count the number of times each spectral variable appears in the K subsets of variables, and select the spectral variables with high frequency of occurrence as the characteristic wavelengths.

[0068] 3. Using the weighted regression coefficient method, the wavelengths with high weights are selected based on the absolute value of the regression coefficients of the PLS model.

[0069] IV. Classification and Prediction Model Construction;

[0070] 1. Sample Classification Model Construction: This invention employs three classic supervised machine learning algorithms to construct the sample classification model: Naive Bayes, Random Forest, and Support Vector Classification. The specific methods are as follows:

[0071] A strategy combining grid search and k-fold cross-validation is employed to systematically optimize the key hyperparameters of the model. For support vector classification, key hyperparameters include the penalty coefficient C (0.1, 1, 10, 100) and the parameter g (0.001, 0.01, 0.1, 1) of the RBF kernel function; for random forest, these include the number of decision trees n_estimators (50, 100, 200, 300) and the number of features considered when splitting nodes max_features ('auto', 'sqrt', 'log2'). These candidate value lists are combined into a multi-dimensional parameter grid, where each node represents a unique combination of hyperparameters.

[0072] To ensure the fairness and robustness of the evaluation and to avoid information leakage on independent test sets, grid search is performed within the training set. This process traverses all hyperparameter combinations in the grid, and for each combination, k-fold cross-validation (k=10 in this study) is used to train and evaluate the model on the training set. Specifically, the training set is divided into k disjoint subsets. Each time, k-1 subsets are used for model training, and the remaining subset is used for validation, repeated k times.

[0073] For each set of hyperparameters, the model's performance in k cross-validations was recorded.

[0074] Ultimately, the hyperparameter combination that yielded the best average performance metric in cross-validation was selected as the optimal configuration for the model.

[0075] 2. Construction of Spectral Quantitative Prediction Model: This invention employs various machine learning regression algorithms to construct the spectral quantitative prediction model, including Bayesian ridge regression, support vector regression, and partial least squares regression. The specific methods are as follows:

[0076] For Support Vector Regression (SVR) and similar models whose performance is highly dependent on hyperparameter selection, we employed a grid search combined with k-fold cross-validation. For SVR, the key hyperparameters for optimization include the regularization parameter C, which controls the penalty for samples exceeding the margin boundary; the kernel function parameter gamma (when using the RBF kernel), which defines the influence range of a single training sample; and the insensitivity interval size epsilon. The core hyperparameter for SVR optimization is the number of latent variables, n_components. Choosing too few latent variables can lead to underfitting, while too many may introduce noise and cause overfitting.

[0077] Unlike the models mentioned above, a core advantage of Bayesian Ridge Regression lies in its internalized, automatic parameter optimization mechanism. This model treats hyperparameters such as regularization strength (λ) and data noise accuracy (α) as probability distributions that need to be learned from the data. During training, the posterior distributions of these hyperparameters are automatically inferred in an iterative manner by maximizing the marginal log-likelihood function of the data.

[0078] V. Model Training and Evaluation System Construction: The overall accuracy and confusion matrix are used to evaluate the model's performance. The formula for calculating the overall accuracy is as follows:

[0079]

[0080] Among them, TP, TN, FP, and FN represent the number of true positive, true negative, false positive, and false negative samples, respectively.

[0081] A confusion matrix is ​​an N×N table (where N is the number of classes) that visually displays the correspondence between the model's predictions and the true class labels. The diagonal elements of the matrix represent the number of correctly classified samples, while the off-diagonal elements represent the number of misclassified samples. By analyzing the confusion matrix, the classes that the model is prone to confuse can be clearly identified.

[0082] The dataset was divided into training and test sets in a 7:3 ratio, and 10-fold cross-validation was used for validation. For the regression model, we used the coefficient of determination (R²). 2 The evaluation is based on the root mean square error (RMSE) and the coefficient of determination R. 2 The definition is given in formula (2-6), where y i For the observed values, These are the model's predicted values. R is the mean of the observed values. 2 The range of values ​​is (-∞, 1], when R 2=1 indicates that the model fits the sample perfectly, 0 = .... <R 2 A value less than 1 indicates that the model can explain a certain proportion of the variance in the observations. 2 A value less than 0 indicates extremely poor model performance.

[0083]

[0084] The root mean square error (RMSE) is defined in formula (2-7) and reflects the average deviation between the model's predicted values ​​and the actual values. A smaller RMSE value indicates a smaller prediction error. By jointly observing R... 2 The performance of RMSE on the training and test sets can be used to determine the model's fitting ability, prediction accuracy, and whether there is overfitting or underfitting.

[0085]

[0086] Select test set R 2 The model combination with the highest RMSE and lowest RMSE (such as SPA screening wavelength + PLSR regression).

[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy, characterized in that, Includes the following steps: S1. Multiple rapeseed varieties are planted under the same natural environment, and different nitrogen fertilizer application levels are set. Sampling is carried out in stages from the seedling stage to the initial flowering stage. The sampling objects include leaves, fruit shells, stems and roots. The collected samples are dried, crushed and sieved. S2. Use a visible and near-infrared spectrometer to perform spectral scanning on the processed sample. The spectral range is 430–2500 nm. The incident light source is a halogen lamp. The acquisition mode is diffuse reflection mode. Each sample is acquired no less than 10 times and the average spectral curve is taken. S3. The nitrogen content of the sample was chemically analyzed according to the standard methods of the agricultural industry. The Kjeldahl method was used to determine the true nitrogen content, which was used as a reference value for model building. S4. Preprocess the spectral data, including denoising, scattering correction, mean centering, standard normal variable transformation, multivariate scattering correction, and first or second derivative transformation, to eliminate spectral baseline drift and noise interference. S5. Based on the preprocessed spectral data, a feature variable selection algorithm is used to screen the feature wavelength range that is highly correlated with the nitrogen content of the sample. S6. Using the characteristic wavelength as input variable, construct a classification model and a regression prediction model, which are used for rapeseed variety and organ identification and quantitative detection of nitrogen content, respectively. S7. Perform cross-validation on the model, calculate the accuracy, coefficient of determination, and root mean square error to evaluate the robustness and generalization ability of the model.

2. The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy according to claim 1, characterized in that, The rapeseed varieties mentioned are Qinyou 908, Zheda 619, Zheda 635, Zheda 649, Zheda Pink, and Zheda Orange, totaling six varieties. The nitrogen fertilizer levels are in six gradients from N0 to N5, and the nitrogen application rate ranges from 0 to 300 kg / hm. 2 A randomized block design was used and repeated 3 times.

3. The method according to claim 1, characterized in that, The sample collection method in S1 is as follows: samples are collected once each at the seedling stage, bolting stage and initial flowering stage of rapeseed, and 8 plants are randomly selected from each plot for mixed samples, resulting in a total of 324 samples for spectral analysis and modeling.

4. The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy according to claim 1, characterized in that, The spectral preprocessing steps further include: removing random noise using smoothing filtering, eliminating particle effects and scattering effects using standard normal variable transformation and multivariate scattering correction, and highlighting spectral detail features using derivative transformation to enhance band information related to nitrogen content.

5. The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy according to claim 1, characterized in that, The feature wavelength selection method includes stepwise regression, variable projection importance, competitive adaptive reweighted sampling, and continuous projection algorithm, and determines the optimal wavelength subset based on the cross-validation results.

6. The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy according to claim 1, characterized in that, The classification model is used to identify rapeseed varieties and different organs. The classification model includes at least one of random forest classifier, support vector machine classifier, and partial least squares discriminant analysis model.

7. The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy according to claim 1, characterized in that, The regression prediction model is used to quantitatively analyze the nitrogen content of rapeseed samples. The regression prediction model includes at least one of partial least squares regression model, support vector regression model, and regression model based on extreme gradient boosting algorithm.

8. The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy according to claim 1, characterized in that, The model evaluation metrics include: ① Overall accuracy, sensitivity, and specificity of the classification model; ② The coefficient of determination R of the regression model 2 This is used to reflect the degree of fit between the predicted value and the measured value; ③ Root mean square error (RMSE) is used to characterize prediction bias; The robustness of the model was verified using the K-fold cross-validation method.

9. The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy according to claim 1, characterized in that, By introducing sample data under different nitrogen fertilizer levels during the modeling process, a training set containing multiple nitrogen fertilizer gradient information is constructed, enabling the established spectral prediction model to adapt to both low and high nitrogen levels, thereby improving the model's applicability and robustness to actual field testing.

10. The method for detecting nitrogen content in organ tissues of different rapeseed varieties based on visible and near-infrared spectroscopy according to claim 1, characterized in that, The detection model enables rapid and non-destructive testing of multiple organ samples, including rapeseed leaves, husks, stems, and roots. It also combines spectral characteristic variables to achieve variety identification and quantitative prediction of nitrogen content, thereby enabling high-throughput, real-time nitrogen nutrition diagnosis under field conditions.

Images