Hyperspectral Estimation Of Tieguanyin Tree Leaf Nutrient Component Content Based On PLS Method

Total nitrogen (TN), total phosphorus (TP), total potassium (TK) are three kinds of important nutrient elements in tea. At present, traditional method of chemical analysis is common method in content detection, this method have disadvantages, such as high cost, slow speed of detection, destroy samples and cannot achieve real-time detection. Compared with traditional method, hyperspectral method have advantages, such as high speed, does not destroy samples, can realize real-time detection, etc. With the method of hyperspectral remote sensing, this study attempt to estimate three kinds of nutrient elements(TN、TP、TK) content in tieguanyin tea. With the content of total nitrogen, total phosphorus and total potassium as testing index develop hyperspectral remote sensing quantitative analysis, the purpose is providing some new reference for estimating three kinds of nutrient elements content rapidly and nondestructively in the tea combined with stoichiometric method. The main contents and conclusions are as follows:This paper had established the total nitrogen, total phosphorus, total potassium content analysis model about tieguanyin tea. This study collected62samples in tieguanyin tea, including31old leaf samples and31new leaf samples. In this experiment, total nitrogen, total phosphorus and total potassium content were detected respectively with kjeldahl determination,HCLO4-H2SO4method and flame photometer method; The spectrum information of each sample was obtained through the portable field features spectrometer(FieldSpec3) which is produced by Analytical Spectral Device (ASD) company.Using spectral effect value(leverage) and chemical residual value(residual) to test possible outlier samples, twice-detection diagnosis method was applied to eliminate the outlier samples, keep samples which were misjudged. On this basis, author use four spectral preprocessing methods (Savitzky-Golay convolution smoothing, multiple scatter correction (MSC), the first derivative and second derivative) respectively to deal with original spectral information, select the appropriate spectral preprocessing methods and improve the forecast ability of calibration model. By the content-grads method, the samples set without outlier samples were divided into calibration set and prediction set in accordance with the proportion of3:1. Calibration models of TN,TP and TK content about the calibration sets were built by partial-least-square(PLS) respectively after various pretreatment,the models had been verified through prediction set and the most optimal high spectral analysis model is as follows:(1)The best fitting high spectral analysis model with total nitrogen content:the optimal fitting model of spectral data were processed by the first derivative and the Savitzky-Golay convolution smoothing. Model determination coefficient (R2) is0.912, root-mean-squares error of cross-validation(RMSECV) is2.12, root-mean-square error of prediction (RMSEP) is1.785,residual predictive deviation (RPD)value is2.299;(2)The best fitting high spectral analysis model with total phosphorus content:the optimal fitting model of spectral data were processed by the second derivative. Model determination coefficient (R2) is0.9451, root-mean-squares error of cross-validation (RMSECV) is0.2653, root mean square prediction error (RMSEP) is0.2799, residual predictive deviation (RPD)value is3.299;(3)The best fitting high spectral analysis model with total potassium content:the optimal fitting model of spectral data were processed by the first derivative. Model determination coefficient (R2) is0.9818, root-mean-squares error of cross-validation (RMSECV) is1.274, root-mean-square error of prediction (RMSEP) is0.974, residual predictive deviation (RPD)value is6.358;From the precision of view of three kinds of nutrient element content in the fitting model, total potassium and total phosphorus are more outstanding, total nitrogen less Internal interaction validation and external validation prove that hyperspectral data quantitative analysis with partial-least-square have higher accuracy, It conforms total nitrogen, total phosphorus and total potassium content of the tea in the test requirements. Its testing process is simpler than conventional chemical methods, which can realize fast and nondestructive testing and provide a new method in rapid detection of total nitrogen, total phosphorus, total potassium content of tea in the future.