Rapid Detection Of Total Nitrogen Content In Fresh Tea Leaves Based On Spectral Analysis

Visible-near Infrared Spectroscopy (VIR-NIRS) technique is a novel and wide-application analysis method. With its advantages of rapidity, high accuracy, non-destruction, good repeatability and real-time on-line measurement, it was appealed to numerous fields including agriculture, food industry, pharmacy and petrochemical industry. Nitrogen is one of the key factors for tea plant growth, yield and quality. VIR-NIRS technique was applied to accurately diagnose tea nitrogen conditions in time, which has important significance to real-time monitor the way that tea plant grows and guide fertilizer management.The paper took fresh tea leaf as the study object. Total nitrogen content of leaves was measured by using VIR-NIRS technology. The mathematical model between VIR-NIRS and total nitrogen content of fresh leaf was built through various pretreatment and regression methods.Firstly, the SPAD value of leaves was measured by SPAD-502, according to the correlative relationship between SPAD value and total nitrogen content. 120 samples in which total nitrogen content has a certain gradient and uniform distribution were chosen. Then the spectrum information of samples was collected by using ASD Field Spec instrument. Finally, the samples were picked up and the true value of total nitrogen content was measured by Kjeldahl method. The spectrum characteristic was analyzed and found one wave crest and two wave troughs in visible area. The spectrum index of reflection presents a high reflection platform in short-wave NIR area. The spectrum characteristic was correlated with water content of leaves in long-wave NIR area. The sensitive spectrum bands were found through correlation coefficient analysis. Combining spectroscopy pretreatment methods such as Normalization, Golay, first and second derivative with calibration methods, the linear model of total nitrogen content was established. Using the wave bands of 449～736nm, 1406～1602nm and 1693～2399nm, the predicting result of PLS model adopting normalization and second derivative pretreatment methods was best. The correlation coefficient (R) of the model was 0.9851 and root mean square error of cross-validation (RMSECV) was 0.0683. The model was used to predict 34 samples of validation set. The correlation coefficient was 0.8897. Root mean square error of prediction (RMSEP) was 0.2135. Average relative error was 7.196%. Maximal error was 19.819%.In order to improve the precision of model, genetic algorithm was introduced to choose characteristic wavelength. Using six wavelength numbers included 701 nm, 695 nm, 1462 nm, 552 nm, 668 nm, 680 nm, RBF nuclear function was used to build the Support Vector Machine (SVM) regression model between VIR-NIRS spectrum and total nitrogen content. The correlation coefficient of the model was 0.9342. RMSECV was 0.1586. The correlation coefficient of predicting model was 0.9284. RMSEP was 0.1903. Average relative error was 6.825%. Maximal error was 14.418%. The result showed that the combination of GA and SVM was an effective method to predict nitrogen content of fresh tea leaf.