Abstract:

A highly accurate and stable calibration model of total nitrogen content in green tea was established based on
partial least square regression. The model demonstrated great predictive performance after the first derivative processing
and smoothing in the spectral range of 4 691–3 959 cm-1 and 5 126–4 848 cm-1. The correlation coefficient of predicted and
chemically measured values in the validation set was 0.993 9 and the RMSEP was 0.092 5%. The prediction performance
was improved by developing sub-models of total nitrogen in green tea according to different tenderness. The prediction
performance of the tenderness level 3 model was the best. The RMSEP in the validation set was the smallest, 0.037 9%,
and the correlation coefficient was 0.996 1. Thus, near infrared spectroscopy can provide a green, rapid and efficient
technique for the detection of total nitrogen in green tea and quality evaluation.

Key words: near infrared spectroscopy, green tea, total nitrogen content, sub-model