FOOD SCIENCE ›› 2020, Vol. 41 ›› Issue (18): 283-287.doi: 10.7506/spkx1002-6630-20190906-089

• Safety Detection • Previous Articles     Next Articles

Selection of Characteristic Near Infrared Spectra and Establishment of Prediction Model for Qingzhuan Tea Quality

WANG Shengpeng, GONG Ziming, ZHENG Pengcheng, LIU Panpan, TENG Jing, GAO Shiwei, GUI Anhui   

  1. (Institute of Fruit and Tea, Hubei Academy of Agricultural Sciences, Wuhan 430064, China)
  • Online:2020-09-25 Published:2020-09-18

Abstract: Near infrared spectroscopy (NIRS) was used for rapid and nondestructive evaluation of the quality of Qingzhuan tea. Under the premise of ensuring the integrity of samples, spectra were acquired and preprocessed, and the characteristic spectral intervals were selected by synergy interval partial least squares?(siPLS). Furthermore, principal component analysis was performed, and a prediction model was established for the sensory evaluation score of Qingzhuan tea by Jordan-Elman back propagation-artificial neural network. The best pretreatment method was multiple scatter correction + 2nd derivative, the characteristic spectral intervals were 4 377.6–4 751.7, 4 755.6–5 129.7, 6 262.7–6 633.9, and 7 386–7 756.3 cm-1, and the cumulative contribution rate of the first three principal components in the characteristic spectral regions was 99.15%. The transfer function of the model was tanh with root mean square error of cross-validation set (RMSECV) and determinant coefficient for prediction (R2p) of 0.386 and 0.973, respectively. The root means square error and the determinant coefficient were respectively 0.393 and 0.971 for unknown Qingzhuan tea samples. The results showed that NIRS combined with Jordan-Elman back propagation-artificial neural network could allow rapid and accurate evaluation the quality of Qingzhuan tea scoring in the range of 75.00–93.00.

Key words: Qingzhuan tea; quality; near infrared spectroscopy; synergy interval partial least squares; back propagation-artificial neural network

CLC Number: