基于可见–近红外光谱技术的红茶等级判别研究

Discrimination Research on Black Tea Grade Based on Visible-Near Infrared Spectroscopy

摘要: 以遵义红茶为研究对象，基于可见–近红外光谱技术的红茶等级判别，并检验模型对红茶的判别效果。首先，将获取的原始光谱数据分别在经过卷积平滑、多元散射校正、去趋势法等8种方法预处理后，比较了八种光谱预处理方法，得到偏最小二乘判别模型为最优光谱预处理方法。然后采用连续投影算法，结合竞争自适应重加权算法和移动窗口法的连续投影算法筛选整个光谱区域的光谱特征波长变量，建立偏最小二乘辨识模型。经过比较模型质量的评价指标，结果显示，以卷积平滑预处理后的光谱数据的偏最小二乘法结合竞争性自适应重加权算法挑选特征波长建立的鉴别模型最优。该方法能较为准确、快速地鉴别出红茶的等级。

Abstract: Taking Zunyi black tea as the research object, the Black tea grade discrimination based on visible-near infrared spectroscopy technology was used, and the discrimination effect of the identification model on black tea was examined. Firstly, the obtained raw spectral data are preprocessed by 8 methods such as SG-Smoothing method, multivariate scattering correction method, detrending method and so on. Comparing these eight spectral preprocessing methods, the results show that the partial least squares discriminant model is the best spectral preprocessing method. Then, the competitive adaptive re-weighting algorithm, combined with the competitive adaptive re-weighting algorithm and the moving window method of continuous projection algorithm is used to filter the spectral characteristic wavelength variables of the entire spectral region, to establish a partial least squares identification model. After comparing the evaluation indexes of model quality, the results show that the partial least squares method of the SG-smoothing pre-processed spectral data combined with the competitive adaptive re-weighting algorithm is the best way to select the characteristic wavelength and establish the identification model. This method can identify the grade of black tea more accurately and quickly.