采用基于线性渐变滤光片分光原理的微型近红外光谱仪对2012年和2013年的西湖龙井和普通扁形茶建立真伪识别模型。 分别对不同年份、 不同保存期样品的近红外光谱数据进行PCA分解, 并根据PCA得分分布的数学特征选取代表性样品, 建立PLS-DA模型, 从数学原理的角度对模型不适应性的原因加以分析, 并对其进行修正, 结合外部盲样的验证, 有效地增强了模型适应性。 研究结果表明, 针对西湖龙井和普通扁形茶, 采用不同年份样品近红外光谱数据共同建模可有效提高模型对不同年份样品的识别正确率; 采用不同保存期样品近红外光谱数据建模结果表明, 普通扁形茶在冷冻保存3个月后, 理化性质发生了较大的变异, 而西湖龙井的理化性质相对较为稳定。 从光谱数据主成分特征的数学原理角度对不同年份以及不同保存期模型的适应性进行了研究, 建立并验证了适合西湖龙井真伪识别的模型, 有效提高了模型的预测准确度。 不仅可为近红外光谱应用于农产品质量安全与品质分级方面提供一定的参考, 而且对提高农产品近红外分级模型的预测准确度亦具有参考价值。

In the present paper, the micro-NIR spectrometer with the splitter of linear variable filter was used to develop the recognition models of the West Lake Longjing tea and the ordinary flat tea of the year 2012 and 2013. The NIR spectral data of different years and different storage times were decomposed by PCA algorithm. The PLS-DA models were developed by the representative samples selected by the mathematical characteristics of PCA-scores’ distribution in order to analyze the reason for the inadaptability of the models according to mathematical principles and find out the solution for its correction. Being examined by the external validation set, the adaptability of the authenticity identification model was enhanced effectively. The result of this research indicated that, for the West Lake Longjing tea and the ordinary flat tea, the correct recognition rate of the model developed by all different-year samples’ NIR spectral data would be enhanced effectively. The model developed by the NIR spectral data of different storage time samples indicated that the physicochemical properties of the ordinary flat tea have changed remarkably after cryopreservation for 3 months, while the physicochemical properties of the West Lake Longjing tea are relatively stable. The model adaptabilities for different years and different storage times were studied according to the mathematical perspective of the principal component characteristics of spectral data. After the authenticity identification model of West Lake Longjing tea was developed, the prediction accuracy was enhanced effectively. This research would provide reference for not only the application of NIR spectroscopy in quality grading and safety of agricultural products, but also the enhancement of the prediction accuracy of the NIR grading models for agricultural products.