为建立薰衣草精油品种品质的快速辨别分析模型, 采用衰减全反射红外光谱法测定三个品种共96个薰衣草精油样品, 对原始光谱数据求二阶导数, 通过方差计算, 确定1　750～900 cm-1波长段为判别分析用数据。 分析结果表明, 主成分分析(PCA)基本能实现精油品种区分, 前三个主成分主要代表着酯、 醇和萜类物质。 使用68个样品的校正集建立正交偏最小二乘判别分析（OPLS-DA）模型, 三个品种薰衣草精油的回归曲线测定系数分别为0.959　2, 0.976　4, 0.958　8, 验证集中三个品种精油预测均方根误差（RMSEP）分别为0.142　9, 0.127　3, 0.124　9, OPLS-DA法建立的模型对校正集和验证集的判别率和预测率都达到100%, 模型对薰衣草精油品种品质有很好的识别能力。 为薰衣草精油品种品质提供一个快速、 直观的方法。

This work aimed to use attenuated total reflectance Fourier transform infrared spectroscopy to identify the lavender essential oil by establishing a Lavender variety and quality analysis model. So, 96 samples were tested. For all samples, the raw spectra were pretreated as second derivative, and to determine the 1 750~900 cm-1 wavelengths for pattern recognition analysis on the basis of the variance calculation. The results showed that principal component analysis (PCA) can basically discriminate lavender oil cultivar and the first three principal components mainly represent the ester, alcohol and terpenoid substances. When the orthogonal partial least-squares discriminant analysis (OPLS-DA) model was established, the 68 samples were used for the calibration set. Determination coefficients of OPLS-DA regression curve were 0.959 2, 0.976 4, and 0.958 8 respectively for three varieties of lavender essential oil. Three varieties of essential oil's the root mean square error of prediction (RMSEP) in validation set were 0.142 9, 0.127 3, and 0.124 9, respectively. The discriminant rate of calibration set and the prediction rate of validation set had reached 100%. The model has the very good recognition capability to detect the variety and quality of lavender essential oil. The result indicated that a model which provides a quick, intuitive and feasible method had been built to discriminate lavender oils.