中文摘要：

将主成分分析方法用于近红外定量分析校正集样本的优选，提出一种根据样本光谱第一主成分得分优选校正集样本的方法，并使用本方法从418个烟草样本中优选得到约105个样本的校正集。通过对烟碱成分实际建模与外部验证，与随机法、含量梯度法两种校正集样本挑选方法的性能进行了对比。结果表明，本方法既克服了随机法挑选样本代表性不足的风险，又可避免含量梯度法必须测定所有样本成分含量而造成的人力物力消耗，具有无需编程、操作简单、易于推广的特点。

英文摘要：

Principal component analysis technique was applied to calibrating sample selection of near-infrared quantitative model. A simple and reliable method in selection of samples for calibration was proposed, which choose calibration samples according to scores of the first principal component of near-infrared spectrum. Calibration sets contained about 105 samples were selected from 418 tobacco samples using the method proposed and the other two methods named Random method and Content Grads method. Quantitative models of nicotine were established and tested respectively to compare the performances of these three selection methods. Results show that the method bring forward has good ability to choose calibration samples using spectrum data only. It can avoid the risk of lacking representativeness caused by Random method as well as the large workload to determine concentrations of all the samples using Content Grads method. In addition, the method is simple and easy to be used.