采用近红外光谱法结合偏最小二乘法（PLS），分别建立了药物复合维生素B片和保健品B族维生素片中维生素B₆的定量分析模型。取药物复合维生素B片或保健品B族维生素片6~12片，研磨，以转速4 000 r·min^-1振荡1 min，将样品粉末分别装入3个样品杯中，采用积分球漫反射分析模块采集其近红外谱图，每个样品采集5次，得到15张近红外光谱图。以高效液相色谱法为参比方法，以不同类型维生素B片为建模的训练集样品，每个样品中，任取12张光谱图作为训练集样本，余下3张光谱图作为内部验证集样本，采用TQ Analyst 9.0软件中的PLS，用一阶导数光谱数据形式，选择3个波段（药物复合维生素B片4 700.46~5 125.86 cm^-1，5 444.60~5 708.26 cm^-1，5 754.69~6 159.52 cm^-1；保健品B族维生素B片4 133.67~5 512.28 cm^-1，5 944.17~6 495.08 cm^-1，8 105.85~9 005.94 cm^-1）同时进行建模，使用诺里斯导数滤噪，节段长度为5，段间间隙为5（药物复合维生素B片）或6（保健品B族维生素B片）。结果显示：药物复合维生素B片定量分析模型（模型2）的校正均值方差（RMSEC）为0.003 0，校正相关系数（R_c）为0.994 7，预测均值方差（RMSEP）为0.002 9，预测相关系数（R_p）为0.995 4，交叉验证均值方差（RMSECV）为0.005 2，交叉验证相关系数（R_cv）为0.994 1，性能指数为92.8；保健品B族维生素B片定量分析模型（模型3）的RMSEC为0.031 4，R_c为0.998 7，RMSEP为0.035 8，R_p为0.998 7，RMSECV为0.038 8，R_cv为0.998 1，性能指数为93.5。利用模型2预测药物复合维生素B片中维生素B₆的含量，预测值的绝对误差为-0.004 4%~0.008 0%，相对误差为-1.7%~3.3%；利用模型3预测保健品B族维生素片中维生素B₆的含量，预测值的绝对误差为-0.007 1%~0.049 7%，相对误差为-1.9%~5.0%。

The models for quantitative analysis of vitamin B₆ in compound vitamin B tablets for drug and vitamin B complex tablets for health products was established respectively by near infrared spectrometry combined with partial least square (PLS) method. 6-12 tablets of compound vitamin B tablets or vitamin B complex tablets were ground and oscillated at 4 000 r·min^-1 for 1 min, and the sample powder was respectively put into 3 sample cups. The near infrared spectra were collected by integrating sphere diffuse reflection analysis module. Each sample was collected 5 times, and 15 near infrared spectra were obtained. High performance liquid chromatography was used as the reference method, and different types of vitamin B tablets were used as modeling samples of the training set. In each sample, 12 spectra were used as the training set samples, and the remaining 3 spectra were used as the internal verification set samples. PLS method in TQ Analyst 9.0 was adopted, and the spectral data of the first derivative was used. 3 band ranges were selected for modeling (ranges of compound vitamin B tablets of 4 700.46-5 125.86 cm^-1, 5 444.60-5 708.26 cm^-1, 5 754.69-6 159.52 cm^-1; ranges of vitamin B complex tablets of 4 133.67-5 512.28 cm^-1, 5 944.17-6 495.08 cm^-1, 8 105.85-9 005.94 cm^-1). Norris derivative noise filter was used. The length of the segment was 5, and the interval between segments was 5 (compound vitamin B tablets) or 6 (vitamin B complex tablets). As shown by the results, quantitative analysis model of compound vitamin B tablets (model 2) of the corrected mean variance (RMSEC) was 0.003 0, with the corrected correlation coefficient (R_c) of 0.994 7, the predicted mean variance (RMSEP) of 0.002 9, the predicted correlation coefficient (R_p) of 0.995 4, the cross validation mean variance (RMSECV) of 0.005 2, the cross validation correlation coefficient (R_cv) of 0.994 1 and the performance index of 92.8. Quantitative analysis model of vitamin B complex tablets (model 3) of RMSEC was 0.031 4, with R_c of 0.998 7, RMSEP of 0.035 8, R_p of 0.998 7, RMSECV of 0.038 8, R_cv of 0.998 1, and the performance index of 93.5. Model 2 was used to predict the content of vitamin B₆ in compound vitamin B samples, and the absolute error and the relative error of prediction values were in the ranges of -0.004 4%-0.008 0% and -1.7%-3.3%, respectively. Model 3 was used to predict the content of vitamin B₆ in the vitamin B complex samples, and the absolute error and the relative error of prediction values were in the ranges of -0.007 1%-0.049 7% and -1.9%-5.0%, respectively.