采用傅里叶变换红外光谱技术结合多元统计分析建立快速鉴别不同年份、 不同产地美味牛肝菌的方法。 采集2011年—2014年云南26个不同地区152个美味牛肝菌样品的红外光谱, 使用正交信号校正(orthogonal signal correction, OSC)、 微波压缩(wavelet compression)方法对原始光谱进行优化处理, OSCW校正前后的光谱数据进行偏最小二乘判别分析(partial least squares discriminant analysis, PLS-DA), 比较光谱预处理前后PLS-DA的分类效果。 将152个美味牛肝菌随机分为训练集(120个)和验证集(32个), 建立OSCW校正前后的PLS分类预测模型。 结果显示, 经OSCW处理后的PLS-DA分类效果明显优于处理前的结果, 主成分得分图能准确区分不同年份、 不同产地美味牛肝菌样品, 表明OSCW处理能有效滤除光谱中的噪音及与因变量无关的干扰信息, 提高光谱分析的准确性和计算速率。 OSCW处理前PLS模型训练集的R2和RMSEE分别为0.790 1和21.246 5, 验证集的R2和RMSEP分别为0.922 5和14.429 2; OSCW预处理后训练集的R2和RMSEE分别为0.852 3和17.238 1, 验证集的R2和RMSEP分别为0.845 4和20.87, 表明OSCW预处理提高了训练集的预测效果, 但OSCW-PLS出现了过拟合现象降低验证集的预测能力, 因此, OSCW不适宜与PLS结合建立模型。 OSCW结合PLS-DA能滤除光谱中大量的干扰信息, 准确区分不同年份、 不同产地美味牛肝菌样品, 为野生食用菌的鉴别分类提供可靠依据。

In order to establish a rapid method for discriminating Boletus edulis mushroom, Fourier transform infrared spectroscopy combined with multivariate statistical analysis were used to study B. edulis which were collected from different origins and different years. The original infrared spectra of all the 152 B. edulis samples collected from 2011 to 2014 and 26 different areas of Yunnan Province were optimized with orthogonal signal correction and wavelet compression (OSCW) method. The spectral data that before and after being preprocessed with OSCW were analyzed with partial least squares discriminant analysis (PLS-DA). The classification results of PLS-DA were compared. Then the 152 B. edulis samples were randomly divided into a training set (120) and a validation set (32) to establish the PLS classification prediction model. The results showed that, after OSCW processing, the classification result of PLS-DA was significantly better than the other one which was not processed by OSCW. Principal component score plot can accurately distinguish B. edulis samples collected from different years and different origins. It indicated that OSCW can effectively eliminate the noise of spectra and reduce the unrelated interference information about the dependent variables to improve the accuracy and calculation speed of spectral analysis. Before OSCW preprocessed, the R2 and RMSEE of PLS model of the training set were 0.790 1 and 21.246 5 respectively while R2 and RMSEP of the model of validation set were 0.922 5 and 14.429 2. After OSCW pretreatment, R2 and RMSEE of the training set were 0.852 3 and 17.238 1 while R2 and RMSEP of validation set were 0.845 4 and 20.87. It suggested that OSCW could improve the predictive effect of the training set, but the over-fitting of OSCW-PLS may reduce the predictive ability of validation set. Therefore, it was unsuitable to establish a model with OSCW combined with PLS. In a conclusion, OSCW combined with PLS-DA can eliminate a large amount of spectrum interference information. This method could accurately distinguish B. edulis samples collected from different years and different origins. It could provide a reliable basis for the discrimination and classification of wild edible fungi.