近年来食品安全问题频发, 消费者愈加重视食品原产地的环境安全, 导致地理标志产品的需求增加。 美味牛肝菌(Boletus edulis)作为一种健康食品, 其产品品质受原产地环境影响较大, 为保护消费者的身体健康, 防止假冒伪劣产品进入市场, 急需一种高效、 廉价的美味牛肝菌产地鉴别技术。 采用数据融合策略结合偏最小二乘判别(PLS-DA)模型对美味牛肝菌的产地进行鉴别。 扫描来自8个产地(昆明、 楚雄、 玉溪、 迪庆、 大理、 保山、 文山和曲靖)141个样品的傅里叶变换近红外光谱和傅里叶变换中红外光谱。 使用Kennard-Stone算法将所有样品划分为2/3的训练集和1/3的预测集, 利用三种融合策略(低级、 中级和高级)对4个单一光谱矩阵: 近红外的菌柄(N-b)、 近红外的菌盖(N-g)、 中红外的菌柄(M-b)、 中红外的菌盖(M-g), 建立偏最小二乘判别(PLS-DA)模型。 用交叉验证均方根误差(RMSECV)和预测均方根误差(RMSEP)评价模型稳定性, 非错误率(NER)、 训练集正确率和预测集正确率评价模型分类性能, 综合多种评价指标, 找出美味牛肝菌产地鉴别的最佳方法。 结果表明: (1)近红外和中红外光谱均能鉴别美味牛肝菌产地; (2)中红外光谱所建立的模型优于近红外光谱所建立的模型; (3)三种融合策略均可提高美味牛肝菌的产地鉴别效果, 产地鉴别效果优劣依次为中级融合、 高级融合、 低级融合、 单一光谱模型。 通过融合近红外和中红外光谱使用PLS-DA进行基于特征值LV的中级融合策略, 建立不同产地美味牛肝菌鉴别模型, 有最少的变量数(49), 最高的产地训练集正确率(100%), 最高的产地预测集正确率(100%), 最低的RMSEP(0.133), 实现了美味牛肝菌产地的快速、 准确鉴别, 可以作为美味牛肝菌产地溯源的一种可靠方法。

In recent years, food safety problems happened frequently, and consumers pay more and more attention to the environmental safety of food origin, which leads to an increase in demand for geographical indication products. As a healthy food, the quality of Boletus edulis is greatly affected by the environment of its origin. In order to protect consumers’ health and prevent fake and inferior products from entering the market, it is urgent to develop an efficient and low-cost identification technology of the origin of delicious Boletus edulis. Data fusion strategy and partial least squares discrimination (PLS-DA) model were used to identify the origin of Boletus edulis. In this paper, Fourier transform near infrared and Fourier transform middle infrared spectra of 141 samples from 8 Origin (Kunming, Chuxiong, Yuxi, Diqing, Dali, Baoshan, Wenshan and Qujing) were scanned. Kennard-stone algorithm was used to divide all samples into 2/3 training set and 1/3 prediction set. Three fusion strategies (low-level, mid-level, high-level) were used to analyze four single spectral matrices spectra: near-infrared average spectra of stipes (N-b), near-infrared average spectra of caps (N-g), mid-infrared average spectra of stipes (M-b), mid-infrared average spectra of caps (M-g) and to establish a partial least squares discriminant (PLS-DA) model. In which root mean square error of cross validation (RMSECV) and the root mean square prediction error (RMSEP) are used to evaluate model stability. The purpose of the non-error ratio (NER), training set classification accuracy and forecast set classification accuracy evaluation model classification performance. It contributes to find out the best way to geographic origin identification of Boletus edulis. The results showed that: (1) near infrared and middle infrared spectra can identify the origin of Boletus edulis; (2) the model established by middle infrared spectrum is better than that in near infrared spectrum; (3) all the three fusion strategies can improve the identification effect of origin of Boletus edulis, and the identification results of producing area from good to bad are in order of mid fusion, high fusion, low fusion and single spectral model. By using PLS-DA intermediate fusion strategy to fuse in near infrared and Mid-infrared spectrum, different origin Boletus edulis identification models are established, with the least number of variables (49), the highest accuracy of training set in producing area (100%), the highest accuracy of prediction set of origin (100%), the lowest RMSEP (0.133). As a reliable method, it can identify the geographical origin of Boletus edulis fast and accurately.