传统食品掺假分析多集中于检测特定已知或者怀疑可能存在的掺假物, 然而由于掺假形式的多样性以及新的掺假物不断出现, 使得传统检测方法具有局限性。 目前, 全蛋粉作为鲜蛋理想替代品掺假现象十分严重, 然而不管是国内还是国外, 其掺假检测都鲜有研究。 因此, 为了探索一种快速检测全蛋粉掺假的方法, 研究尝试使用最近快速发展起来的具有绿色、 无损等优点的高光谱技术来检测全蛋粉掺假的可行性。 从不同地区收集不同品牌的鸡蛋全蛋粉, 按不同比例分别掺入淀粉、 大豆分离蛋白、 麦芽糊精以及三种掺假物的混合物进行试验样品的制备。 样品进行光谱采集后, 采用ENVI软件选取感兴趣区域(ROI)后提取出平均光谱。 根据获得的光谱数据建立全波段下支持向量机(SVM)模型进行掺假的判别并采用偏最小二乘回归(PLSR)模型建立全波段与掺假浓度之间的关系。 结果显示, 采用径向基核函数所建立的SVM模型, 其分类的正确率达到90%以上, 基于PLSR建立掺假模型实际值与预测值相关系数R2P均高于0.90。 为了简化模型, 采用回归系数法(RC)及连续投影法(SPA)提取特征波长, 根据特征波长下的光谱数据建立RC-PLSR和SPA-PLSR模型, 结果显示, 经简化的模型依然具有良好的性能, 说明使用高光谱技术来检测全蛋粉掺假是可行且高效的。

Traditional analysis of food adulteration is more concentrated in the detection of specific known or suspected adulterants which may exist. However, due to the variety of adulteration and the emergence of new adulterants, the traditional detection methods have limitations. Currently, as an ideal substitute for fresh egg, the adulteration of egg powder is serious, but the problem is rarely studied both at home and abroad. In order to explore a rapid detection method of whole egg powder adulteration, this study attempted to use hyperspectral technology green and nondestructive in its advantages to detect the feasibility of whole egg powder with several adulterants. Different brands of egg powder were collected from different area and the common adulterants (starch, soy isolate protein, maltodextrin and mixture) were added in in proportion. After spectral acquisition, the region of interest (ROI) was extracted by ENVI and the mean spectra were extracted. Firstly, the support vector machines (SVM) models were founded to identify the adulteration and the Partial least squares (PLSR) model was used to establish the relationship between the full bands and adulteration concentration. The results showed that the correctness of the SVM model based on RBF kernel function was more than 90%, and the correlation coefficient between the actual value and the prediction value of the adulteration model based on PLSR was higher than 0.90. In order to simplify the model, the regression coefficient method (RC) and the successive projections algorithm (SPA) were used to extract the characteristic wavelengths, and the RC-PLSR model and SPA-PLSR were established according to the spectral data at the characteristic wavelength. The results showed that the simplified models still have good performance, indicating that the hyperspectral technique to detect adulteration of whole egg powder is feasible.