基于电子鼻的玉米霉变程度鉴别及玉米赤霉烯酮和黄-中国粮油学报.DOCVIP

电子鼻融合BP神经网络预测玉米赤霉烯酮和黄曲霉毒素B1含量模型研究 于慧春 彭盼盼 殷勇 刘云宏 （河南科技大学食品与生物工程学院，河南 洛阳 471023） 摘 要 为探究玉米赤霉烯酮和黄曲霉毒素B1的无损快速定量测定方法，用电子鼻对7级不同霉变程度玉米样品进行检测，并用理化分析方法分别测定霉变玉米中的玉米赤霉烯酮与黄曲霉毒素B1含量；在提取电子鼻响应信号的积分值作为特征参量的前提下，采用BP神经网络建立不同霉变程度下玉米样品中的玉米赤霉烯酮与黄曲霉毒素B1含量的预测模型。同时，为了获得较为可靠的BP神经网络预测模型，在神经网络结构不变的条件下，对比分析了不同训练集、测试集构建的预测模型。结果发现在各预测模型的70组测试样本中，相对误差控制在5%以内的样本数量都在60个以上，最大相对误差控制在15%以内，从而证明了BP神经网络预测模型的有效性、可靠性。该研究为实施玉米霉变毒素的快速无损检测提供了一种途径。 关键词 电子鼻 玉米 玉米赤霉烯酮 黄曲霉毒素B1 BP神经网络 霉变 预测模型 中图分(类号：TP212 文献标识码：A 文章编号： Coupled electronic nose and BP neural network to study on the predicting model of zearalenone and aflatoxin B1 YU Huichun PENG Panpan YIN Yong (College of Food and Bioengineering, Henan University of Science and Technology, Luoyang 471023) Abstract In order to explore the fast, quantitative and nondestructive test method for zearalenone and aflatoxin B1, corn samples with 7 different levels of mold were tested by electronic nose (e-nose), at the same time the content of zearalenone and aflatoxin B1 were tested using biochemical analysis method. The integral value of the e-nose response signal was extracted and acted as the characteristic parameter, BP neural network was adopted to establish prediction model for the content of zearalenone and aflatoxin B1 of different degree of mildew corn samples. In addition, in order to obtain a more reliable BP neural network prediction model, under the premise that the structure of the neural network was unchanged, the prediction model based on different training sets and test sets was compared and analyzed. The results show that in each prediction model of 70 groups of test samples, the relative error control within 5% of the sample quantity are over 60, maximum relative error was controlled within 15%, it proved the validity and reliability of the BP neural network prediction model, the study provides a method of fast nondestructive testing corn mycotoxin. Key words: electronic nose, corn, zearalenone, aflatoxin B1, BP neural network,