中文摘要：本文采用太赫兹（THz）光谱技术结合极端学习机（ELM）对转基因大豆油进行快速无损检测，且无需任何样品预处理。为了在识别精度、泛化能力和训练复杂度之间达到最佳平衡，本文采用一种交叉验证方法对惩罚系数C和隐节点L的参数组合进行优化。在选定的最佳参数对（C＝0.12，L＝70）下，转基因和非转基因大豆油样品在训练集和预测集上THZ ELM模型的分类精度可以同时超过89%。研究结果表明，THz光谱技术有望成为未来PCR、ELISA等方法的优良替代品，并为检测非转基因和转基因大豆油提供了一种新的综合工具。
外文摘要：In this paper terahertz (THz) spectroscopy methodology combined with extreme learning machine (ELM) was developed for the fast and nondestructive identification of transgenic soybean oils without any sample pretreatment. In order to achieve the optimum balance among the recognition accuracy, generalization ability and training complexity, leave-one-out cross validation was employed to optimize the parameter combination of penalty coefficient C and the number of hidden nodes L. With the selected optimal parameter pairs, (C = 0.12, L = 70), the classification accuracy of THz-ELM model for the transgenic and non-transgenic soybean oil samples in training set and predicting set can exceed 89% simultaneously. The results demonstrate that THz spectroscopy could be in the future an excellent replacement of the antecedent procedures, such as PCR and ELISA, and supply a new comprehensive tool to distinguish non-transgenic and transgenic soybean oils.
外文关键词：Terahertz (THz) Spectroscopy, Rapid Detection, Non-Transgenic and Transgenic Soybean Oils
作者： Zhou, Sheng-Ling; Zhu, Shi-Ping; Liu, Guang-Hao; Zhu, Jie
作者单位：Southwest Univ
期刊名称：JOURNAL OF NANOELECTRONICS AND OPTOELECTRONICS
期刊影响因子：1.069
出版年份：2018
出版刊次：1