基于太赫兹光谱及1DCNN-BiLSTM的黄蜀葵花金丝桃苷含量预测  

Prediction of Hyperoside Content in Abelmoschus Manihot Based on Terahertz Spectroscopy and 1DCNN-BiLSTM

作　　者：叶华清 郑成勇[1] YE Hua-qing;ZHENG Cheng-yong(School of Mathematics and Computer Science,Wuyi University,Jiangmen 529020,China)

机构地区：[1]五邑大学数学与计算科学学院,广东江门529020

出　　处：《五邑大学学报（自然科学版）》2024年第2期48-54,共7页Journal of Wuyi University(Natural Science Edition)

摘　　要：太赫兹(THz)光谱具有信噪比高、光子能量低、穿透性强、安全和快速等优点,已广泛应用于食药检测.现有基于HPLC等方法的黄蜀葵花金丝桃苷含量检测耗时长、操作复杂.提出一种基于太赫兹光谱及1DCNN-BiLSTM的黄蜀葵花金丝桃苷含量预测方法:首先采集黄蜀葵花的太赫兹时域谱,并通过傅里叶等变换获取其7种光谱数据;然后利用主成分分析(PCA)对7种光谱数据降维,以获得维数一致的7元样本数据;接着将降维对齐后的7元样本数据输入设计好的1DCNN-BiLSTM网络,以获得金丝桃苷含量预测.与1DCNN、BiLSTM的对比实验结果表明,1DCNN-BiLSTM网络具有较高的预测精度,10次随机实验的平均决定系数达0.970 5.Terahertz(THz)spectroscopy has the advantages of high signal-to-noise ratio,low photon energy,strong penetration,and being safe and quick,and has been widely used in food and drug detection.The existing methods based on high performance liquid chromatography(HPLC)and other methods for detecting the content of hyperoside in Abelmoschus manihot are time-consuming and complex to operate.A prediction method of hyperoside content in Abelmoschus manihot based on THz spectroscopy and 1DCNN-BiLSTM was proposed.Firstly,the THz time-domain spectral data of Abelmoschus manihot were collected,and seven kinds of spectral data were obtained through Fourier transform.Then,principal component analysis(PCA)was used to reduce the dimensionality of 7 spectral data to obtain dimensionally aligned 7-variable sample data.Next,the aligned 7-variable sample data after dimensionality reduction are input into the designed 1DCNN-BiLSTM network to obtain the prediction of hyperoside content.Compared with 1DCNN and BiLSTM,the experimental results show that 1DCNN-BiLSTM network has higher prediction accuracy, and the average coefficient of determination of 10 random experiments is 0.970 5 .

关 键 词：太赫兹光谱 一维卷积神经网络 双向长短时记忆网络 黄蜀葵花 金丝桃苷 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]