光谱学与光谱分析, 2017, 37 (7): 2199, 网络出版: 2017-08-30  

基于激光诱导击穿光谱技术的咖啡豆中咖啡因含量快速检测方法

Determination of Caffeine Content in Coffee Beans Based on Laser Induced Breakdown Spectroscopy
作者单位
浙江大学生物系统工程与食品科学学院, 浙江 杭州 310058
摘要
应用激光诱导击穿光谱(LIBS)技术研究了快速检测咖啡豆中咖啡因含量的可行性。 将咖啡豆磨粉压成片状作为采集LIBS光谱数据的样本, 应用原子吸收分光光度计测量每个样本中咖啡因的含量。 应用基线校正, 小波变换和归一化等数据预处理方法; 针对基于全部变量的偏最小二乘(PLS)模型会出现过拟合, 分别应用回归系数和主成分分析(PCA)选择特征变量, 并建立了基于特征变量的PLS和BP神经网络模型。 结果表明: 基于回归系数所选特征变量的PLS模型中, 建模集相关系数Rc=096, 预测集Rp=091; 基于PCA提取特征变量的PLS模型中, Rc=094, Rp=090; 基于PCA所选特征变量的BP神经网络模型中, Rc=096, Rp=096。 两种方法所提取特征变量均对应C, H, O, N, Na, Mn, Mg, Ca和Fe, 且基于上述两种方法所选特征变量的PLS模型均对预测集样本有较好的预测结果, 说明上述元素与咖啡因含量存在联系, 应用回归系数和PCA选择的特征变量是有效的, 但是咖啡豆内C, H, O, N, Na, Mn, Mg, Ca, Fe与咖啡因含量的确切关系需要进一步研究。 基于PCA所选特征变量的BP神经网络模型有更优的预测结果, 说明所选特征变量适用于不同的建模方法。 研究表明LIBS技术结合化学计量学方法可以实现咖啡豆中咖啡因含量的快速检测。
Abstract
The feasibility of fast detection of caffeine content in coffee beans based on laser induced breakdown spectroscopy (LIBS) combined with chemometrics methods was studied. The coffee beans were grinded. 05 g of powerd material was transformed into a disk by manual tableting machine. 60 disks of coffee bean material were prepared for LIBS data acquiring. The samples were pretreated by acid wet digestion, and actual caffeine content of each sample was obtained by automic absorption spectrometer (AAS). Baseline correction was applied on the original spectral data to eliminate the negative values. Wavelet transform (WT) was used to reduce the noise, wavelet basis function is Daubechies 5 (db5) and decomposition level is 10. Normalization were employed to deal with variations caused by matrix effects and experimental conditions. The partial least squares (PLS) model on full data appeared to over-fitting. Regression coefficients and principal component analysis (PCA) were used to select characteristic variables, respectively. PLS models and back propagation (BP) neural network model were built by the variables selected. In the PLS model on the variables selected by regression coefficients, correlation coefficient of calibration set (Rc) was 096, correlation coefficient of prediction set (Rp) was 091. In the PLS model on the variables selected by PCA, Rc=094, Rp=090. In the BP neural network model on variables selected by PCA, Rc=096, Rp=096. The characteristic variables selected by two methods correspond to C, H, O, N, Na, Mg, Ca, Fe, Mn. The PLS models on the variables selected by regression coefficients and PCA both performed well on prediction samples. It demonstrated the certain relationship existing between the elements and caffeine content and the selected variables were effective. But the precise relationship bwtween C, H, O, N, Na, Mg, Ca, Fe, Mn and caffeine content needs further study. The BP neural network model on variables selected by PCA performed better than the PLS model, which demonstrated the selected variables were suitable for different modeling methods. The study showed LIBS could be applied to fast determination of caffeine content within coffee bean combined with chemometrics methods. The method of caffeine content detection presented by this study is innovative.

宋坤林, 张初, 彭继宇, 叶蓝韩, 刘飞, 何勇. 基于激光诱导击穿光谱技术的咖啡豆中咖啡因含量快速检测方法[J]. 光谱学与光谱分析, 2017, 37(7): 2199. SONG Kun-lin, ZHANG Chu, PENG Ji-yu, YE Lan-han, LIU Fei, HE Yong. Determination of Caffeine Content in Coffee Beans Based on Laser Induced Breakdown Spectroscopy[J]. Spectroscopy and Spectral Analysis, 2017, 37(7): 2199.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!