摘要
采用近红外光谱的离散余弦变换和BP神经网络相结合,建立了白酒近红外光谱与其酒精度之间的数学关系模型,为快速检测白酒的酒精度提供了一种新的方法。结果表明,经离散余弦变换后建立的模型比全光谱偏最小二乘建立的模型具有更精确的预测效果。模型的相关系数由原来的0．9611提高到0．9744,预测标准偏差由原来的1．3891降低到0．9542。
A mathematical relation model between liquor near infrared spectrum and liquor alcoholicity was established by use of the combination of discrete cosine transformation （DCT） and BP neural networks, which was a new approach for rapid liquor alcoholicity determination. The results suggested that the predicted standard deviation was 0.9542 and the correlative coefficient was 0.9744 by this method, meanwhile, 1.3891 and 0.9611 respectively by PLS. It proved that this mathematical model could be used in practice. （Tran. by YUE Yang）
出处
《酿酒科技》
北大核心
2006年第3期52-54,共3页
Liquor-Making Science & Technology
基金
2004年度江苏省博士后资助项目
关键词
白酒
定量分析
离散余弦变换
神经网络
近红外光谱
liquor
quantitative analysis
discrete cosine transformation
BP neural network
near infrared spectrum
