摘要
在苹果酒发酵过程中，由于糖度变化幅度很大且酒体基质始终处于动态变化之中，文章采用分阶段处理结合人工神经网络法对不同发酵阶段的糖度检测、监测近红外光谱模型的建立进行了探讨。不同建模方法的比较结果表明采用减去一条直线法预处理光谱，阶段Ⅰ建模光谱范围选择75026472．1cm^-1阶段Ⅱ建模光谱范围选择6102-5446．2cm。时，阶段Ⅰ模型的R^2为98．93％，RMSECV为4．42g·L^-1；阶段Ⅱ模型的彤为99．34％，RMSECV为1．21g·L^-1；进一步对模型进行验证和评价，结果表明阶段Ⅰ模型验证集的RMSEP为4．07g·L^-1；阶段Ⅱ模型验证集的RMSEP为1．13g·L^-1本研究结果表明利用近红外光谱法建立的模型具有良好的预测效果，能满足苹果酒工业生产中检测、监测精度要求。
The sugar content and the matrix always are being changed during cider-making fermentation. In order to measure and monitor sugar content accurately and rapidly, it is necessary for the spectra to be sorted. Calibration models were established at different fermentation stages based on near infrared spectroscopy with artificial neural network. NIR spectral data were collected in the spectral region of 12 000-4 000 cm^-1 for the next analysis. After the different conditions for modeling sugar content were analyzed and discussed, the results indicated that the calibration models developed by the spectral data pretreatment of straight line subtraction（SLS） in the characteristic absorption spectra ranges of 7 502-6 472.1 cm^-1 at stage Ⅰ and 6 102-5 446. 2 cm^-1 at stage Ⅱ were the best for sugar content. The result of comparison of different data pretreatment methods for establishing calibration model showed that the correlation coefficients of the models （R2） for stage Ⅰ and Ⅱ were 98. 93% and 99.34% respectively, and ＂the root mean square errors of cross validation（RMSECV） for stage Ⅰ and Ⅱ were 4. 42 and 1.21 g · L^-1 respectively. Then the models were tested and the results showed that the root mean square error of prediction （RMSEP） was 4.07 g· L^-1 and 1.13 g · L^-1 respectively. These demonstrated that the models the authors established are very well and can be applied to quick determination and monitoring of sugar content during cider-making fermentation.
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2009年第3期652-655,共4页
Spectroscopy and Spectral Analysis
基金
国家食品安全关键技术重大项目(2006BAK02A24)
国家自然科学基金项目(20806062)
西北农林科技大学青年学术骨干支持计划项目资助
关键词
近红外光谱
苹果酒
糖度
人工神经网络
偏最小二乘法
Near infrared spectroscopy
Cider
Sugar content
Artificial neural network
Partial least square
