基于近红外光谱技术的米糠粕主要成分分析  

Analysis on nutrients of rice bran meal based on near infrared spectroscopy

作　　者：姜红[1,2] 赵志行 李小定[1,2] 江叔俭 

机构地区：[1]华中农业大学食品科学技术学院,湖北武汉430070 [2]环境食品学教育部重点实验室,湖北武汉430070 [3]武汉度微生物科技有限公司,湖北武汉430070

出　　处：《食品工业科技》2015年第16期86-90,172,共6页Science and Technology of Food Industry

基　　金：国家科技支撑计划项目(2013BAD20B06)

摘　　要：为探索米糠粕营养成分的近红外快速测定方法，采集261个米糠粕样品的近红外光谱，分别经过标准正态变量变换、去趋势校正、多元散射校正等20种方法进行预处理，在1000—1799nm波长范围内，结合化学方法测定数据采用偏最小二乘法、主成分分析结合人工神经网络法、偏最小二乘结合人工神经网络法建立米糠粕营养成分近红外定量模型。结果发现，在3种建模方法中，偏最小二乘法结合人工神经网络法建立的模型效果最好，预测精度最高，所得的水分、灰分和粗蛋白近红外定量模型的相关系数分别为0．9593、0．9168和0．96261.To explore fast analysis method on nutrients of rice bran meal by near infrared spectroscopy,the near infrared spectra of 261 rice bran samples was collected. The spectra was performed with 20 different preprocessing methods,such as standard normal variate,detrend and multiplicative scatter correction. The near infared spectroscopy quantitative models of rice bran meal nutrients in 1000~1799 nm were respectively established by partial least square,principal component analysis combined with artificial neural network and partial least square combined with artificial neural network with the data measured by chemical method. The results showed that the models obtained by partial least square combined with artificial neural network were the best and its prediction accuracy was the highest. The correlation coefficients of the models were 0.9593,0.9168 and 0.9626 for moisture, ash content and crude protein, respectively.

关 键 词：米糠粕 近红外光谱技术 偏最小二乘法 主成分分析 人工神经网络 

分 类 号：TS210.9[轻工技术与工程—粮食、油脂及植物蛋白工程]