近红外光谱波段优化选择在驴奶成分分析中的应用

近年来,驴奶引起了越来越多研究者的注意.与牛奶相比,驴奶的营养成分更接近母乳,且有着许多独特的优势.由于驴奶与牛奶成分差别较大,适用于牛奶的模型无法直接应用于驴奶的成分分析中.但目前还未见将近红外光谱分析技术应用到驴奶成分分析中的研究报道.文章采用傅里叶变换近红外光谱法,快速测定了新疆疆岳驴奶中脂肪、蛋白质、能量和灰分的含量.其含量分布范围分别为:1.15%～2.54%,0.34%～2.67%,355.87～565.17 cal·kg-1,0.28%～0.57%.光谱扫描区为3 899.6～12 493.4 cm-1,扫描间隔1 cm-1.采用PLS回归算法对光谱信息阵X提取主成分时附加约束,使X的主成分与待分析组分Y相关.并利用优化波段和不同预处理方法优化组合,建立了PLS回归预测模型,并与PLS全谱区建模预测进行了比较.结果表明波段优化组合建模分析效果整体优于全谱区建模结果,驴奶样品中脂肪、蛋白质、能量和灰分的近红外光谱定量分析模型预测值与化学分析实测值在水平α=0.05下显著相关,其含量分析模型的校验集(RMSEP)值分别为0.18,0.117,23.5,0.040 6,表明预测值有较好的精度.结果表明建立近红外光谱定量分析模型用于驴奶样品成分分析是可行的,波段优化选择与全谱建模分析效果比较表明,建立定量分析模型对波段优化组合选择是必要的,当模型中包含了与组分无关的信息时,对模型定量分析将起干扰的作用,会影响模型的分析效果.因此进行谱段信息选择建立相应组分分析模型是数据预处理的有效环节.样品各组分标准值的测定结果的准确度和精确度都影响近红外定量分析的准确度.以近红外光谱法进行定量建模分析时,扩大组分含量的分布范围,提高标准数据的分析精度和准确度都是很必要的.

Optimizing Spectral Region in Using Near-Infrared Spectroscopy for Donkey Milk Analysis

Donkey milk has aroused more attention in recent years since its nutrition composition shows a higher similarity to human milk than others. Due to the composition difference between cow milk and donkey milk, the present models available for cow milk analysis could not be applied to donkey milk without modifications. A rapid and reliable analysis method is required to measure the nutrition composition of donkey milk. Near infrared spectroscopy is a newly developed method in food industry, but no literature report was found regarding to its application in the analysis of donkey milk. Protein, fat, ash contents and energy value are the major nutrition factors of milk. In the present paper, these factors of donkey milk were investigated by Fourier transform near-infrared (FT-NIR) spectroscopy. The ranges of protein, fat and ash contents, and energy value in donkey milk samples were 1.15%-2.54%, 0.34%-2.67%, 0.28%-0.57% and 355.87-565.17 cal x kg(-1), respectively. The IR spectra ranged f from 3 899.6 to 12 493.4 cm(-1), with a 1 cm(-1) scanning interval. When the principal least square (PLS) regression algorithm is used for spectral regions information extraction, the additional constraint makes the principal components of matrix X to be related with the components of Y which is to be analyzed. Various spectral regions and data pretreatment methods were selected for principal least square (PLS) regression model development. A comparison of the whole and optimized spectral region NIR indicated that the models of selecting optimum spectral region were better than those of the whole spectral region. It was shown that the protein, fat and ash contents, and energy value in donkey milk obtained by chemical methods were well correlated to the respective values predicted by the NIR spectroscopy quantitative analysis model (alpha = 0.05). The RMSEP values were 0.18, 0.117, 0.040 6 and 23.5 respectively, indicating that these predicted values were reliable. These results suggested that FT-NIR spectroscopy could be used for the rapid detection of the composition of donkey milk by establishing NIR spectroscopy quantitative analysis models. Selecting an optimum spectral region and establishing a special NIR analysis model accordingly are key steps during the data pretreatment. The models of the optimum spectral region were better than the models of the whole spectral region. When irrelevant information was included in the models, it would interfere with the analysis and give less reliable results. Therefore, the selection of a right spectra region plays an important role in the set-up of quantitative analysis models. The accuracy and reliability of the standard data used in model settings are also critical to the final results. In order to improve the reliability and accuracy of the NIR methods, a wide range of component contents and more accurate standard data are definitely required.