Protein content of single kernels of wheat by near-infrared reflectance spectroscopy
Protein content of single kernels of wheat by near-infrared reflectance spectroscopy
复制标题
DOI：
10.1006/jcrs.1997.0165
复制
发表时间：
1998-05-01
影响因子：
3.8
通讯作者：
Delwiche, SR
中科院分区：
文献类型：
--
作者：
Delwiche, SR
众所周知，蛋白质含量会影响加工小麦产品的功能特性。传统上，这些方法对大小从 30-40 g（Eor 燃烧和凯氏定氮分析）到几百克（用于全谷物近红外分析）的样品进行等分试样（0.25-2.2 g）进行，这些方法本质上不提供有关单核蛋白质变异性的信息。美国农业部对小麦分级和分类的检查程序正在发生变化，以便为加工商或最终用户提供有关硬度、水分、重量和小麦等级等多种单粒特性变化的信息。本研究的重点是论证通过近红外反射率测量单个小麦籽粒粗蛋白含量的可行性。选取了 1992 年美国收获的 300 多个商业小麦样品，代表 6 个（硬质小麦除外）市场类别中的 5 个（硬红冬小麦、硬红春小麦、软红冬小麦、硬白小麦和软白小麦），并从中随机选择 10 粒并按单粒进行处理。处理包括反射扫描（1100-2498 nm）、干燥（用于水分补偿）和燃烧（用于参考蛋白质含量测定）。当偏最小二乘法和多元线性回归模型应用于从校准中排除的样品时，表现出蛋白质的标准误差范围为 0.462% 至 0.720%，具体取决于建模技术、用于开发模型的类别数量以及测试的小麦类别。汇集小麦类别来生成通用模型并没有降低模型的准确性。 1100-1400 nm 区域获得了最佳结果。随着波长区域变宽或最小波长从 1100 nm 移至更高值，模型性能会恶化。
Protein content is well known to affect the functional properties of processed wheat products. Traditionally performed on aliquots (0.25-2.2 g) from samples ranging in size from 30-40 g (Eor combustion and Kjeldahl analyses) to several hundred grams (for whole-grain near-infrared analysis), these methods inherently do not provide information on single-kernel protein variability. Inspection procedures by the United States Department of Agriculture for grading and classification of wheat are undergoing change to provide the processor or end user with information on the variability of several single-kernel properties including hardness, moisture, weight, and wheat class. The present study has focused on demonstrating the feasibility of measuring crude protein content of single wheat kernels by near-infrared reflectance. More than 300 commercial wheat samples from the 1992 U.S. harvest, representing five (hard red winter, hard red spring, soft red winter, hard white, and soft white) of the six (durum excluded) market classes were chosen, from which 10 kernels were randomly selected and handled on a single-kernel basis. Handling consisted of reflectance scanning (1100-2498 nm), drying (for moisture compensation), and combustion (for reference protein-content determination). Partial least squares and multiple linear regression models, when applied to samples excluded from calibration, demonstrated standard errors of performance ranging from 0.462 to 0.720% protein depending on the modeling technique, number of classes used to develop the model, and the wheat class tested. The pooling of wheat classes to produce a general model did not diminish model accuracy. Best results were achieved with an 1100-1400-nm region. Model performance worsened as the wavelength region widened or as the minimum wavelength shifted from 1100 nm to higher values.