DTU Informatics, Technical University of Denmark, Richard Petersens Plads Bldg 321, DK-2800 Kgs. Lyngby, Denmark.
J Agric Food Chem. 2011 Nov 9;59(21):11385-94. doi: 10.1021/jf202122y. Epub 2011 Oct 13.
Pregermination is one of many serious degradations to barley when used for malting. A pregerminated barley kernel can under certain conditions not regerminate and is reduced to animal feed of lower quality. Identifying pregermination at an early stage is therefore essential in order to segregate the barley kernels into low or high quality. Current standard methods to quantify pregerminated barley include visual approaches, e.g. to identify the root sprout, or using an embryo staining method, which use a time-consuming procedure. We present an approach using a near-infrared (NIR) hyperspectral imaging system in a mathematical modeling framework to identify pregerminated barley at an early stage of approximately 12 h of pregermination. Our model only assigns pregermination as the cause for a single kernel's lack of germination and is unable to identify dormancy, kernel damage etc. The analysis is based on more than 750 Rosalina barley kernels being pregerminated at 8 different durations between 0 and 60 h based on the BRF method. Regerminating the kernels reveals a grouping of the pregerminated kernels into three categories: normal, delayed and limited germination. Our model employs a supervised classification framework based on a set of extracted features insensitive to the kernel orientation. An out-of-sample classification error of 32% (CI(95%): 29-35%) is obtained for single kernels when grouped into the three categories, and an error of 3% (CI(95%): 0-15%) is achieved on a bulk kernel level. The model provides class probabilities for each kernel, which can assist in achieving homogeneous germination profiles. This research can further be developed to establish an automated and faster procedure as an alternative to the standard procedures for pregerminated barley.
发芽前处理是大麦用于制麦时的许多严重退化现象之一。在某些条件下，发芽的大麦籽粒可能无法重新发芽，从而降低为质量较低的动物饲料。因此，早期识别发芽前处理对于将大麦籽粒分离为低质量或高质量至关重要。目前，定量发芽前处理大麦的标准方法包括使用视觉方法（例如，识别根芽）或使用胚胎染色方法，这些方法都需要耗时的程序。我们提出了一种使用近红外（NIR）高光谱成像系统在数学建模框架中识别发芽前处理大麦的方法，该方法可以在大约 12 小时的发芽前处理早期识别发芽前处理。我们的模型仅将发芽前处理作为单个籽粒无法发芽的原因，并不能识别休眠、籽粒损伤等。该分析基于超过 750 个罗莎琳娜大麦籽粒，根据 BRF 方法在 0 到 60 小时之间进行了 8 个不同持续时间的发芽前处理。重新发芽这些籽粒揭示了将发芽前处理的籽粒分为三组：正常、延迟和有限发芽。我们的模型采用基于一组对籽粒方向不敏感的提取特征的有监督分类框架。当将单个籽粒分为三组时，模型的分类错误率为 32%（置信区间（95%）：29-35%），而在批量籽粒水平上的错误率为 3%（置信区间（95%）：0-15%）。该模型为每个籽粒提供类别概率，可以帮助实现均匀的发芽曲线。这项研究可以进一步发展，以建立一种自动化和更快的程序，作为发芽前处理大麦的标准程序的替代方法。
