Discriminating Oat and Groat Kernels from Other Grains Using Near-Infrared Spectroscopy Paul R. Armstrong, 1,† Francesco Dell’Endice, 2 Elizabeth B. Maghirang, 1 and Alisa Rupenyan 2 ABSTRACT Cereal Chem. 94(3):458–463 Oats and groats can be discriminated from other grains such as barley, wheat, rye, and triticale (nonoats) with near-infrared spectroscopy. The two instruments tested herein were the manual version of the United States Department of Agriculture–Agricultural Research Service single-kernel near-infrared (SKNIR) instrument and the automated QualySense QSorter Explorer high-speed sorter, both used in similar near-infrared spectral ranges. Three linear discriminate self-prediction models were developed: 1) oats versus groats + nonoats, 2) oats + groats versus nonoats, and 3) groats versus nonoats. For all three models, the SKNIR instrument showed high correct classification of oats or groats (94.5–100%), which was similar to results of the QSorter Explorer at 95.0–99.4%. The amount of nonoats that were misclassified as oats or groats was low for both instruments at 0–0.2% for the SKNIR instrument and 0.8–3.7% for the QSorter Explorer. Linear discriminate models from independent prediction and validation sets yielded classification accuracies of 91.6–99.3% (SKNIR) and 90.5–97.8% (QSorter Explorer). Small differences in classification accuracy were attributed to processing speeds between the two instruments: 3 kernels/s for the SKNIR instrument and 35 kernels/s for the QSorter Explorer. This indicated that both instruments are useful for quantifying grain sample compositions of oat and groat samples and that both could be useful tools for meeting consumer demand for gluten-free or low-gluten products. Discrimination between grains will help producers and manufacturers meet various regulatory requirements. Examples include requirements such as those from the U.S. Food and Drug Administration and the Commission of European Communities, in which gluten-free oats or other products can only be labeled as nongluten if they contain gluten at less than 20 ppm, the established safe consumption limit for people with celiac disease. The QSorter Explorer is currently being used to meet these requirements. Substantial increases in oat production in the United States were recorded recently, with 85.5 million bushels produced in 2015, a 22% increase over the 70.2 million bushels produced during 2014. This increase was on top of the 8% increase in 2014 compared with the 64.6 million bushels of oat production in 2013 (NASS-USDA 2016). The significant increase in production for 2015 was attributed to the increase in yield per acre of 2.3 bushels/acre over the 2014 average yield of 67.9 bushels/acre, and a 19% increase in harvested area. The International Grains Council (IGC 2013), in its 5-year global supply and demand projection, forecasts feed consumption to remain steady, whereas the supply and demand for food is pro- jected to expand by 2% per annum as consumers capitalize on health benefits from oats and the trend toward increased consumption of breakfast cereals in developing countries. Additionally, the increasing demand for oats may also be related to the results of numerous clinical trials in which oat grains were found to be well tolerated by celiac patients and, hence, have been included as part of gluten- free diet in several countries, including the United States and Canada (Størsrud et al. 2003). Celiac disease, also known as gluten-sensitive enteropathy, is a genetic autoimmune disorder that affects at least 1 in 133 Americans, and studies have shown that a 100% gluten-free diet is the only existing treatment. Oats are one of the gluten-free grains that was reported to be well tolerated by celiac patients. However, several studies have confirmed that commercial oat supplies can be heavily contaminated with gluten from other grains (Thompson et al. 2010; Koerner et al. 2011; Sharma et al. 2015) such as wheat, rye, and barley, the most commonly found contaminants. Thompson (2004) reported proof of gluten contamination in products by Quaker, Country Choice, and McCann’s that were labeled as gluten free. Cross-contamination that can occur in the field or during transport, storage, and processing makes marketing of gluten-free oats difficult. Results of a study by Hollon et al. (2013) showed that a daily intake of 10 mg of gluten by adults with celiac disease, equivalent to daily ingestion of a pound of gluten-free product containing gluten at 20 ppm, caused no intestinal damage, whereas 50 mg was harmful to the majority of the patients. This served as a strong basis for the United Stated Food and Drug Administration (FDA) regulation issued in 2013 for gluten-free food labeling that established a gluten limit of less than 20 ppm for foods that carry the label “gluten free,” “no gluten,” “free of gluten,” or “without gluten” (FDA 2013). The Commission of the European Communities has issued regulations concerning the composition and labeling of foodstuffs suitable for people intolerant to gluten that closely mirror FDA regulations, in which gluten-free and very-low-gluten labels indicate gluten content not exceeding 20 and 100 mg/kg, respectively (Vassiliou 2009). The low level of tolerance makes detecting gluten-containing grains in oats very challenging. Instrumented approaches employed in the United States to obtain pure oats have consisted of sorting by mechanical or optical means. Numerous studies have been conducted on oat inspection, identi- fication, and purification techniques even prior to the increased demand for pure oats. Sapirstein et al. (1987) used digital image analysis to discriminate and classify grain samples of barley, oats, and rye present in hard red spring wheat. They also employed ca- nonical discriminant analysis and showed that wheat and oats had excellent (100%) discrimination from all the grain types. Wu et al. (2013) used color vision to discriminate between different rice va- rieties, with good discrimination for most varieties. Majumdar and Jayas (2000a, 2000b) used machine vision based on kernel mor- phology and color models to develop a grain classification model and found excellent oat classification (approximately 97–100%). Paliwal et al. (2004) used a flatbed scanner as an inexpensive machine vision system to identify and classify various grains; they reported that, for single-kernel images, a set of at least 30 features (morpho- logical, color based, and textural) was required to achieve approxi- mately 96% classification accuracy for oats. Adjemout et al. (2007) used pattern recognition methods based on shape and texture features for corn, oat, barley, and lentil and reported good separation for corn and lentil but overlapped projections for barley and oats. Choudhary et al. (2008) combined all morphological, color-based, textural, and wavelet features by using linear discriminant classification and † Corresponding author. Phone: +1.785.776.2728. E-mail: paul.armstrong@ars.usda.gov 1 U.S. Department of Agriculture–Agricultural Research Service, Center for Grain and Animal Health Research, Stored Product Insect and Engineering Research Unit, 1515 College Avenue, Manhattan, KS 66502, U.S.A. 2 QualySense AG, Glattbrugg, Switzerland. http://dx.doi.org/10.1094/CCHEM-06-16-0162-R This article is in the public domain and not copyrightable. It may be freely reprinted with customary crediting of the source. AACC International, Inc., 2017. 458 CEREAL CHEMISTRY