Hardness Measurement of Bulk Wheat by Single‐Kernel Visible and Near‐Infrared Reflectance Spectroscopy

Reflectance spectra (400 to 1700 nm) of single wheat kernels collected using the Single Kernel Characterization System (SKCS) 4170 were analyzed for wheat grain hardness using partial least squares (PLS) regression. The wavelengths (650 to 700, 1100, 1200, 1380, 1450, and 1670 nm) that contributed most to the ability of the model to predict hardness were related to protein, starch, and color differences. Slightly better prediction results were observed when the 550–1690 nm region was used compared with 950–1690 nm region across all sample sizes. For the 30‐kernel mass‐averaged model, the hardness prediction for 550–1690 nm spectra resulted in a coefficient of determination (R²) = 0.91, standard error of cross validation (SECV) = 7.70, and relative predictive determinant (RPD) = 3.3, while the 950–1690 nm had R² = 0.88, SECV = 8.67, and RPD = 2.9. Average hardness of hard and soft wheat validation samples based on mass‐averaged spectra of 30 kernels was predicted and compared with the SKCS 4100 reference method (R² = 0.88). Compared with the reference SKCS hardness classification, the 30‐kernel (550–1690 nm) prediction model correctly differentiated (97%) between hard and soft wheat. Monte Carlo simulation technique coupled with the SKCS 4100 hardness classification logic was used for classifying mixed wheat samples. Compared with the reference, the prediction model correctly classified mixed samples with 72–100% accuracy. Results confirmed the potential of using visible and near‐infrared reflectance spectroscopy of whole single kernels of wheat as a rapid and nondestructive measurement of bulk wheat grain hardness.     

Selecting and Sorting Waxy Wheat Kernels Using Near‐Infrared Spectroscopy

An automated single kernel near‐infrared (NIR) sorting system was used to separate single wheat (Triticum aestivum L.) kernels with amylose‐free (waxy) starch from reduced‐amylose (partial waxy) or wild‐type wheat kernels. Waxy kernels of hexaploid wheat are null for the granule‐bound starch synthase alleles at all three Wx gene loci; partial waxy kernels have at least one null and one functional allele. Wild‐type kernels have three functional alleles. Our results demonstrate that automated single kernel NIR technology can be used to select waxy kernels from segregating breeding lines or to purify advanced breeding lines for the low‐amylose kernel trait. Calibrations based on either amylose content or the waxy trait performed similarly. Also, a calibration developed using the amylose content of waxy, partial waxy, and wild‐type durum (T. turgidum L. var durum) wheat enabled adequate sorting for hard red winter and hard red spring wheat with no modifications. Regression coefficients indicated that absorption by starch in the NIR region contributed to the classification models. Single kernel NIR technology offers significant benefits to breeding programs that are developing wheat with amylose‐free starches.     

Single Wheat Kernel Color Classification by Using Near-Infrared Reflectance Spectra

An optical radiation measurement system, which measured reflectance spectra, log (1/R), from 400 to 2,000 nm, was used to quantify single wheat kernel color. Six classes of wheat samples were used for this study, including red wheat that appears white and white wheat that appears red. Partial least squares regression and multiple linear regression were used to develop classification models with three wavelength regions, 500–750, 500–1,700, and 750–1,900 nm, and three data pretreatments, log (1/R), first derivative, and second derivative. For partial least squares models, the highest classification accuracy was 98.5% with the wavelength region of 500–1,700 nm. The log (1/R) and the first derivative yielded higher classification accuracy than the second derivative. For multiple linear regression models, the highest classification accuracy was 98.1% obtained from log (1/R) spectra from the visible and near-infrared wavelength regions.     

Determining Vitreous Subclasses of Hard Red Spring Wheat Using Visible/Near‐Infrared Spectroscopy

The percentage of dark hard vitreous (DHV) kernels in hard red spring wheat is an important grading factor that is associated with protein content, kernel hardness, milling properties, and baking quality. The current visual method of determining DHV and non‐DHV (NDHV) wheat kernels is time‐consuming, tedious, and subject to large errors. The objective of this research was to classify DHV and NDHV wheat kernels, including kernels that were checked, cracked, sprouted, or bleached using visible/near‐infrared (Vis/NIR) spectroscopy. Spectra from single DHV and NDHV kernels were collected using a diode‐array NIR spectrometer. The dorsal and crease sides of the kernels were viewed. Three wavelength regions, 500–750 nm, 750–1,700 nm, and 500–1700 nm were compared. Spectra were analyzed by using partial least squares (PLS) regression. Results suggest that the major contributors to classifying DHV and NDHV kernels are light scattering, protein content, kernel hardness, starch content, and kernel color effects on the absorption spectrum. Bleached kernels were the most difficult to classify because of high lightness values. The sample set with bleached kernels yielded lower classification accuracies of 91.1–97.1% compared with 97.5–100% for the sample set without bleached kernels. More than 75% of misclassified kernels were bleached. For sample sets without bleached kernels, the classification models that included the dorsal side gave the highest classification accuracies (99.6–100%) for the testing sample set. Wavelengths in both the Vis/NIR regions or the NIR region alone yielded better classification accuracies than those in the visible region only.     

Moisture Effects on the Prediction Performance of a Single‐Kernel Near‐Infrared Deoxynivalenol Calibration

Fusarium head blight (FHB) is a serious disease in wheat that affects grain quality owing to the accumulation of mycotoxins such as deoxynivalenol (DON) in grains. Near‐infrared (NIR) spectroscopy has been used to develop techniques to estimate DON levels in single wheat kernels to facilitate rapid, nondestructive screening of FHB resistance in wheat breeding lines. The effect of moisture content (MC) variation on the accuracy of single‐kernel DON prediction by NIR spectroscopy was investigated. Sample MC considerably affected accuracy of the current NIR DON calibration by underestimating or overestimating DON at higher or lower moisture levels, respectively. DON in single kernels was most accurately estimated at 13–14% MC. Major NIR absorptions related to Fusarium damage were found around 1,198–1,200, 1,418–1,430, 1,698, and 1,896–1,914 nm. Major moisture related absorptions were observed around 1,162, 1,337, 1,405–1,408, 1,892–1,924, and 2,202 nm. Fusarium damage and moisture related absorptions overlapped in the 1,380–1,460 and 1,870–1,970 nm regions. These results show that absorption regions associated with water are often close to absorption regions associated with Fusarium damage. Thus, care must be taken to develop DON calibrations that are independent of grain MC.     

Assessment of Heat-Damaged Wheat Kernels Using Near-Infrared Spectroscopy

Heat damage is a serious problem frequently associated with wet harvests because of improper storage of damp grain or artificial drying of moist grain at high temperatures. Heat damage causes protein denaturation and reduces processing quality. The current visual method for assessing heat damage is subjective and based on color change. Denatured protein related to heat damage does not always cause a color change in kernels. The objective of this research was to evaluate the use of nearinfrared (NIR) reflectance spectroscopy to identify heat-damaged wheat kernels. A diode-array NIR spectrometer, which measured reflectance spectra (log (1/R)) from 400 to 1,700 nm, was used to differentiate single kernels of heat-damaged and undamaged wheats. Results showed that light scattering was the major contributor to the spectral characteristics of heat-damaged kernels. For partial least squares (PLS) models, the NIR wavelength region of 750–1,700 nm provided the highest classification accuracy (100%) for both cross-validation of the calibration sample set and prediction of the test sample set. The visible wavelength region (400–750 nm) gave the lowest classification accuracy. For two-wavelength models, the average of correct classification for the classification sample set was >97%. The average of correct classification for the test sample set was generally >96% using two-wavelength models. Although the classification accuracies of two-wavelength models were lower than those of the PLS models, they may meet the requirements for industry and grain inspection applications.     

Near‐Infrared Hyperspectral Imaging of Fusarium‐Damaged Oats (Avena sativa L.)

The feasibility of hyperspectral imaging (HSI) to detect deoxynivalenol (DON) content and Fusarium damage in single oat kernels was investigated. Hyperspectral images of oat kernels from a Fusarium‐inoculated nursery were used after visual classification as asymptomatic, mildly damaged, and severely damaged. Uninoculated kernels were included as controls. The average spectrum from each kernel was paired with the reference DON value for the same kernel, and a calibration model was fitted by partial least squares regression (PLSR). To correct for the skewed distribution of DON values and avoid nonlinearities in the model, the DON values were transformed as DON* = [log(DON)]³. The model was optimized by cross‐validation, and its prediction performance was validated by predicting DON* values for a separate set of validation kernels. The PLSR model and linear discriminant analysis classification were further used on single‐pixel spectra to investigate the spatial distribution of infection in the kernels. There were clear differences between the kernel classes. The first component separated the uninoculated and asymptomatic from the severely damaged kernels. Infected kernels showed higher intensities at 1,925, 2,070, and 2,140 nm, whereas noninfected kernels were dominated by signals at 1,400, 1,626, and 1,850 nm. The DON* values of the validation kernels were estimated by using their average spectra, and the correlation (R) between predicted and measured DON* was 0.8. Our results show that HSI has great potential in detecting Fusarium damage and predicting DON in oats, but it needs more work to develop a model for routine application.     

Determining Weight and Moisture Properties of Sound and Fusarium‐Damaged Single Wheat Kernels by Near‐Infrared Spectroscopy

Single kernel moisture content (MC) is important in the measurement of other quality traits in single kernels because many traits are expressed on a dry weight basis. MC also affects viability, storage quality, and price. Also, if near‐infrared (NIR) spectroscopy is used to measure grain traits, the influence of water must be accounted for because water is a strong absorber throughout the NIR region. The feasibility of measurement of MC, fresh weight, dry weight, and water mass of single wheat kernels with or without Fusarium damage was investigated using two wheat cultivars with three visually selected classes of kernels with Fusarium damage and a range of MC. Calibration models were developed either from all kernel classes or from only undamaged kernels of one cultivar that were then validated using all spectra of the other cultivar. A calibration model developed for MC when using all kernels from the wheat cultivar Jagalene had a coefficient of determination (R²) of 0.77 and standard error of cross validation (SECV) of 1.03%. This model predicted the MC of the wheat cultivar 2137 with R² of 0.81 and a standard error of prediction (SEP) of 1.02% and RPD of 2.2. Calibration models developed using all kernels from both cultivars predicted MC, fresh weight, dry weight, or water mass in kernels better than models that used only undamaged kernels from both cultivars. Single kernel water mass was more accurately estimated using the actual fresh weight of kernels and MC predicted by calibrations that used all kernels or undamaged kernels. The necessity for evaluating and expressing constituent levels in single kernels on a mass/kernel basis rather than a percentage basis was elaborated. The need to overcome the effects of kernel size and water mass on single kernel spectra before using in calibration model development was also highlighted.     

Evaluation of a Near‐Infrared Reflectance Spectrometer as a Granulation Sensor for First‐Break Ground Wheat: Studies with Six Wheat Classes

In flour milling, a granulation sensor for ground wheat is needed for automatic control of a roller mill's roll gap. A near‐infrared (NIR) reflectance spectrometer was evaluated as a potential granulation sensor of first‐break ground wheat using offline methods. Sixty wheat samples, ground independently, representing six classes and five roller mill gaps, were each used for calibration and validation sets. Partial least squares regression was used to develop the models with cumulative mass of size fraction as the reference value. Combinations of four data pretreatments (log (1/R), baseline correction, unit area normalization, and derivatives) and three wavelength regions (700–1,500, 800–1,600, and 600–1,700 nm) were evaluated. Unit area normalization combined with baseline correction or second derivative yielded models that predicted well each size fraction of first‐break ground wheat. Standard errors of performance of 4.07, 1.75, 1.03, and 1.40 and r² of 0.93, 0.90, 0.88, and 0.38 for the >1,041‐, >375‐, >240‐, and >136‐μm size ranges, respectively, were obtained for the best model. Results indicate that the granulation sensing technique based on NIR reflectance is ready for online evaluation.     

Near‐Infrared Spectroscopic Method for Identification of Fusarium Head Blight Damage and Prediction of Deoxynivalenol in Single Wheat Kernels

Fusarium Head Blight (FHB), or scab, can result in significant crop yield losses and contaminated grain in wheat (Triticum aestivum L.). Growing less susceptible cultivars is one of the most effective methods for managing FHB and for reducing deoxynivalenol (DON) levels in grain, but breeding programs lack a rapid and objective method for identifying the fungi and toxins. It is important to estimate proportions of sound kernels and Fusarium‐damaged kernels (FDK) in grain and to estimate DON levels of FDK to objectively assess the resistance of a cultivar. An automated single kernel near‐infrared (SKNIR) spectroscopic method for identification of FDK and for estimating DON levels was evaluated. The SKNIR system classified visually sound and FDK with an accuracy of 98.8 and 99.9%, respectively. The sound fraction had no or very little accumulation of DON. The FDK fraction was sorted into fractions with high or low DON content. The kernels identified as FDK by the SKNIR system had better correlation with other FHB assessment indices such as FHB severity, FHB incidence and kernels/g than visual FDK%. This technique can be successfully employed to nondestructively sort kernels with Fusarium damage and to estimate DON levels of those kernels. Single kernels could be predicted as having low (<60 ppm) or high (>60 ppm) DON with ≈96% accuracy. Single kernel DON levels of the high DON kernels could be estimated with R² = 0.87 and standard error of prediction (SEP) of 60.8 ppm. Because the method is nondestructive, seeds may be saved for generation advancement. The automated method is rapid (1 kernel/sec) and sorting grains into several fractions depending on DON levels will provide breeders with more information than techniques that deliver average DON levels from bulk seed samples.     

Differentiating Vitreous and Nonvitreous Durum Wheat Kernels by Using Near‐Infrared Spectroscopy

The vitreousnss of durum wheat is used by the wheat industry as an indicator of milling and cooking quality. The current visual method of determining vitreousness is subjective, and classification results between inspectors and countries vary widely. Thus, the use of near‐infrared (NIR) spectroscopy to objectively classify vitreous and nonvitreous single kernels was investigated. Results showed that classification of obviously vitreous or nonvitreous kernels by the NIR procedure agreed almost perfectly with inspector classifications. However, when difficult‐to‐classify vitreous and nonvitreous kernels were included in the analysis, the NIR procedure agreed with inspectors on only 75% of kernels. While the classification of difficult kernels by NIR spectroscopy did not match well with inspector classifications, this NIR procedure quantifies vitreousness and thus may provide an objective classification means that could reduce inspector‐to‐inspector variability. Classifications appear to be due, at least in part, to scattering effects and to starch and protein differences between vitreous and nonvitreous kernels.     

Detecting and Segregating Black Tip‐Damaged Wheat Kernels Using Visible and Near‐Infrared Spectroscopy

Detection of individual wheat kernels with black tip symptom (BTS) and black tip damage (BTD) was demonstrated with near‐infrared reflectance spectroscopy (NIRS) and silicon light‐emitting‐diode (LED) based instruments. The two instruments tested, a single‐kernel NIRS instrument (SKNIRS) and a silicon LED‐based single‐kernel high‐speed sorter (SiLED‐SKS) were both developed by the Stored Product Insect and Engineering Research Unit, Center for Grain and Animal Health Research, USDA Agricultural Research Service. BTD was classified into four levels for the study ranging from sound, symptomatic (BTS) at two levels, and damaged (BTD). Discriminant analysis models for the SKNIRS instrument could distinguish sound undamaged kernels well, correctly classifying kernels 80% of the time. Damaged kernels were classified with 67% accuracy and symptomatic kernels at about 44%. Higher classification accuracy (81–87%) was obtained by creating only two groupings: 1) combined sound and lightly symptomatic kernels and 2) combined heavily symptomatic and damaged kernels. A linear regression model was developed from the SiLED‐SKS sorted fractions to predict the percentage of combined BTS and BTD kernels in a sample. The model had an R² of 0.64 and a standard error of prediction of 7.4%, showing it had some measurement ability for BTS and BTD. The SiLED‐SKS correctly classified and sorted out 90% of BTD and 66% of BTS for all 28 samples after three passes through the sorter. These instruments can serve as important tools for plant breeders and grading facilities of the wheat industry that require timely and objective determination and sorting of different levels of black tip present in wheat samples.     

Detection of Sprouted and Midge‐Damaged Wheat Kernels Using Near‐Infrared Hyperspectral Imaging

Sprout damage which results in poor breadmaking quality due to enzymatic activity of α‐amylase is one of the important grading factors of wheat in Canada. Potential of near‐infrared (NIR) hyperspectral imaging was investigated to detect sprouting of wheat kernels. Artificially sprouted, midge‐damaged, and healthy wheat kernels were scanned using NIR hyperspectral imaging system in the range of 1000–1600 nm at 60 evenly distributed wavelengths. Multivariate image analysis (MVI) technique based on principal components analysis (PCA) was applied to reduce the dimensionality of the hyperspectral data. Three wavelengths 1101.7, 1132.2, and 1305.1 nm were identified as significant and used in analysis. Statistical discriminant classifiers (linear, quadratic, and Mahalanobis) were used to classify sprouted, midge‐damaged, and healthy wheat kernels. The discriminant classifiers gave maximum accuracy of 98.3 and 100% for classifying healthy and damaged kernels, respectively.     

An Automated Near‐Infrared System for Selecting Individual Kernels Based on Specific Quality Characteristics

An automated sorting system was developed that nondestructively measured quality characteristics of individual kernels using near‐infrared (NIR) spectra. This single‐kernel NIR system was applied to sorting wheat (Triticum aestivum L.) kernels by protein content and hardness, and proso millet (Panicum miliaceum L.) into amylose‐bearing and amylose‐free fractions. Single wheat kernels with high protein content could be sorted from pure lines so that the high‐protein content portion was 3.1 percentage points higher than the portion with the low‐protein kernels. Likewise, single wheat kernels with specific hardness indices could be removed from pure lines such that the hardness index in the sorted samples was 29.4 hardness units higher than the soft kernels. The system was able to increase the waxy, or amylose‐free, millet kernels in segregating samples from 94% in the unsorted samples to 98% in the sorted samples. The portion of waxy millet kernels in segregating samples was increased from 32% in the unsorted samples to 55% after sorting. Thus, this technology can be used to enrich the desirable class within segregating populations in breeding programs, to increase the purity of heterogeneous advanced or released lines, or to measure the distribution of quality within samples during the marketing process.     

Predicting Scab,Vomitoxin, and Ergosterol in Single Wheat Kernels Using Near-Infrared Spectroscopy

Near-infrared spectroscopy (NIRS) was used to detect scab damage and estimate deoxynivalenol (DON) and ergosterol levels in single wheat kernels. Results showed that all scab-damaged kernels identified by official inspectors were correctly identified by NIRS. In addition, this system identified more kernels with DON than did a visual inspection. DON and ergosterol were predicted with standard errors of ≈40 and 100 ppm, respectively. All samples with visible scab had single kernels with DON levels >120 ppm, and some kernels contained >700 ppm of DON. This technology may provide a means of rapidly screening samples for potential food safety and quality problems related to scab damage.     

Automated Color Classification of Single Wheat Kernels Using Visible and Near-Infrared Reflectance

Modification of an existing single kernel wheat characterization system allowed collection of visible and near-infrared (NIR) reflectance spectra (450–1,688 nm) at a rate of 1 kernel/4 sec. The spectral information was used to classify red and white wheats in an attempt to remove subjectivity from class determinations. Calibration, validation, and prediction results showed that calibrations using partial least squares regression and derived from the full wavelength profile correctly classed more kernels than either the visible region (450–700 nm) or the NIR region (700–1,688 nm). Most results showed >99% correct classification for single kernels when using the visible and NIR regions. Averaging of single kernel classifications resulted in 100% correct classification of bulk samples.     

Single Kernel Near‐Infrared Analysis of Tetraploid (Durum) Wheat for Classification of the Waxy Condition

Plant breeding programs are active worldwide in the development of waxy hexaploid (Triticum aestivum L.) and tetraploid (T. turgidum L. var. durum) wheats. Conventional breeding practices will produce waxy cultivars adapted to their intended geographical region that confer unique end use characteristics. Essential to waxy wheat development, a means to rapidly and, ideally, nondestructively identify the waxy condition is needed for point‐of‐sale use. The study described herein evaluated the effectiveness of near‐infrared (NIR) reflectance single‐kernel spectroscopy for classification of durum wheat into its four possible waxy alleles: wild type, waxy, and the two intermediate states in which a null allele occurs at either of the two homologous genes (Wx‐1A and Wx‐1B) that encodes for the production of the enzyme granule bound starch synthase (GBSS) that controls amylose synthesis. Two years of breeders' samples (2003 and 2004), corresponding to 47 unique lines subdivided about equally into the four GBSS genotypes, were scanned in reflectance (1,000–1,700 nm) on an individual kernel basis. Linear discriminant analysis models were developed using the best set of four wavelengths, best four wavelength differences, and best four principal components. Each model consistently demonstrated the high ability (typically >95% of the time) to classify the fully waxy genotype. However, correct classification among the three other genotypes (wild type, wx‐A1 null, and wx‐B1 null) was generally not possible.     

Estimation of the Deoxynivalenol and Moisture Contents of Bulk Wheat Grain Samples by FT‐NIR Spectroscopy

Deoxynivalenol (DON) levels in harvested grain samples are used to evaluate the Fusarium head blight (FHB) resistance of wheat cultivars and breeding lines. Fourier transform near‐infrared (FT‐NIR) calibrations were developed to estimate the DON level and moisture content (MC) of bulk wheat grain samples harvested from FHB screening trials. Grains in a rotating glass petri dish were scanned in the 10,000–4,000 cm⁻¹ (1,000–2,500 nm) spectral range using a Perkin Elmer Spectrum 400 FT‐IR/FT‐NIR spectrometer. The DON calibration predicted the DON levels in test samples with R ² = 0.62 and root mean square error of prediction (RMSEP) = 8.01 ppm. When 5–25 ppm of DON was used as the cut‐off to classify samples into low‐ and high‐DON groups, 60.8–82.3% of the low‐DON samples were correctly classified, whereas the classification accuracy of the high‐DON group was 82.3–94.0%. The MC calibration predicted the MC in grain samples with R ² = 0.98 and RMSEP = 0.19%. Therefore, these FT‐NIR calibrations can be used to rapidly prescreen wheat lines to identify low‐DON lines for further evaluation using standard laboratory methods, thereby reducing the time and costs of analyzing samples from FHB screening trials.     

Reduction of Aflatoxin and Fumonisin Contamination in Yellow Corn by High‐Speed Dual‐Wavelength Sorting

A high‐speed dual‐wavelength sorter was tested for removing corn contaminated in the field with aflatoxin and fumonisin. To achieve accurate sorting, single kernel reflectance spectra (500–1,700 nm) were analyzed to select the optimal pair of optical filters to detect mycotoxin‐contaminated corn during high‐speed sorting. A routine, based on discriminant analysis, was developed to select the two absorbance bands in the spectra that would give the greatest classification accuracy. In a laboratory setting, and with the kernels stationary, absorbances at 750 and 1,200 nm could correctly identify >99% of the kernels as aflatoxin‐contaminated (>100 ppb) or uncontaminated. A high‐speed sorter was tested using the selected filter pair for corn samples inoculated with Aspergillus flavus; naturally infested corn grown in central Illinois; and naturally infested, commercially grown and harvested corn from eastern Kansas (2002 harvest). For the Kansas corn, the sorter was able to reduce aflatoxin levels by 81% from an initial average of 53 ppb, while fumonisin levels in the same grain samples were reduced an average of 85% from an initial level of 17 ppm. Similar reductions in mycotoxin levels were observed after high‐speed sorting of A. flavus inoculated and naturally mold‐infested corn grown in Illinois.