Location: Food Quality Laboratory2018 Annual Report
1: Enable new or refine commercial optical, viscometry and physical technologies that integrate indicators of starch soundness in wheat and barley. A. Identify sources of variation in viscometry-based methodologies (e.g., falling number) that are used to indicate starch soundness, then develop and test alternative procedures to reduce precision error. B. Develop a near-infrared spectroscopy model, or alternatively a Raman spectroscopy model, for ascertaining mixture levels of conventional and waxy hexaploid wheat. 2: Enable new, real-time rapid optical methods to measure defects and vitreousness in commercial wheat kernels. A. Develop hyperspectral imaging procedures for identification of wheat kernels damaged by scab (Fusarium head blight), black point, heat, and frost, as defined by official inspection criteria. B. Develop imaging system for assessing the percentage of vitreous kernels in durum and hard red spring wheat.
Determine the underlying precision of the falling number procedure under ideal conditions, that is, the test when run with samples for which sampling error has been minimized. Samples of soft white wheat and three samples of club wheat that are representative pre-harvest sprouting will be evaluated. Within each subclass, the three samples are designed to be representative of low (< 300 s), moderate (300-375 s), and high (> 375 s) falling number. All thermal conditions of the falling number instrument within a test day will be held constant as practicable, including the starting temperature of the meal-water mixture, the temperature of the instrument’s bath, the volume of water within the bath, and the mass and volume of the meal-water test sample. In addition to the runs that are based on the conventional amount of meal and water (treatment A: 7 g + 25 mL), three alternate preparations will be examined. Treatment (A, B, C, D) and wheat class or subclass (SWW, Club) will be specified as fixed factors. Samples within wheat class, grind, and portions will be specified as random factors. Minimum-width 95% confidence intervals for the variance estimates will be calculated as wil best linear unbiased predictors (BLUPs) obtained from the mixed model provide estimates of FN averages for each treatment and wheat class sample. Once the precision of the FN method under standard and modified conditions has been established, the research will proceed in two tacks. First, the issue of sampling error will be addressed. Sampling will occur at typical transfer points, such as the truck dump bin at the country elevator, railcar fill sites, and the entry point to the mill. Second, if under standard operation the procedure produces coefficients of variation greater than 3%, modifications to the mixing and heating stage will be explored in an attempt to reduce air bubble entrainment, with rheological measurements (conducted in parallel) performed using a rheometer under constant shear and heating rate conditions. NIR calibrations for mixture levels of conventional and waxy wheat will be developed with accuracies < 5%. Both Linear and non-linear quantitative modeling will be performed on both whole kernel and ground meal binary mixtures. In hyperspectral imaging research of wheat kernels for damage or vitreousness, PCA will be examined to identify the wavelengths at local minima and maxima which inherently possess the greatest contrast. The 10 most sensitive wavelengths will be examined in paired combinations exhaustively (45 trials) using reflectance ratios. Linear discriminant analysis will be used to identify the wavelength pair whose image band ratio produces the greatest percentage of correctly classified kernels, and so on for the next pair. Three regions of interest (ROI) on the kernel, namely the endosperm, germ, and entire kernel, will be separately examined. With each ROI, image processing will be done at the pixel level, whereby subregions of damage in the ROI are first identified; then, depending on the size of the subregion, a decision will be made on whether to categorize the ROI and/or the kernel as normal or damaged.
Significant progress was made in all objectives of this project, which falls under National Program 306, Component 1, Problem Statement 1.A: Define, Measure, and Preserve/Enhance/Reduce Attributes that Impact Quality and Marketability. Addressing the project’s Sub-objective 1.A, we performed a study that evaluated the effect of barometric pressure on ‘falling number’ readings. Falling number is the name given to a wheat quality test that indirectly measures viscosity of a heated mixture of ground wheat and water. Viscosity is strongly influenced by alpha-amylase, the enzyme that breaks down starch into smaller chain molecules and, eventually, into glucose. Because barometric pressure affects water boiling point and boiling water is the source of heat for the falling number test, it has long been known that laboratories located at elevations quite higher than sea level, say 3000 feet, will have falling number readings that are higher than if the test were run at sea level. On a well-defined set of U.S.-grown hard and soft wheat, we performed falling number tests at simulated elevations ranging from 0 to 5,000 feet. From these readings, a universal mathematical correction equation was developed that outputs a falling number as though the test was performed at sea level. The model, by means of simply changing the value of one coefficient, is structured to operate on barometric pressure, land elevation, or water bath temperature, with the choice among these three measurements defined by what is most conveniently measured. We subsequently applied the correction model to a historical set of data of more than 2,000 readings collected at about 25 government laboratories and found that the correction reduced falling number variation caused by elevation by more than 70%. The correction model is in the process of being incorporated in an official directive of the USDA on the falling number procedure. Three additional studies were conducted on falling number, with one of these completed and the other two ongoing. For the completed study, we explored the possibility of developing a substitute procedure for falling number that is based on near-infrared (NIR) spectroscopy. Although the falling number procedure is relatively accurate, its cost, analysis time and required operator skill often preclude this test from where it is needed the most, the first point of sale, typically the country elevator. By contrast, near-infrared (NIR) spectroscopy is a technology that enjoys widespread use at the country elevator for analysis of moisture and protein. Past studies using early NIR instrumentation and statistical algorithms were not successful in producing workable NIR models for falling number. Our study, conducted at the behest of the U.S. Pacific Northwest wheat industry, sought to revisit this challenge using up-to-date equipment and software. Based on a very large collection of soft white wheat breeders' samples grown in multiple locations in Washington State, NIR modeling of falling number was explored by two approaches. The first approach was based on statistical regression modeling to emulate falling number directly, while the second approach was based on providing a pass/fail decision on whether the estimated falling number fell above or below a prescribed threshold. In either approach, the results indicated that NIR spectroscopy is not a suitable replacement to the actual falling number procedure. For the ongoing studies, one is examining how microclimate variation within a field due to differences in elevation and solar exposure affects falling number. This is of particular interest in the Pacific Northwest, where large variations in terrain within a single field cause large daily temperature extremes (low and high), which in turn are thought to cause the synthesis of alpha-amylase during seed development that is generally (though inconclusively) thought to be unfavorable in the same sense as when the enzyme is synthesized during rainy conditions right before harvest. In this first year, three commercial winter wheat fields, each greater than 100 acres in size, were hand-sampled at harvest at 17 to 21 geospatially referenced sites within each field. Based on the geospatial coordinates of each site, land elevation, slope, and aspect were determined using 1/3 arc-second digital elevation model (DEM) raster files obtained from the USGS National Elevation Dataset and coordinated with commercial geo spatial software. Likewise, direct solar radiation flux for Julian days 128-218 was calculated for each site. To date, we have found that an elevation effect on falling number is not significant despite the sites differing by as much as 50 m within a field. Contrarily, direct solar flux, a function of aspect, slope and shadowing from nearby geographical features, is positively correlated with falling number (r = 0.48 to 0.69). Research continues in the 2018 season at three additional fields. Ultimately, this work may provide a tool that can be used at time of harvest for making decisions on the binning of low falling number wheat from susceptible regions within a field. The second ongoing study began in late spring of 2018 on developing a standard material for the falling number procedure that can be used at multiple laboratories for monitoring the performance of a network of falling number instruments, such as in official government operations. Presently, we are developing a protocol for testing our own standard material among four laboratories, in which within laboratory precision (repeatability), instrument alignment, and storage time stability will be assessed. The project’s Sub-objective 1.B addresses the development of near-infrared spectral imaging models of intact wheat kernels to measure mixture levels of conventional wheat and very low amylose, or waxy, wheat, which is a relatively new product in the United States. Unique processing characteristics and end-use products for waxy wheat give it a premium value in the marketplace. Contingent with the premium, however, will be a mechanism for ensuring the purity of a waxy wheat lot, with spectral imaging explored as a possibility. In addition to traditional linear mathematical functions (linear discriminant analysis, partial least square discriminant analysis) we previously developed, we recently examined non-linear mathematical functions. Four classifiers were used, including stats linear classifier, k-nearest neighbor classifier, decision tree classifier, and nearest mean classifier. While the stats linear classifier had performance on par with the best of the traditional models, the other three classifiers demonstrated significantly poorer performance. With either traditional or the newly examined models, the results of performing waxy-conventional sorting based on spectral imaging are not as good as those obtained by us using conventional (non-imaging) spectrometry; however, the potential advantage with the imaging technique is in its adaptability to sorting operations. Addressing the project’s second objective, there is one active sub-objective, this being on the development of hyperspectral imaging (HSI) for evaluation of mold damage in wheat kernels, particularly that caused by scab, also known as Fusarium head blight. Older work described in the literature relied on color digital imaging on bulk samples of wheat (i.e., not individual kernels), with the level of fusarium damage equated to the proportion of pixels whose saturation values were within a user-determined range indicative of infection. However, the saturation threshold must be adjusted according to variety and environmental conditions. Our work in the past year was intended to avoid this level of fine tuning by using longer wavelength radiation than the visible, and doing the analysis on individual kernels identified in HSI analysis. We developed a linear discriminant analysis (LDA) model based on mean of reflectance values of the interior pixels of each kernel at four wavelengths (1100, 1197, 1308, and 1394 nm) to differentiate between sound and scab-damaged kernels. We examined other input variables in the LDA, including kernel morphological properties and histogram features from the pixel responses of selected wavelengths of each kernel within a ‘hypercube’, but found that the mean response at the selected wavelengths was comparable, if not superior. The benefit of this finding is that future spectral imaging hardware may be based on multi-spectral instead of hyperspectral design, which translates into less expense. The results of the classification modeling indicate the strong potential of HSI in determining fusarium-damaged kernels. At the same time, however, improvement in aligning this procedure to visual analysis is hampered by the inherent level of subjectivity in visual analysis. Work is now being directed toward other damage conditions affecting wheat.
1. A new USDA standard protocol for determination of wheat quality. “Falling number” (FN) is a procedure used to gauge the level of naturally occurring alpha-amylase in wheat (the enzyme responsible for breaking down starch), a critical determinant of wheat quality. Because the procedure uses boiling water as a heat source as well as the medium for starch break down, it is affected by barometric pressure and, in turn, the land elevation of the laboratory. This can lead to falsely high FN values when the test is performed at elevations above 1,000 feet. Working at simulated elevations between 0 and 5,000 feet, ARS scientists in Beltsville, Maryland, developed a correction equation for FN that allows this value to be reported on a sea-level basis. Initially requested by USDA’s Federal Grain Inspection Service, the equation was turned over to that agency for incorporation into a directive that guides federal, state, and private laboratories on the FN procedure. This issue is important because overseas customers of U.S. wheat often have requirements on the FN value of cargo lots, thus making the procedure’s accuracy a paramount concern to our exporters, especially in the Pacific Northwest.
2. Estimation of fusarium-damaged wheat kernels by near-infrared hyperspectral imaging. Fusarium head blight (FHB) is among the most common fungal diseases affecting wheat. This preharvest disease causes decreased yield, low-density kernels and, most concerning, the potential for occurrence of the mycotoxin deoxynivalenol, a compound toxic to humans and livestock. Human visual analysis of wheat samples has been the traditional method for FHB assessment in both official inspection and plant breeding operations. While not requiring specialized equipment, visual analysis is dependent on a trained and consistent workforce. ARS scientists in Beltsville, Maryland have developed an alternative to visual analysis that uses spectral imaging at discrete wavelengths in the near-infrared region. Combined with image processing of the randomly arranged wheat kernels in samples, each containing a few hundred kernels, this alternative procedure estimates the percentage of fusarium-damaged kernels in a field sample (for plant breeding programs) or grade sample (for inspection programs). The benefits of the procedure are it removes the subjective component of damage assessment, and it is scalable to automatic sorting operations.
Delwiche, S.R., Qin, J., Graybosch, R.A., Rausch, S.R., Kim, M.S. 2018. Near-infrared hyperspectral imaging of blends of conventional and waxy hard wheats. Journal of Spectral Imaging. 7(a2):1-13.
Delwiche, S.R., Steber, C.M. 2018. Falling number of soft wheat wheat by near-infrared spectroscopy: a challenge revisited. Cereal Chemistry. 95(3):469-477.
Delwiche, S.R., Rausch, S.R., Vinyard, B.T. 2018. Correction of wheat meal falling number to a common barometric pressure at simulated laboratory elevations of 0 to 1500 meters. Cereal Chemistry. 95(3):428-435.