Skip to main content
ARS Home » Research » Publications at this Location » Publication #173229
Title: HIGH-SPEED OPTICAL SORTING OF SOFT RED WINTER WHEAT FOR REMOVAL OF FUSARIUM-DAMAGED KERNELS

Author
item Delwiche, Stephen - Steve
item Pearson, Thomas
item Gaines, Charles

Submitted to: Meeting Abstract
Publication Type: Proceedings
Publication Acceptance Date: 12/11/2004
Publication Date: 12/11/2004
Citation: Delwiche, S.R., Pearson, T.C., Gaines, C.S. 2004. High-speed optical sorting of soft red winter wheat for removal of Fusarium-damaged kernels. Second International Symposium on Fusarium Head Blight Proceedings. Dec. 11-15, 2004, Orlando, Florida. p. 397.

Interpretive Summary:

Technical Abstract: Our previous work has examined the accuracy of a semi-automated wheat scab inspection system that is based on near-infrared (NIR) reflectance (1000 to 1700 nm) of individual kernels. Classification analysis has involved the application of various statistical classification techniques, including linear discriminant analysis (LDA), soft independent modeling of class analogy (SIMCA), partial least squares (PLS) regression, and non-parametric (k-nearest-neighbor) classification. Recent research has focused on the determination of the most suitable visible or near-infrared wavelengths that could be used in high-speed sorting for removal of FHB-infected soft red winter wheat kernels. Current technology in high-speed sorters limits the number of spectral wavelengths (regions) of the detectors to no more than two. Hence, the critical aspect of this study has been the search for the single wavelengths and best two-wavelength combinations that maximize class separation, using LDA. Four thousand eight hundred kernels from 100 commercial varieties, equally divided between normal and scab-damaged categories, were individually scanned in the extended visible (410-865 nm) and near-infrared (1031-1674 nm) regions. Single- and all combinations of two-wavelength LDA models were developed and characterized through cross-validation by the average correctness of classification percentages. Short visible (~420 nm) and moderate near-infrared (1450-1500 nm) wavelengths produced the highest single-term classification accuracies (at approximately 77% and 83%, respectively). The best two-term models occurred near the wavelengths of 500 and 550 nm for the visible region alone (94% accuracy), 1152 and 1248 nm for the near-infrared region alone (97%), and 750 and 1476 nm for the hybrid region (86%). These wavelengths are, therefore, considered of importance in the design of monochromatic and bichromatic high-speed sorters for scab-damage reduction. Ongoing research is presently examining the efficiency of high-speed sorting for Fusarium-damaged kernels, as measured by reduction in DON concentration. Approximately 40 5-kg commercial samples of soft red winter wheat have undergone as many as three successive sorts, using a commercial sorter outfitted with filters at 675 and 1470 nm. Results indicate a significant reduction in DON is achieved through sorting; however, this comes at the expense of false positives (good kernels diverted to reject stream) and the overall reduction in material available for processing.