U.S. flag

An official website of the United States government


Main content area

A Multispectral Sorting Device for Wheat Kernels

Pearson, Tom, Maghirang, Elizabeth, Dowell, Floyd
American Journal of Agricultural Science and Technology 2013 v.1 no.2 pp. 45
air, animal feeders, chutes, color, economic costs, hard red winter wheat, image analysis, near-infrared spectroscopy, protein content, scab diseases, seeds, sorting, spectral analysis, wavelengths
A multispectral sorting device was constructed using three visible and three near-infrared light-emitting diodes (LED) with peak emission wavelengths of 470 nm (blue), 527 nm (green), 624 nm (red), 850 nm, 940 nm, and 1070 nm. The multispectral data were collected by rapidly (~12 kHz) blinking one LED at a time and subsequently measuring the reflected light from wheat kernels as they dropped off a feeder chute. A microcontroller was used to direct the LED pulses, digitize the analog signal from the photodiode, perform signal processing, and apply classifications. Sorting was accomplished through activation of an air valve that diverted the wheat kernels per the classification assignment. Three applications were tested; the separation of red from white wheat kernels, the separation of Fusarium head blight (FHB)-damaged kernels from undamaged kernels, and the separation of kernels classified as having high, medium, and low protein content. The performance of this LED-based sorter was compared with those of a high-speed color image-based sorter and a single-kernel near-infrared reflectance (SKNIR) sorter (950-1650 nm). The results indicate that the accuracy of the LED-based sorter was comparable to or better than the color image-based sorter. For the sorting of red from white wheat, the LED-based instrument removed 98% of the white wheat while also removing 23.7% of the red wheat in two subsequent passes. In contrast, the color image-based sorter removed less white wheat (83%) while removing more of the red wheat (42.4%). For FHB-damaged kernels, both the LED- and color image-based sorters removed approximately 90% of the kernels with visible symptoms of FHB-damage. However, the LED-based sorter removed fewer undamaged kernels (1.9%) compared to the color image-based sorter (7.0%). For isolating kernels with high protein content, both the LED and color image instruments diverted ~40% of the original wheat sample, which contained ~1.0% higher protein content on average than the original sample. However, the LED sorter was more consistent in isolating the higher-protein kernels across seven different varieties of hard red winter wheat. In comparison, the SKNIR sorter yielded a 1.9% increase in protein content compared with the original unsorted wheat sample. The throughput of the LED sorter was approximately 20 kernels/s compared with ~200 kernels/s for the image sorter and ~0.5 kernels/s for the SKNIR sorter. The cost of the LED-based sorter is expected to be less than either of the other two instruments; in total, the parts for the construction of the LED-based sorter cost approximately $1000. The LED-based sorter will likely be most effectively employed to separate the desired traits of small lots of seed (e.g., breeder size samples), assist in sample purification, and help breeders in selecting the kernels with higher protein and less damage due to FHB.