LSU AgCenter researchers are continuing to develop an imaging system to uniquely and quickly analyze soybeans with the objective of determining, among other things, potentially valuable qualities that have remained hidden during the inspection process such as oil and protein content.

Typically, damaged and disease-affected soybeans are marked as damaged during inspection and grading. But, this highly automated prototype imaging system, with the use of machine learning algorithms and shortwave infrared reflectance (SWIR) imaging, could help identify previously unrealized value and potentially open up a new avenue to analyze postharvest soybean quality for producers, inspectors, buyers and researchers.

“We’ve investigated different ways of essentially getting inside the soybean,” said LSU AgCenter engineer Kevin Hoffseth. “We found success building some machines to do the imaging, and we continue to refine and build new machines.”

The complex automated tabletop imaging system for soybean analysis is now in the patent application process. It not only depends upon quality lighting and camera technologies for image acquisition, but mechanical mechanisms to automatically move and position the soybean samples.

Additionally, this prototype includes algorithm-based, machine-learning functions built off a large database of diverse soybean images, including images of damaged and diseased ones, to consistently replicate accurate results for rapidly analyzing soybean samples.

“Along with trying to speed up the imaging in general, we’re then using different algorithms and numerical approaches to analyze it in different ways,” said Hoffseth.

“We put a lot of time into exploring imaging parameters and numerically, methodically evaluating them,” Hoffseth said. “We have been standardizing the automatic lighting and imaging for both visible spectra and infrared spectra.”

Building an effective image database of disease-damaged soybeans is crucial to the project, but it has created a challenge. Many farmers toss damaged soybean harvests, but Hoffseth and his team of researchers have gotten many of their samples from LSU AgCenter research stations throughout the state, and from helpful Louisiana farmers through assistance from the Louisiana Farm Bureau.

“One big focus is building and creating soybean grain image libraries, because you need that as the data to train new algorithms,” said Hoffseth.

AgCenter station researchers could find the soybean-analyzing system useful for their future experiments, said Hoffseth. Large scale, in-depth analysis and measurement of grain-by-grain, postharvest single grain quality for crop research purposes has not always been done, because without an efficient automated process, it would be excessively time consuming and labor intensive to complete.

“Researchers treat soybean plants to target certain pests or disease. And then, my lab can come in and say, ‘Then how does that also affect the postharvest grain quality?’” he said.

Hoffseth said he hopes the technology can help create tools that farmers, buyers and inspectors can all use to consistently analyze soybean samples, while achieving the same results, in quicker time.

“We’re making really good progress on developing fundamental tools and methods that should play critical roles in next generation automation of grain grading,” he added.

Comparisons of reflectivity of cut soybean sections at a specific infrared wavelength shows spatial difference in magnitude, connected to oil and protein content. Photo by Hobbs McAllister