Can AI help tell the difference between a good and bad sweet potato?

Scientists used hyperspectral imaging to sort produce.

Researchers used a hyperspectral camera to create images of 141 potatoes and inspect their firmness and dry matter content. Llez/Wikimedia

Most grocery store patrons take for granted just what it takes to transport a humble sweet potato out of the ground and into a shopping basket. The slightly-sweet red root vegetable can come in various sizes and flavor profiles but consumers have come to expect a level of consistency. To meet that market demand, sweet potatoes are subjected to rounds of laborious and time-consuming quality assessments to root out undesirable batches that are either too firm, not sweet enough, or otherwise deemed unlikely to sell. This process is currently performed methodically by humans in a lab, but a new study suggests hyperspectral cameras and AI could help speed up that process.

In a study published this week in Computers and Electronics in Agriculture, researchers from the University of Illinois set out to see if data collected by a hyperspectral imaging camera could help narrow down certain potato attributes typically determined by manual inspectors and tests. Hyperspectral cameras collect vast amounts of data across the electromagnetic spectrum and are often used to help determine the chemical makeup of materials. In this case, the researchers wanted to see if they could analyze data from the potato images to accurately determine a spud’s firmness, soluble solid content, and dry matter content—three key attributes that contribute to the vegetable’s overall taste and market appeal. Ordinarily, this process requires tedious, sometimes wasteful testing that can include leaving test potatoes heated in a 103 degrees celsius oven for 24 hours. 

“Traditionally, quality assessment is done using laboratory analytical methods,” University of Illinois College of Agricultural, Consumer and Environmental Sciences assistant professor Mohammed Kamruzzaman said in a statement. “You need different instruments to measure different attributes in the lab and you need to wait for the results.”

The researchers gathered 141 defect-free sweet potatoes and took photos from multiple angles. Hyperspectral imaging produces torrents of data, which can be both blessing and curse for researchers looking for specific variables. To solve that problem, the researchers used an AI model to help filter down the noisy data into several wavelengths. They were then able to connect those wavelengths to the specific desirable sweet potato attributes they were looking for. 

“With hyperspectral imaging, you can measure several parameters simultaneously. You can assess every potato in a batch, not just a few samples,” Kamruzzaman added.

AI and hyperspectral cameras could speed up vegetable inspection

The researchers argue farmers and food inspectors could use their combination of hyperspectral imaging and AI to accurately and cost effectively scan sweet potatoes for key attributes while simultaneously cutting down on food waste created as a byproduct of traditional testing. And while this particular study focused on sweet potatoes, it’s possible similar tactics could be used to find desired features in a host of other vegetables and fruits as well. Kamruzzaman says he and his colleagues eventually want to create quickly and easily scan sweet potato batches. On the consumer side, the researchers envision one-day building out an app grocery store patrons could use to scan a potato and look up its particular attributes. Such an app, in theory, could cut down on patrons awkwardly fondling their produce. 

“We believe this is a novel application of this method for sweet potato assessment,” doctoral student and study lead author Toukir Ahmed wrote. “This pioneering work has the potential to pave the way for usage in a wide range of other agricultural and biological research fields as well.”

The agriculture industry is increasingly turning to AI solutions to try and ramp up efficiency and head off growing farm labor shortages. From autonomous Tulip-inspecting machines in Holland to self-driving John Deere tractors, farmers across the world are hoping these new innovations can eventually drive down food prices and increase their own profitability at the same time. How exactly that will all play out, however, remains to be seen. Agriculture gains derived from AI solutions may also take longer to benefit economically developing countries, where some farming is still done by hand.