Researchers from the University of Illinois suggest that a combination of artificial intelligence (AI) and hyperspectral cameras could automate quality assessments of sweet potatoes.

To ensure the quality of sweet potatoes, humans in the lab typically subject them to rounds of time-consuming assessments to sort out undesirable batches according to characteristics including firmness, sweetness and appearance, among other qualities.

Source: Llez

To expedite such assessments, the researchers from the University of Illinois examined whether data collected using a hyperspectral imaging camera might identify specific potato attributes that are traditionally determined by manual inspectors and tests.

The researchers explained that hyperspectral cameras — which can be used to gather large amounts of data across the electromagnetic spectrum to help determine the chemical makeup of certain materials — could be used to accurately analyze data from the potato images. The data could then be used to identify the three key attributes contributing to the vegetable’ taste and market appeal — namely its firmness, soluble solid content and dry matter content. Done manually, this process tends to be tedious and wasteful, the researchers explained.

To test their automated approach, the team gathered 141 defect-free sweet potatoes and took photos of them from assorted angles. As the images are captured, hyperspectral imaging created a deluge of data. The researchers explained that an AI model was subsequently used to filter out the noisy data into different wavelengths and those wavelengths were then linked to the desired sweet potato attributes.

In addition to accurately and cost effectively scanning sweet potatoes for those key attributes while also reducing food waste created as a byproduct of traditional testing, the team suggests that the combination of AI and hyperspectral imaging could also be used for the same purpose on other vegetables and fruits.

An article detailing the technology, “Advancing sweet potato quality assessment with hyperspectral imaging and explainable artificial intelligence,“ appears in the journal Computers and Electronics in Agriculture.