May 5, 2021Machine vision system created for almond grading and safety
Researchers at the University of South Australia (UniSA) have developed a world-first automated technique for simultaneously grading almond quality and detecting potentially serious mycotoxin contamination in kernels.
In 2019-2020, Australia’s almond crop was worth just over $1 billion AUD, and the value of the sector is expected to expand to $1.5 billion AUD in coming years, with Australian almond growing conditions among the best in the world.
Given the local industry is now exporting to more than 50 nations, accurate and consistent grading of almonds is paramount, ensuring international markets can trust the Australian product.
Traditionally, almonds have been graded manually, with samples taken hourly from production lines to check for consistency of appearance, chips and scratches, double kernels, insect and mold damage, and other defects.
This process, however, is labor intensive, slow, and subjective, all of which can lead to inaccurate and inconsistent grading, particularly from season to season due to staff turnover.
In conjunction with industry partner SureNut, researchers at the University of South Australia have developed a machine that dramatically improves the accuracy of almond grading, in addition to detecting potentially fatal contaminants common in almond kernels.
Funded through the Cooperative Research Centres Projects program, a research team led by Associate Professor Sang-Heon Lee combined two high definition cameras, a hyperspectral camera and purpose developed AI algorithms to create a system that can examine almond quality in far greater detail than the human eye.
The system can accurately assess physical defects such as chips and scratches and detect harmful contaminants, including the presence of aflatoxin B1, a potent carcinogen that may be implicated in more than 20% of global liver cancer cases.
“Our goal with this innovation was not to simply replicate what a human being could do, but to go far beyond that,” Lee said.
“So, in respect to physical appearance, this machine can detect defects more quickly and more accurately than manual grading, and by using two high definition cameras and a transparent viewing surface, it can also view both sides of the nut simultaneously.”
While this visual functionality alone puts the SureNut system at the forefront of innovation in this field, the addition of the hyperspectral camera for contamination detection is a ground-breaking world first.
UniSA engineering researchers, Wilmer Ariza and Gayatri Mishra, developed the hyperspectral system used on the SureNut machine, and Ariza says almonds presented a unique challenge for the technology.
“We are the first team to successfully use hyperspectral imaging this way with almonds, even though other researchers have tried,” Ariza says.
“Certain characteristics of the nut kernels – which we are keeping a secret – made the process extremely difficult, but we overcame that, and now we can deliver highly accurate analysis using hyperspectral imaging.
“Through the process, we also discovered some new information about hyperspectral imaging in general, which we will share with the wider research community in time.”
Thanks to this hyperspectral innovation, the SureNut system can monitor four key indicators in almonds – moisture content; free fatty acid content (FFA) and peroxide value (PV), which are associated with rancidity; and aflatoxin B1 content.
“Moisture, FFA, PV and aflatoxin B1 content were all correctly predicted by the developed model with accuracy of 95%, 93%, 91% and 94%, respectively,” Lee says.