Microplastics are now found almost everywhere, from the environment to drinking water and food.

Detecting microplastics in seafood is a slow, labor-intensive process that involves dissolving the fish with chemicals and examining the remaining debris piece by piece under a microscope. That makes routine monitoring for microplastics impractical for the seafood industry.

A new study led by Virginia Tech scientists used hyperspectral imaging and artificial intelligence (AI), to detect tiny plastic particles directly on the surface of fish with 95% accuracy. The study was published in the Journal of Chemometrics.

“Now we can monitor microplastic particles in real time,” said lead researcher Yiming Feng, associate professor in the College of Agriculture and Life Sciences and Extension specialist in food process engineering at the Virginia Seafood Agricultural Research and Extension Center. “Instead of sampling one fish and sending it through days of lab work, you could scan thousands of fish as they move down a production line and immediately detect potential contamination.”

Feng and his team used hyperspectral imaging, a type of camera that reads the chemical makeup of materials, to scan the surface of fish fillets for microplastics. Because plastic and fish tissue can appear almost identical using this type of imaging, the team trained an AI model to detect subtle spectral differences between plastic and fish tissue.

“What’s different here is that we’re using the material’s own chemical signature to identify it,” Feng said. “That allows us to separate plastic from fish, even when they look the same.”

To test the approach, the researchers placed small plastic particles on the surface of tilapia fillets and scanned them. The system correctly identified plastic particles 95-96% of the time, down to particles as small as 300 micrometers—approximately the size of a fine grain of salt.

"We wanted to understand the limits of the technology," Feng said. "We found that at 300 micrometers, it’s very accurate. Around 200 micrometers, it became more challenging. But even reaching that range in real time on a food surface was almost impossible in the past."

From the lab to the production line

The technology could offer a more efficient way to monitor products as they move through processing facilities.

“Many seafood processors, in Virginia and nationwide, are small businesses,” Feng said. “A single recall can be devastating. At the same time, they face labor shortages and rising costs. Smart sensing technologies like this could help strengthen food safety while reducing dependence on manual inspection.”

The technology has applications beyond microplastic detection in seafood. Feng’s lab previously worked with PepsiCo, using hyperspectral imaging to monitor oil content on potato chip production lines in real time. He said the same system could be adapted to detect metal fragments, foreign debris and other contaminants across a range of food products.

“Sensing is just the first step,” Feng said. “Once you can detect contamination automatically, you can connect that data to robotics for sorting and quality control. That’s where advanced manufacturing in food systems is heading.”

The method is not yet ready for use on processing lines. The imaging systems are expensive, and the large data files they produce take up to 10 seconds to process—still too slow for fast-moving production environments.

Feng and his team are now working to improve speed and efficiency, expand the system to different types of seafood and plastics and improve detection of smaller particles. He said detecting particles more than 1 centimeter below the surface is more challenging and may require combining this approach with other sensing technologies.

“The transition from the lab bench to the production line is where this technology will make its true impact,” Feng said. “As these imaging systems become more affordable and our algorithms faster, we can drastically reduce the industry's reliance on manual inspection, making real-time, rigorous food safety accessible to everyone.”

The research was supported by 4-VA, a collaborative partnership advancing innovation across Virginia, and the Hatch program of the US Department of Agriculture’s National Institute of Food and Agriculture.

Co-authors included Nikhita Sai Nayani, who recently completed a master’s degree in the Department of Computer Science; Ran Yang, a presidential postdoctoral fellow at the Virginia Seafood Agricultural Research and Extension Center; Lihong Yang, a doctoral student in biological systems engineering; and Lifeng Zhou, an assistant professor of electrical and computer engineering at Drexel University and 2020 graduate of Virginia Tech’s Department of Electrical and Computer Engineering.

Source: Virginia Tech