VELIKIY NOVGOROD, September 7. /TASS/. Scientists of the Novgorod State University and the Russian Academy of Sciences' Institute of Oceanology created the world's biggest open database of infrared spectra of microplastics from the Eastern Arctic waters - Microscan. By using it, neural networks will be able to distinguish with an accuracy of more than 96% polyethylene from polypropylene right in the ocean, and environmentalists will have an accurate tool for counting plastic debris, leader of the Polymer Modeling Laboratory at the University's Microplastic Research Center Sergey Lyulin told TASS.
"The database contains infrared spectra of 2,010 real particles ranging in size from half a millimeter to five millimeters," the expert said. "Right now, it features micro particles' spectral data of polyethylene, polypropylene, and polystyrene - the most common synthetic polymers. The database's starting material was micro plastic particles found in surface waters in the White, Barents, Kara, East Siberian Seas and the Laptev Sea in 2020-2021."
Microplastics are tiny particles of ordinary plastic that can remain in the environment for quite a long time. They enter the water as a result of decomposition of plastic debris, car tire wear, washing synthetic clothing, destruction of fishing nets, and from other sources. Polymers, being of different density, behave differently in water: for example, polyethylene floats up, while other types may sink and remain on the ocean floor for a long time.
Until recently, researchers faced two major problems. Firstly, most literature sources contain information about "ideal" granules of pure plastic. However, plastics age in the sea, bacteria settle on it, plastics become overgrown with a "biocorona", crack in the sun, and the properties change. Therefore, at first, micro particles found in the sea must be identified - whether they are pieces of plastics or of some other particles. Secondly, there are not many bases of microplastics infrared spectra, and usually they are much smaller. In order to speed up the process of identifying microplastics, it is necessary to train neural network models, and the quality of their "work" significantly improves with an increase in the size of databases used for training. According to the scientists, the developed Microscan database is not only the biggest in the world. It is the first database on microplastics in the Arctic.
The solution
Each found particle was examined in infrared light without any additional chemical purification. The identification results were verified by using three different approaches: spectral matching, artificial intelligence, and expert control: experienced chemists rechecked manually all disputed spectra.
The database may be called a "universal training tool for machine learning." It allows algorithms to learn how to work with real particles "from nature", without additional chemical treatment, which is extremely important in the field. Using models, trained on such a database, will significantly speed up identification right during the expedition.
The research was carried under a mega grant, dubbed "New approaches to solving the problem of microplastics as a potential threat to humans and the environment."