Site icon Visionify

How can Machine Learning detects Microplastic in Ocean?

How can Machine Learning detects Microplastic in Ocean

When you put a plastic bottle or plastic bag in the recycling bin you expect it to be recycled and turn into a new plastic bottle. That is how many people think. But the reality is that plastic does not actually break down fully into new materials. And for certain kinds of plastics this means that tons of tiny plastic fragments can escape from recycling plants into rivers, oceans and even the air. Those pieces accumulate in landfills and pollute soils, groundwater, wildlife and human food supplies. So it comes as no surprise that bottled water results in far more microplastic waste than other drinks like tea, coffee or wine. 

Microplastic which is plastic debris that is less than five millimeters long has become a major environmental problem. Changing the way fish and other animals look throughout the world’s oceans. In our blog we will go through what is microplastic and how ML is used to detect microplastic.

What is Micro Plastic?

Microplastic is the primary plastic polluting oceans. The term refers to any piece of plastic less than 5 mm (0.2 inches) in size. There are many ways the plastics get into the ocean and once they are there some are eaten by fish or birds which we then eat. Microplastics do not biodegrade but can be broken into nanoparticles. As these get consumed by marine life they move up the food chain. Eventually we could be ingesting them ourselves when we eat fish.

Ways to detect microplastic

To assist researchers in identifying microplastic prediction analyses have devised. Many of these methods are based on most of two types of machine learning: model-based or instance-based. Model-based approaches start with a series of examples labeled spectroscopic reference data used to develop the statistical model. Researchers then use the model to anticipate properties of novel spectra (polymer or matrix). Instance-based algorithms allocate microplastic samples to related groups by using collections of known spectra, referred to as instances in machine learning. Spectral library searching is a popular method. Researchers examine a spectrum and search for similar ones in a database.

Both types of machine learning models have advantages and disadvantages. In general instance-based learning has the advantage of enhancing or adjusting spectroscopic reference data by switching the reference spectra in the library. On the other hand model-based machine learning obviously beats that in terms of analytical speed. 

Another model-based machine learning strategy for interpreting FPA-based hyperspectral images uses random decision forest (RDF) classifiers to create various polymers models. This model already detects 20 different polymers. And taught to be relevant to a wide range of matrices, including soil, sludge, and air, compared to existing model-based learning approaches for FPA-based hyperspectral imaging. 

What happens after detection?

Particles can detected by scanning for pixels that belong to the same polymer class and connecting them into lines. For each particle we allot distinct ID and is classified according to geometric attributes such as length, breadth, image resolution, volume, and direction. The classification value reflects the classification’s reliability.

The particle detection and characterization model can depicted in the image, which includes a list of individual particles and a list of total particle counts per class. In addition each class’s MPs can highlighted in a distinct color.

You can utilize the particle editor to evaluate and revise any MPs obtained from the previous analysis procedure. Achieved by comparing the averaged spectrum of every particle to a database of known polymer reference spectra. You can also manually add or remove pixels.

Finally once the MP list has formed and can downloaded as a CSV file to use in other software for postprocessing or data visualization.
Check this out Opportunities of Artificial Intelligence in Redefining Food Inspection Industry


Artificial intelligence could help solve the problem of microplastics. Machine learning algorithms can accurately detect the minuscule plastic particles and provide an accurate image of how they dispersed. ML has a wide array of applications. ML can applied to many different fields using vast amounts of data to help predict future trends and improve efficiency by making better decisions. AI could be one solution to detecting microplastic particles accurately.

Machine vision solutions are a reliable solution to many of the problems in manufacturing. Among the many things we provide our clients is a proven way to streamline their inventory process, saving them time and money while keeping quality uncompromised. Call us to get a demo.

Exit mobile version