Micro-plastics dispersion in water is one of the major global threats due to the potential of plastic items to affect the food chain and reproduction of marine organisms. However, reliable and automatic recognition of micro-plastic in water is still an unmatched goal. Here we identify micro-plastics in water samples through digital holography microscopy combined to machine learning. We exploit the rich content of information of the holographic signature to design new distinctive features that specifically characterize micro-plastics and allow distinguishing them from marine plankton of comparable size. We use these features to train a plain support vector machine, remarkably improving its performance. Thus, we obtain a very accurate classifier using a simple machine learning approach, which does not require a large amount of training data and identifies micro-plastics of various morphology and optical properties over a wide range of characteristic scales. This is a first mandatory step to develop sensor networks to map the distribution of micro-plastics in water and their flows.
High-accuracy identification of micro-plastics by holographic microscopy enabled support vector machine
Bianco V;Memmolo P;Merola F;Distante C;Ferraro P
2019
Abstract
Micro-plastics dispersion in water is one of the major global threats due to the potential of plastic items to affect the food chain and reproduction of marine organisms. However, reliable and automatic recognition of micro-plastic in water is still an unmatched goal. Here we identify micro-plastics in water samples through digital holography microscopy combined to machine learning. We exploit the rich content of information of the holographic signature to design new distinctive features that specifically characterize micro-plastics and allow distinguishing them from marine plankton of comparable size. We use these features to train a plain support vector machine, remarkably improving its performance. Thus, we obtain a very accurate classifier using a simple machine learning approach, which does not require a large amount of training data and identifies micro-plastics of various morphology and optical properties over a wide range of characteristic scales. This is a first mandatory step to develop sensor networks to map the distribution of micro-plastics in water and their flows.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.