Classifying polymers with mid-IR spectra and machine learning: From monitoring to detection

Monitoring and identifying environmental microplastics is of great importance for the scientific world, environmental agencies, and water authorities, to estimate the environmental impact and increase efforts to decrease emissions. As one of the infrared spectroscopy techniques, Laser Directed Infrared (LDIR) imaging can observe various microplastics, in terms of spectroscopical signals. Such signals are useful for follow-up analyses, particularly, identification by machine learning (ML) algorithms. Based on medium or large-sized datasets, past studies applied a variety of ML models to detect microplastics from their LDIR spectra. To tackle it, we first propose a practical data augmentation technique to generate synthetic samples when only a few samples are available. Then a comprehensive comparison of multiple models, including both machine learning and deep learning models, is presented. Our results show that the ensemble ML model, compared to neural network models, can take the least training time to achieve the best performance, i.e., a classification accuracy of 99.5%, even with a small dataset (210 samples collected from aquatic systems). This study provides a generic framework for monitoring and detecting microplastics by combining LDIR and ML.