Deep-learning-assisted near-infrared hyperspectral imaging for microplastic classification

Microplastics are small plastics with a size between a few microns and about 5 mm. Due to their small size, microplastics can be ingested by living organisms including humans, which has become a global concern and a heated area of research. To detect and characterize microplastics, various methods have been used, among which Fourier transform infrared (FTIR) spectroscopy and Raman spectroscopy offer nondestructive solutions. In this study, using deep-learning-assisted hyperspectral imaging (HSI) in the near-infrared (NIR) range of 1100-1650 nm, we demonstrate high-throughput, nondestructive classification of microplastics. Because NIR light is barely absorbed by most plastics and highly scattered by small particles, NIR-HSI has mostly been used for microplastics larger than 100 mu m. Using deep learning in combination with Fourier transform spectroscopy, here we show NIRHSI can classify microplastics in the 10-100 mu m range with an accuracy over 99 % and at a speed much faster than FTIR or Raman spectroscopy. The demonstrated method offers a new solution for high-throughput detection and classification of microplastics.