Near-infrared hyperspectral circular polarization imaging and object classification with machine learning

Abstract

We constructed a hyperspectral circular polarization (S3) imaging system in the near-infrared (NIR) region comprising a circularly polarized broadband light source, a polarization grating, and a commercial hyperspectral camera. With this system, we captured hyperspectral S3 images of plastic samples. We then demonstrated the classification with machine learning and found that the hyperspectral S3 images showed higher classification precision than the conventional NIR hyperspectral images. This result indicates that the hyperspectral S3 imaging has potential for object classification even for samples with similar absorption spectra. This hyperspectral S3 imaging system can be applied in garbage classification in recycling plants.