Electrostatic separation, which is one plastic sorting process, has the problem of irregularity in recovery rates and purity due to fluctuations in the raw material composition ratio.In this study, we investigated a method for real-time estimation of the plastic composition ratio using a near-infrared hyperspectral camera, with the aim of enabling automatic control of sorting conditions based on material variability.We investigated data preprocessing and machine learning methods to identify plastic types from near-infrared spectra for the three main types of plastics (ABS, PS, PP) that are used as raw materials for electrostatic separation in home appliance recycling, and achieved identification results with 99 % accuracy.By applying this method to hyperspectral imaging data and taking into account differences in shape and dark color ratio between each plastic type, we were able to estimate the composition ratio of each plastic type with an accuracy within 1.2 pt of the true value (weight ratio) .
Kuroda et al. (Thu,) studied this question.