Classification of plastics using GMM-UBM on high-frequency data

Plastic recycling, an ever-growing application, requires a certain purity of the polymers to re-use them as secondary material. Polymers can be separated by their material characteristics determined in the terahertz range. Recent scientific interest is arising in the identification of the materials employing terahertz spectroscopy. In this paper, we investigate a large dataset of polymers recorded by a THz-spectrometer. To separate single polymers, material groups and additives we use acoustic sub-band features based on the Hilbert envelope and a fast and well-established classification method, which is also used e.g. in speaker identification. Recognition rates of 95% for single polymers, 86% for selected polymer categories and 65% for additive detection were achieved. These results are considered to be promising and we plan further investigations on this task.