Classification of plastic materials using machine-learning algorithms and microwave resonant sensor

Abstract

The identification of different plastic materials in pellet form using a microwave negative-order-resonance sensor along with the evaluation of classification algorithms from machine learning is presented in this paper. Operation of the sensor is within the unlicensed ISM 2.5 GHz band, and identification of the materials is based on the measured resonant parameters from the sensor. Several classifiers are used to process the resonant parameters having uncertainty factors involved in pellet measurements (air gaps, pellet positions, dimensions and shapes), and performance comparison between the algorithms is carried out in terms of accuracy in the classification. Moreover, the presented measurement method is proposed as a fast, non-destructive, and low-power consumption way to identify plastic raw materials using a low-profile circuit having a high potential of being used in industrial processes.