LusTra: waste type prediction through sound and accelerometer data

In recent years, advancements in waste sorting have been significantly enhanced by the integration of deep learning algorithms. In this regard, LusTra proposes a waste type recognition system using sound and accelerometer data. Two waste types are considered: Polyethylene Terephthalate (PET) bottles and Aluminium cans. Predictions through sound data, converted to Mel spectrograms, with Convolutional Neural Networks (CNN), are promising and result in an accuracy of ${8 9 \%}$ for PET waste and ${9 0 \%}$ for aluminium waste. Random forest with features extracted from the accelerometer data provide an accuracy of ${7 6 \%}$ and $86 \%$ for PET and aluminium respectively.