Automatic classification of packaging waste at source using acoustic technology

Abstract

Sorting packaging waste at the source is key to effective recycling. Reverse vending machines (RVMs) can encourage recycling by offering incentives, but they rely on costly sensors like barcode scanners and computer vision, which face issues such as unreadable barcodes, high computational demands, and lighting sensitivity. This paper proposes an alternative approach using acoustic technology. The idea is to use a parametric array loudspeaker (PAL) to emit ultrasonic and audible sound waves towards the recyclable object, and by measuring its interaction, classify the item with the help of machine learning or deep learning. The exponential sine sweep method is used to measure the ultrasonic and audible impulse response of each item, creating a dataset for various containers. Classical and deep learning models are trained to classify items into categories like plastic, glass, cardboard, and metal. The system was tested in a controlled environment featuring a scaled replica of a reverberation chamber, with an omnidirectional parametric loudspeaker (OPL) serving as the sound source. Preliminary results demonstrate high classification accuracy, highlighting the potential of acoustic-based methods to improve the accessibility and efficiency of RVMs in promoting recycling initiatives