Evaluation of practical edge computing CNN-based solutions for intelligent recycling bins

Abstract

Rapid economic growth has given rise to the urgent demand for more efficient waste recycling systems. An innovative smart recycling bin is proposed that automatically separates urban waste to increase the recycling rate. Over 1800 recycling waste images were collected and combined with an existing public dataset to train neural network classification models for two embedded systems, one incorporating a Jetson Nano and the other a K210 unit. The model developed reached an accuracy of 93.99% on the Jetson Nano and 94.61% on the K210. A user interface application was also designed to collect feedback from users during their interaction with the smart bin. In terms of power consumption, the system employing the Jetson Nano consumed 4.7 W, representing a 30% reduction in power consumption compared to previous work, while the K210 required just 0.89 W of power to operate. In summary, our work demonstrated a small-scale, fully functional prototype of an energy-efficient, high-accuracy smart recycling bin, with the potential of commercialisation for the purpose of improving urban waste recycling.