Period: Mar 26, 2022

Team: Jeongmin Yoo, Yeeun Sohn, Yooha Bae, Byungkon Kang

Publication Description: International Conference on Industrial Application Engineering 2022 at Matsue, Japan

Abstract: Recycled transparent PET bottles have the highest added value among plastics, but in South Korea, only 10% of PET bottles are recycled with high quality. Therefore, the South Korea Ministry of Environment planned to impose a new policy that makes it mandatory to separate clear PET bottles. We propose Clean Bottle, which is a reverse vending machine that helps the new policy via image recognition so that only clean PET bottles without seals and content are accepted. The vending machine uses a convolutional neural network to determine whether the given bottle is a pass or fail case. We built a prototype of this machine, designed by using a Raspberry Pi 4B and PiCamera. The testing accuracy of the model is over 99% and is able to correctly classify new data of PET. In addition, we also study various reduced-size models that could be helpful when porting to inexpensive embedded systems.