In-line sorting system with battery detection capabilities in e-waste using combination of X-ray transmission scanning and deep learning

Abstract

In the recycling process of e-waste, fires caused by the accidental crushing of batteries are a serious problem. Currently, the presence of batteries in e-waste is estimated based on the experience of the staff involved, which lacks speed and accuracy. To solve this problem, an in-line sorting system that can detect batteries by using a combination of X-ray scanning and deep learning was developed. The novel and unique feature of this system is its three-stage deep learning processing. First, the type of e-waste item is estimated from X-ray transmission images. Second, the batteries are detected by networks pre-trained specifically for the estimated item types. And third, batteries overlooked in the image process are detected by a follow-up network trained by a variety of situations. Through a validation study, it was confirmed that the program achieved high accuracy (0.967 for trained e-waste categories and 0.770 for untrained), surpassing a comparative program with a single deep learning network (0.902 for trained e-waste categories and 0.716 for untrained).