Computer vision and optical character recognition for the classification of batteries from WEEE

Abstract

Today, the sorting of battery packs and pouch cells from waste electrical and electronic equipment (WEEE) is performed manually as the batteries vary in shape and weight. Therefore, the feasibility to classify waste batteries by first recognizing text on an image of the battery with deep learning object character recognition (OCR), and then comparing the extracted text with a database of known text information representing battery types is investigated. In addition, the feasibility to sort beyond what is manually viable today is evaluated by examining the text IEC code on lithium-ion batteries (LIBs) to classify by cathode material. Results are promising, with an achieved precision of 0.98 and recall of 0.67 for classification by battery group, and an achieved precision of 0.83 and recall of 0.34 to classify by cathode material, demonstrating the potential for a human-machine collaboration for battery sorting.