Detection and recognition of batteries on X-Ray images of waste electrical and electronic equipment using deep learning

Abstract The trend of increased use of lithium-ion batteries, challenges the cost-effectiveness and safety of manual battery separation during the end-of-life treatment of Waste Electric and Electronic Equipment (WEEE). Therefore, the need for novel techniques to separate and sort batteries from WEEE is increasingly important. For this reason, the presented research investigates the potential to facilitate the development of novel techniques for battery extraction and sorting by examining the technical feasibility of predicting the presence, location, and type of batteries inside electronic devices with a deep learning object detection network using X-Ray images of the internal structure of WEEE. To determine the required X-ray imaging parameters, 532 electronic devices were arbitrarily collected from a recycling facility. From each product, two X-Ray Transmission (XRT) images were captured at two different X-Ray source configurations. Results obtained with the limited dataset are promising, demonstrating a 91% true positive rate and only a 6% false positive rate for classifying battery-containing devices. Moreover, a precision of 89% and a recall of 81% are demonstrated for battery detection, and an average precision of 85% and an average recall of 76% are demonstrated to distinguish amongst the following six battery technologies: cylindrical nickel-metal hydride or nickel-cadmium, cylindrical alkaline, cylindrical zinc-carbon, cylindrical lithium-ion, pouch lithium-ion, and button cell batteries. These results demonstrate the potential of using deep learning object detection on XRT-generated images for both automated battery extraction and sorting, regardless of the condition or shape of the products.