Gold detection from printed circuit boards for eco-friendly e-waste recycling

Electronic waste is the fastest-growing waste stream in the world and contains valuable materials such as plastics and precious metals, such as gold. Gold found in electronic devices is used extensively in mobile phones, connectors, and circuit boards, and has a high recovery potential. Existing recovery processes are either performed with chemicals through manual or automated processes. However, they are error-prone in the identification of gold regions and face challenges due to the absence of standard datasets with Printed Circuit Boards (PCBs). These challenges must be addressed to develop sustainable and efficient methods for gold recovery from waste PCBs. This study presents a lightweight Region of Interest (RoI)-guided deep learning framework for detecting regions printed in gold on PCBs. A dataset of PCB images containing gold was created, pre-processed, and annotated to show the location of components with gold content. An RoI is generated by fusing local entropy with Hue, Saturation, Value (HSV) thresholding. This fused RoI map is applied as a mask over the original images to guide the detection of the required regions. The experimental results validate the robustness and effectiveness of the proposed RoI-guided YOLOv9 model through a comprehensive quantitative evaluation. The proposed model consistently achieved higher detection accuracy and better localization of gold-bearing regions than baseline methods and multiple YOLO variants, while maintaining computational efficiency. These results demonstrate the model's superior capability for the reliable and precise detection of gold components in PCB images.