AI-driven detection of metal contaminants in construction and demolition wood waste for enhanced valorisation

Wood waste contaminated with metal residues presents major challenges for its valorisation, hindering reuse, recycling, and recovery. Conventional metal removal methods, such as magnetic separation after shredding, are ineffective when the dimensional integrity of wood is required, necessitating manual inspection and removal with tools, which is a labour-intensive and time-consuming process. As computer vision technology advances automation, it is promising to explore its capability to detect metal contaminants in wood waste, enabling efficient removal through precise localisation. Accordingly, this study evaluates state-of-the-art, edge-deployable object-detection models in two categories: axis-aligned detectors (CNN-based: YOLOv11, YOLOv12; and transformer-based: RT-DETR) and oriented bounding-box detectors (YOLOv8-OBB, YOLOv11-OBB) for localising metal contaminants in real time. Under axis-aligned models, RT-DETR records a higher performance over metal contaminant detection, while YOLOv11-OBB performed well under oriented bounding box models. Since both top models perform remarkably across the dataset from different perspectives, selecting one model is inefficient. Therefore, detection fusion was conducted by integrating the detection outputs from both top models along two paths: box confidence and oriented bounding box prioritising. Overall, detection fusion strategies increased mAP@50 to about 70% and mAP@50:95 to approximately 53%, surpassing individual models by identifying metal contaminants that each model had missed. Therefore, "Oriented Bounding Box Prioritising" configuration was identified as the best AI-driven approach for detecting metal contaminants, as it accurately localises contaminants at the correct angle, minimising damage to surrounding wood during removal. This provides the AI-driven localisation prerequisite for dimension-preserving removal, enabling the valorisation of metal-free wood waste.