Multi-scale directional attention for AI-based WPCB component detection: A lightweight vision approach toward automated e-waste recycling

High-precision component identification on waste printed circuit boards (WPCBs) is crucial for e-waste value assessment. However, complex backgrounds, dense distributions, and directional structures pose boundary localization challenges for existing vision models. This study proposes a lightweight, directionally-enhanced object detection method for industrial WPCB sorting. By integrating multi-angle structural priors into a one-stage detection framework, the model enhances directional feature extraction at a minimal computational cost. Additionally, a multi-scale feature enhancement and progressive residual fusion strategy ensures training stability under complex conditions. Evaluated on a hybrid industrial-grade dataset of 13 core component categories, the optimized model achieves an mAP 50-95 of 89.44%. Feature space analysis confirms reduced prediction variance and improved inference stability. With millisecond-level inference speed, this robust method meets the real-time, high-throughput demands of e-waste recycling, facilitating automated non-destructive dismantling and precise pricing.