LAUSR

Abstract. With the advancement of contemporary technology and the evolution of industrial paradigms, industrial intelligence has emerged as a critical factor in facilitating the enhancement of industrial processes. As core components in electronic manufacturing, printed circuit boards (PCBs)’ quality directly determines product reliability. However, automated defect detection faces two major challenges: (1) the coexistence of small targets and complex background textures and (2) the inherent trade-off between high-accuracy detection and real-time processing requirements. To address these challenges, we propose an enhanced lightweight framework based on YOLOv5s, featuring two technical innovations: (1) the HS-LSKA module combines hierarchical split architecture with separable large kernel attention, enabling efficient multi-scale feature fusion with 17.1% fewer parameters than standard YOLOv5s and (2) the global-efficient multi-scale attention mechanism integrates global contextual information through hybrid attention paths, which is particularly effective for small defect detection in cluttered industrial environments. Experimental results demonstrate that the improved model achieves state-of-the-art performance on multiple PCB datasets, with mAP50 of 94.8% and 98.8% on industrial and public datasets, respectively.