SHNet architecture for detecting small defects on printed circuit boards. Reproduced from the authors' paper.
METHOD / FINDINGS
SHNet
A lightweight PCB defect detector that preserves shallow high-resolution features through multi-scale adaptation, semantic alignment, and wavelet enhancement.
A YOLOv8s-based architecture removes the low-resolution detection head and retains deeper features as semantic guidance. MKAM provides scale-aware features; SSAM aligns shallow spatial detail and deeper semantics; WEM strengthens weak defect cues in the frequency domain.
2.3 MB model with 0.97M parameters. Reported model artifact size and parameter count, consistent across the two evaluated PCB datasets. (Table 4, p. 8.)
A zero-shot template-matching framework that combines visual foundation models, aspect-ratio-preserving normalization, cross-attention, and geometric refinement across image scales.
