FAA-YOLO: frequency-augmented attention mechanism for oracle bone inscription detection

This paper presents FAA-YOLO, a novel YOLOv8s enhancement framework for detecting archaic characters in oracle bone rubbings. Our method addresses two key challenges in this domain: the high-frequency stroke patterns characteristic of ancient scripts, and the low-contrast backgrounds commonly found in historical artifacts. The proposed architecture integrates wavelet transform-based frequency analysis with a dual-domain attention mechanism, creating synergistic spatial-frequency feature representations. Specifically, we develop a Shuffle Frequency Attention (SFA) module that concurrently processes spatial features and their wavelet-transformed frequency components through parallel attention pathways. Evaluations on the OBIMD benchmark demonstrate 5.7% higher [email protected] compared to baseline YOLOv8s.