Oracle Bone Inscriptions (OBIs) are essential for understanding early Chinese civilization and preserving cultural heritage. Accurate OBI detection is fundamental to digital archiving, structural analysis, and historical interpretation. However, automated OBI detection remains challenging because of background interference, imbalanced spatial distribution, and scale variation. To better adapt detection models to OBI images, OracleDet is built around three core modules that enable structure-guided enhancement, decoupled positive-negative region attention, and adaptive multi-scale spatial-frequency modeling. Extensive experiments on the YinqiwenyuanOBI and OBIM datasets show that OracleDet outperforms general-purpose detectors and existing OBI-specific methods, particularly in degraded and densely distributed inscription scenes. Moreover, OracleDet supports practical applications in annotation correction and expert-assisted analysis, indicating its value for digital heritage preservation and OBI research.
Zhang et al. (Mon,) studied this question.