OracleDet

OracleDet is a lightweight and efficient object detection framework designed for Oracle Bone Inscription (OBI) detection in complex cultural heritage scenes. Built upon YOLOv11, OracleDet introduces three task-specific modules: Structure-guided Feature Enhancement Module (SFEM), Positive–Negative Region Attention (PNRA), and Multi-Scale Spatial-frequency Module (MSFM). These components improve robustness against background interference, dense and sparse inscription distributions, and scale variations. Experiments on the YinqiwenyuanOBI and OBIM datasets demonstrate that OracleDet achieves strong detection accuracy while maintaining a compact model size, making it suitable for digital heritage preservation, annotation correction, and expert-assisted oracle bone research.

Project Overview

This project contains three core modules specifically designed for Oracle Bone Inscription detection tasks. These plug-and-play modules can be integrated into any model to enhance feature extraction and attention mechanisms in object detection networks:

• SFEM (Structure-guided Feature Enhancement Module) - Structure-guided feature enhancement module
• PNRA (Positive-Negative Region Attention) - Positive-negative region attention mechanism
• MSFM (Multi-Scale Spatial-frequency Module) - Multi-scale spatial-frequency module

Datasets

YinqiwenyuanOBI Dataset

Dataset download: https://jgw.aynu.edu.cn/home/down/index.html

OBIM Dataset

The OBIM dataset is a multi-modal Oracle Bone Inscription dataset that provides comprehensive and high-quality annotations, including bounding boxes, character categories, transcription content, corresponding inscription groups, and reading order within groups for each oracle bone character group.

Citation:

@article{li2024oracle,
  title={Oracle bone inscriptions multi-modal dataset},
  author={Li, Bang and Luo, Donghao and Liang, Yujie and Yang, Jing and Ding, Zengmao and Peng, Xu and Jiang, Boyuan and Han, Shengwei and Sui, Dan and Qin, Peichao and others},
  journal={arXiv preprint arXiv:2407.03900},
  year={2024}
}

YOLO Format Dataset Download:

We have processed the OBIM dataset to retain only the object detection-related parts and converted it to YOLO format.

• Shared file: OBIM_YOLO.rar
• Link: https://pan.baidu.com/s/1TUxmHzUqwBBGsAFSjotN-w?pwd=ZCZC
• Extraction code: ZCZC

Visualization

Detection Results

Citation

If this project is helpful for your research, please cite:

DOI: 10.1038/s40494-026-02621-5

@article{oracledet2026,
  title={OracleDet: Structure-Guided and Spatial–Frequency Aware Detection of Oracle Bone Inscriptions in Complex Heritage Scenes},
  author={Cheng Zhang，Cheng Xu，Yuxuan Liu，Yu Yang，Lu Chen，Jiazheng Yuan，Bofeng Mo},
  journal={npj Heritage Science},
  year={2026},
  doi={10.1038/s40494-026-02621-5}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contact

If you have any questions or suggestions, please contact us through:

• Submit a GitHub Issue
• Email: 935894505@qq.com