An improved EfficientNetV2 with category-aware attention for ancient ceramic dating
Source abstract
Abstract
Accurate dating of ancient ceramics is a long-standing archaeological challenge constrained by two bottlenecks: extreme long-tail distribution (50.10% Qing vs 0.3% pre-Qin) and complex fine-grained classification. Conventional models often underperform in underrepresented classes and overlook nuanced features. We propose a framework integrating a category-aware attention (CAA) module and focal loss into an improved EfficientNetV2 backbone. A comprehensive dataset of 50 391 high-resolution images across eight historical periods (pre-Qin to Qing) was curated, using stratified sampling (7:1:2 split) to ensure tail-class representation (e.g., 158 pre-Qin samples). CAA directs the model toward period-specific attributes (e.g., Song celadon glaze) via category prototypes, whereas focal loss mitigates majority-class gradient bias. Our framework achieves 88.59% overall accuracy and 70.35% Macro-F1 score, outperforming the baselines (ResNet50 and ConvNeXtV2) by 5.5%–8.3% in accuracy and 12.1%–15.7% in Macro-F1. It significantly enhances tail-class robustness (e.g., F1 = 0.77 for Wei–Jin/Northern and Southern Dynasties). Ablation studies confirmed that CAA and focal loss contributed 11.2% and 9.8% to Macro-F1, respectively. This study resolves long-tail challenges in ceramic dating, supporting preservation and chronology refinement. Future research will explore multimodal fusion and few-shot learning for data-scarce cultural-heritage tasks.
Source: crossref · Quality: baseline
Research context
OpenAlex topics: Image Processing and 3D Reconstruction · Cultural Heritage Materials Analysis · Generative Adversarial Networks and Image Synthesis
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
Affiliations
School of Information Engineering, Jingdezhen Ceramic University , Jingdezhen 333403,
Funding
Science and Technology Project of the Jiangxi Provincial Department of Education · Jingdezhen Science and Technology Bureau · Graduate Innovation Fund of Jingdezhen Ceramic University
References (30)
Show source-provided references
1. A brief discussion on ‘human evidence’ and ‘material evidence’ in the field of ancient ceramic authentication
2. A brief introduction to ancient ceramic authentication methods
3. Ancient ceramic kiln identification based on convolutional neural network
4. Research on intelligent chronology of ancient Chinese ceramics based on convolutional neural network10.13957/j.cnki.tcxb.2022.01.017 ↗
5. T. Mu and F.Wang, “Research on image feature recognition of ancient ceramic shapes and patterns based on artificial intelligence,” Ph.D. thesis, Shaanxi University of Science & Technology, 2020.
6. R. Zhang and J.Wu, “Research on image recognition of bottom marks of Ming and Qing blue-and-white porcelain based on convolutional neural network,” M.S. thesis, Nanchang University, 2018.
7. Squeeze-and-excitation networks
8. CBAM: Convolutional block attention module
9. Findings on machine learning for identification of archaeological ceramics: A systematic literature review10.1109/access.2024.3429623 ↗
10. Ceramic shape classification based on cross multi-scale deep residual network
11. Identification of porcelain ewers in Tang, Song, and Yuan dynasties by digital shape characterization10.1016/j.ceramint.2023.01.011 ↗
12. Typical ceramic pattern extraction based on convolutional neural network
13. Ceramic decoration extraction method based on computer vision and image processing
14. Research on pattern and shape feature extraction methods for chronological identification of ancient pottery
15. Applications of deep learning to decorated ceramic typology and classification: A case study using Tusayan white ware from northeast Arizona10.1016/j.jas.2021.105375 ↗
16. Ancient ceramic pattern classification model based on improved E-EfficientNet
17. Smote: Synthetic minority over-sampling technique10.1613/jair.953 ↗
18. The foundations of cost-sensitive learning
19. Focal loss for dense object detection10.1109/ICCV.2017.324 ↗
20. Can deep learning identify early Chinese ceramics using only 2D images?10.3390/s26041312 ↗
21. EfficientNetV2: Smaller models and faster training
22. Prototypical networks for few-shot learning
23. Decoupling representation and classifier for long-tailed recognition
24. On the surprising behavior of distance metrics in high dimensional space10.1007/3-540-44503-X ↗
25. CosFace: Large margin cosine loss for deep face recognition10.1109/cvpr.2018.00552 ↗
26. Dynamic few-shot visual learning without forgetting
27. Deep residual learning for image recognition10.1109/CVPR.2016.90 ↗
28. ConvNeXt v2: Co-designing and scaling ConvNets with masked autoencoders
29. Swin transformer: Hierarchical vision transformer using shifted windows10.1109/ICCV48922.2021.00986 ↗
30. Class-balanced loss based on effective number of samples10.1109/CVPR.2019.00949 ↗