Improved Glass Composition Analysis and Identification of Cultural Heritage with Limited Data Using Data Augmentation and CatBoost

Hu, X., Zhang, J., Wang, Z. and ZHAO, J. (2024) Improved Glass Composition Analysis and Identification of Cultural Heritage with Limited Data Using Data Augmentation and CatBoost. In: 5th International Conference of Artificial Intelligence Technologies and Applications, Changchun, Jilin, China, 30 Jun-02 Jul 2023, pp. 735-748. (doi: 10.3233/faia231369)

Abstract

Glass artifacts play a significant role in cultural heritage, offering valuable insights into ancient craftsmanship and cultural exchange. However, accurately analyzing and identifying ancient glass objects presents challenges due to limited data. This study aims to enhance the analysis and identification of glass compositions in cultural heritage by employing data augmentation techniques and the CatBoost prediction model. Firstly, data augmentation techniques are applied to expand the limited dataset, increasing sample quantity and diversity to improve the model’s generalization capability. The TOPSIS method is employed to comprehensively evaluate different augmentation factors and select the most suitable ones. Subsequently, the CatBoost prediction model is utilized, and the model parameters are optimized using a random search method to further enhance predictive performance. Experimental research on ancient glass artifacts validates the effectiveness and feasibility of the proposed methods. The final model demonstrates high predictive performance and a good fit on the training set, cross-validation set, and test set. For example, when predicting the sodium oxide content before weathering in glass artifacts, the average R-squared(R2) reaches 0.998, and the Mean Squared Error(MSE) is 0.003. These results signify the accurate prediction of glass artifact compositions and the model’s stable predictive capabilities across different datasets. Utilizing the predicted chemical composition, the identification of glass artifacts achieves a classification accuracy of 100%, indicating the excellence of the model. In conclusion, this study presents an improved approach for analyzing and identifying glass compositions by overcoming the limitations posed by limited data through data augmentation and the CatBoost model. These advancements provide valuable tools and methods for preserving and researching cultural heritage, contributing to the progress of ancient civilization studies and technological development.

• Item Type: Conference Proceedings
• Additional Information: Published in Proceedings of the 5th International Conference (ICAITA 2023), Frontiers in Artificial Intelligence and Applications Vol. 382: 735-748.
• Status: Published
• Refereed: Yes
• Glasgow Author(s) Enlighten ID: ZHAO, Jiahui
• Authors: Hu, X., Zhang, J., Wang, Z., and ZHAO, J.
• College/School: College of Science and Engineering > School of Engineering
• Journal Name: Frontiers in Artificial Intelligence and Applications
• Publisher: IOS Press
• ISSN: 1879-8314
• Copyright Holders: Copyright: © 2024 The Authors
• First Published: First published in Proceedings of the 5th International Conference (ICAITA 2023), Frontiers in Artificial Intelligence and Applications Vol. 382: 735-748
• Publisher Policy: Reproduced under a Creative Commons licence