PagodaNet: Light Weight AI Model for Chinese Calligraphy Styles Recognition

The unique style of Chinese calligraphy, with its complex structures and historical evolution, presents a challenge for the public in learning and appreciating ancient Chinese characters. Besides, in the fields of textual archaeology and calligraphy education, calligraphy style recognition is the primary task. To address this issue, a dataset containing 12 calligraphy styles had been constructed, a light weight network called PagodaNet also been proposed for Chinese calligraphy style recognition. The experiment result demonstrates that PagodaNet achieves robust performance and outperforms popular models through transfer learning on the dataset. With an average recognition accuracy of 94.81% on the dataset proposed. PagodaNet can effectively recognize Chinese calligraphy styles with fewer computational resources. Furthermore, this study highlights that transfer learning with heavy weight models may not always be the most optimal approach for pattern recognition tasks. users can enhance their understanding and appreciation of ancient Chinese characters in a more efficient and effective manner with the aid by PagodaNet.