Cultural element recognition in ancient Chinese poetry using Pre-Trained Models

This work addresses the limitations of traditional methods that fail to comprehensively capture the multidimensional cultural features of ancient Chinese poetry. It introduces the Multi-Head Attention (MHA) mechanism and Contrastive Language-Image Pre-training (CLIP) algorithm, proposing an optimised method based on the Pre-Trained Model (PTM) to enhance the accuracy of cultural element recognition in ancient poetry. The results indicate that the PTM-based optimised model achieves a modality alignment consistency of 0.92 in integrating text and calligraphy images, demonstrating strong capabilities in combining image and text information. The alignment consistency between calligraphy images and video materials reaches 0.89, showcasing good cross-modal alignment ability. The overall alignment consistency of the optimised PTM is 0.90, indicating high accuracy in processing multimodal information. Furthermore, the optimised PTM achieves a detail retention rate of 0.92 and a detail description completeness of 0.89. Moreover, its detail-capturing ability and presentation consistency exceed 0.92, reflecting its superior performance in extracting and presenting complex cultural details. This work aims to enhance the recognition and accuracy of artistic elements in ancient poetry, thereby promoting the digital preservation and application of cultural heritage. The work provides more precise analytical tools for educators and cultural heritage institutions, supporting the in-depth study and widespread dissemination of ancient poetry in modern society.