Innovative pathways for contemporary expression of landscape imagery in classical Chinese poetry: A deep learning-based automatic recognition model

Classical Chinese poetry provides a critical basis for revealing historical landscape characteristics and for understanding aesthetic concepts in premodern China. Effectively extracting landscape imagery information from such poems is therefore a prerequisite and key step for guiding the spatial production of poetic ambience. To address the limitations of existing text-mining approaches in terms of both recognition efficiency and accuracy, this study focuses on Zhuzhici (bamboo-branch poems), which is related to the Jiangnan region, and introduces deep learning techniques to develop an automated landscape imagery recognition model tailored to the cultural specificity and linguistic nonstandardness of classical Chinese poetry. The results indicate that the proposed DA-BERT-BiLSTM-CRF model achieves a precision of 87.87%, a recall of 83.05%, and an F1 score of 85.39%, significantly improving the recognition accuracy of landscape imagery elements, including those expressed in single-character forms, functioning as geographically specific proper names, and embedded with implicit emotional connotations or behavioural activities. Furthermore, this study explores potential application pathways of the model for the digital inventory and in-depth mining of historical landscape resources, the refined construction of a place-based landscape imagery knowledge base, and the rapid collaborative reconstruction of cross-modal scene information. These efforts broaden the avenues through which landscape-imagery theories can be translated into spatial planning practices, thereby supporting the creation of poetic dwelling environments.