Application of deep learning for symbol detection on historical maps to explore spatiotemporal changes in the regional tea industry of early 20th-century Taiwan

Abstract Focusing on the early history of the tea industry in Tamsui, Taiwan, this study uses land used data extracted from maps of the 1900s and 1920s to explore the regional characteristics and changes in the distribution of tea plantations. Map symbol detection modeling was performed using Artificial Intelligence deep learning techniques, which have been growing in the field of map research in recent years. Through the constructed symbol detection model, the land use annotation data of historical maps can be automatically retrieved for GIS-based spatiotemporal analysis. Thus, the study presents the impact of global economic panic and the failure of tea exportation in the 1920s on the local tea industry and reflects the tea plantation landscape in response strategies.