Extraction of Tree Symbology from Historical Maps Using Machine Learning Techniques
2023
Haichuan, Wang | William, Mackaness
The objective of this study was to employ Convolutional Neural Network (CNN) in order to extract tree canopy cover from a historical Ordnance Survey map of Edinburgh dating back to the 1890s. Through the utilisation of a well prepared dataset, a model was constructed that showed proficiency in the detection and classification of trees across diverse map segments. The model demonstrated optimal performance after undergoing training for a duration of 30 epochs. It revealed a noteworthy level of expertise in accurately identifying trees within unambiguous situations. Nevertheless, there were difficulties encountered in the precise identification of faint tree symbols and the differentiation between trees and other features on maps, especially in areas with textual labels or rugged topography. The undertaking, notwithstanding its efficacy in comparison to conventional approaches, brought attention to aspects that require more improvement and optimisation.
显示更多 [+] 显示较少 [-]