Generative adversarial network application for cartographic heritage translation to satellite images of Java, Indonesia

This research examines the application of the Conditional Generative Adversarial Network (cGAN) on historic map datasets of Java, Indonesia. The procedure is divided into four stages: input, pre-processing, processing, and output. The observed data are historical maps from various years of creation from the 1500s to the 1900s. The method used is Pix2Pix which aims to translate the cartographic archives into satellite images. The map sample size has a resolution of 256 × 256 pixels. The generator adopts the U-Net architecture, while the discriminator adopts a patch-based fully convolutional network. The results of the translation are evaluated in both quantitative and qualitative, highlighting that different cartographic characteristics over time, such as map styles, geographic information, scales, and creators vary their response to built models in generator and discriminator performance. Furthermore, this research is expected to facilitate the creation of a database of translated geographic information from historical maps towards satellite imagery as well as authorising the leverage of cartographic heritage collections as part of archival resources to create sustainable cultural heritage in the future.