In this work, we present a proof of concept study of multi-layer thematic map generation. A novel framework is developed to extract a multi-layer thematic map that better represents urban land cover distributions. The Swin transformer obtains the preliminary semantic segmentation result for the tree and building; and Dense-U-Net-121 for the road. Subsequently, we recover the hidden information separately from three main urban semantic classes, namely building, tree, and road. The proposed workflow brings possibilities to an improved urban morphological representation.