This paper describes an innovative computational approach for comparing old maps. Maps older than 20 years remain a vast treasure of geographic information in many parts of the world with potential applications in many environmental and social analyses, e.g., establishing road construction over the past 80 years or identifying settlement growth since the middle ages. Semantic segmentation has developed into a viable computational method for analysing old maps from previous centuries. It allows for the discrete identification of elements, e.g., lakes, forests, and roads, from cartographic sources and their computational modelling. Semantic segmentation uses convolutional neural networks to extract elements. With this technique, we create a computational approach to compare old maps systematically and efficiently.