Abstract
Abstract. Archive topographical maps are a key source of geographical information from past ages, which can be valuable for several science fields. Since manual digitization is usually slow and takes much human resource, automatic methods are preferred, such as deep learning algorithms. Although automatic vectorization is a common problem, there have been few approaches regarding point symbols. In this paper, a point symbol vectorization method is proposed, which was tested on Third Military Survey map sheets using a Mask Regional Convolutional Neural Network (MRCNN). The MRCNN implementation uses the ResNet101 network improved with the Feature Pyramid Network architecture and is developed in a Google Colab environment. The pretrained network was trained on four point symbol categories simultaneously. Results show 90% accuracy, while 94% of symbols detected for some categories on the complete test sheet.
Highlights
Historical maps contain vast amounts of geographical information, which can help us understand environmental and physical changes that happened over past ages
Iosifescu et al proposed a vectorization methodology for extracting areal and linear features using binary image segmentation and vectorizing tools provided by the open source GDAL and OGR libraries (Iosifescu, Tsorlini, and Hurni 2016)
Quan et al focused on point symbol recognition combined with image segmentation on a pretrained Convolutional Neural Networks (CNN) of AlexNet architecture and achieved 98.97% accuracy on single test images (Quan et al 2018)
Summary
Historical maps contain vast amounts of geographical information, which can help us understand environmental and physical changes that happened over past ages. These maps, are usually in printed form and do not have a vector geo-data representation, which is critical in modern geospatial analysis since these methods are mostly digital. A similar study by Gede et al focused on vectorizing linear hydrographic features using QGIS (Gede et al 2020) These studies provide scanning and rectifying techniques as well as several raster pre-processing and vector cleaning methods. Quan et al focused on point symbol recognition combined with image segmentation on a pretrained CNN of AlexNet architecture and achieved 98.97% accuracy on single test images (Quan et al 2018).
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
