Deep Learning for Historical Cadastral Maps Digitization: Overview, Challenges and Potential

Abstract

Cartographic heritage of historical cadastral maps represent remarkable geospatial data. Historical cadastral maps are generally regarded as an essential part of the land management infrastructure (buildings, streets, canals, bridges, etc.). Today these cadastral maps are still in use in a digital raster form (scanned maps). Digitization of cadastral maps is time consuming and it is a challenge for scientists and engineers to find ways to automatically convert raster into vector maps. The process of map digitization typically involves several stages: preprocessing, visual object detection and classification, vector representation postprocessing and extracting information from text. Although neural networks have had a long history of use in the domain, their applications remain limited to extracting the information from text. Recent convergence of advancements in the domains of training deep neural networks (DNN) and GPU hardware allowed DNNs to achieve state-of-the-art results in computer vision applications, beyond hand-written text recognition. This paper provides an overview of different approaches to historical cadastral maps digitization, focusing of the challenges and the potential of using deep neural networks in map digitization.

A Digital 4D Information System on the World Scale: Research Challenges, Approaches, and Preliminary Results

“…Although building footprints are still mainly manually extracted from historical maps, various AI-based approaches support this task. CNNs and-more recently-transformer approaches are used for the segmentation of historical maps [145][146][147][148][149]. Another approach to automatically generating 3D/4D models comprises building footprint recognition and parametric modelling.…”

Segmentation of historical maps without annotated data

“…Recent convergence of advancements in the domains of training deep neural networks significantly improved results, inspired by computer vision applications [8]. Thus, map digitization requires visual object and pattern recognition, such as the identification of edges (parcel and building lines), symbols, text, and patterns.…”

“…They propose a Bi-Directional Cascade Network (BDCN) architecture, where an individual layer is supervised by labeled edges at its specific scale, rather than directly applying the same supervision to different layers. The main limit of using neural networks for historical map vectorization is the constraint of the limited amount of training data available [8].…”

EKSTRAKSI FITUR BANGUNAN SECARA OTOMATIS PADA CITRA SATELIT RESOLUSI TINGGI MENGGUNAKAN PENDEKATAN DEEP LEARNING BERBASIS MASK R-CNN (Studi Kasus di Kota Ambon)

“…Secara umum, cara kerja ekstraksi fitur bangunan pada citra satelit yaitu dengan cara mengenali bentuk bangunan pada citra kemudian merekonstruksikannya (Ghasemi Nejad et al, 2019;Theng, 2006). Skema seperti ini merupakan artificial intelligence (AI) yang menggunakan metode machine learning (ML) dengan teknik berupa neural network (NN) (Ahmad, 2017;Ignjatić et al, 2018). Neural network dapat bekerja berdasarkan data training dalam artian hasil rekonstruksi tergantung pada data trainingnya.…”