Semi-supervised learning for topographic map analysis over time: a study of bridge segmentation

• Published in: Scientific reports 2022-11, Vol.12 (1), p.18997-12, Article 18997
• Main Authors: Wong, Cheng-Shih, Liao, Hsiung-Ming, Tsai, Richard Tzong-Han, Chang, Ming-Ching
• Format: Article
• Language: English
• Subjects: 704/172; 704/172/4081; Annotations; Anthropogenic factors; Humanities and Social Sciences; Image Processing, Computer-Assisted - methods; multidisciplinary; Neural networks; Neural Networks, Computer; Science; Science (multidisciplinary); Segmentation; Supervised Machine Learning; Topographic mapping; Topography; Transfer learning
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-022-23364-w

Geographical research using historical maps has progressed considerably as the digitalization of topological maps across years provides valuable data and the advancement of AI machine learning models provides powerful analytic tools. Nevertheless, analysis of historical maps based on supervised learning can be limited by the laborious manual map annotations. In this work, we propose a semi-supervised learning method that can transfer the annotation of maps across years and allow map comparison and anthropogenic studies across time. Our novel two-stage framework first performs style transfer of topographic map across years and versions, and then supervised learning can be applied on the synthesized maps with annotations. We investigate the proposed semi-supervised training with the style-transferred maps and annotations on four widely-used deep neural networks (DNN), namely U-Net , fully-convolutional network (FCN) , DeepLabV3 , and MobileNetV3 . The best performing network of U-Net achieves F 1 i n s t : 0.1 = 0.725 and F 1 i n s t : 0.01 = 0.743 trained on style-transfer synthesized maps, which indicates that the proposed framework is capable of detecting target features (bridges) on historical maps without annotations. In a comprehensive comparison, the F 1 i n s t : 0.1 of U-Net trained on Contrastive Unpaired Translation (CUT) generated dataset ( 0.662 ± 0.008 ) achieves 57.3 % than the comparative score ( 0.089 ± 0.065 ) of the least valid configuration (MobileNetV3 trained on CycleGAN synthesized dataset). We also discuss the remaining challenges and future research directions.