TOPOGRAPHIC MAP PRODUCTION AUTOMATION USING DEEP LEARNING METHODS
DOI:
https://doi.org/10.24867/23KG01BasaricKeywords:
Deep learning, Image segmentation, Topographic map, Topograpgic data extraction, Building footprint extractionAbstract
The production of digital topographic maps requires the use of data from various sources, within the framework of strict cartographic rules. The need to develop an advanced methodology for the automatic extraction of building data is justified. Using the U-net convolutional neural network implemented in Python, a building prediction mask was obtained on a 30 cm resolution satellite orthophoto. Vector processing of the mask was performed, creating polygons with given rotation data, from which corresponding point vectors representing buildings were obtained. Generated data require minimal manipulation and correction. The tool was developed for a 1:25,000 scale topographic mapsheet. The mapsheet covers an area of approximately 120 km2. With minor modifications, this tool can be used on a wide range of topographic maps.
References
[2] „Deep Learning vs. Machine Learning – What’s The Difference?,“ Levity, 26 Jул 2022. [На мрежи]. Доступно на: https://levity.ai/blog/difference-machine-learning-deep-learning#:~:text=Machine%2 0learning%20means%20computers%20learning,as%20documents%2C%20images%20and%20text.,[Последњи приступ 28 Јул 2022].
[3] O. Ronneberger, P. Fischer и T. Brox, „U-Net: Convolutional Networks for Biomedical Image Segmentation,“ MICCAI 2015 - Medical Image Computing and Computer-Assisted Intervention, т. Lecture Notes in Computer Science, бр. 9351, pp. 234-241, 2015.
[4] Z. Kokeza, M. Vujasinovic, M. Govedarica, B. Milojevic and G. Jakovljević, Automatic building footprint extraction from UAV images using neural networks, Geodetski Vestnik 64 (2022) 545.
[5] L. Shapiro и G. Stockman, „Computer Vision,“ New Jersey, Prencitce-Hall, 2001, pp. 279-325.
[6] Inria, „Inria Aerial Image Labeling Dataset,“ The dataset, [На мрежи]. Доступно на: https://project.inria.fr/aerialimagelabeling/). . [Последњи приступ 11 Септембар 2022].
