AERIAL PHOTOGRAPH RECOGNITION AND SEGMENTATION METHODS FOR DIGITAL MAPS

Authors

DOI:

https://doi.org/10.32689/maup.it.2025.4.32

Keywords:

aerial photographs, object recognition, image segmentation, YOLO v8, GeoAI, OpenStreetMap (OSM), deep learning, geographic information systems, mapping automation, polygon orthogonalization, geoJSON

Abstract

The article examines the methodology of recognition and segmentation of aerial photographs with the aim of automating the mapping of buildings and structures on OpenStreetMap (OSM). The relevance of the research is determined by the need to reduce the manual work of volunteers who fill the OSM database and to increase the accuracy of spatial object representation. To achieve this goal, the author created a custom dataset containing satellite images of settlements in central and western Ukraine with geographic reference, building masks, and a spatial resolution of approximately 0.6 m/pixel. The study compares three deep learning models for object recognition: GeoAI (ResNet-34), YOLO v8m, and YOLO v8n. The training was conducted for 200 epochs with an 80/20 data split. According to the experimental results, the YOLO v8m model showed the best generalized accuracy (Precision ≈ 0.797, mAP50 ≈ 0.777), while GeoAI demonstrated lower values (IoU = 0.63 and Dice = 0.68). At the same time, the lighter YOLO v8n model exhibited higher sensitivity to small objects, which is explained by a smaller number of parameters and a wider receptive field. Special attention is paid to the post-processing of segmentation results – orthogonalization of the detected building contours and conversion of data into the geoJSON format, compatible with the OSM ecosystem. An algorithm for simplifying polygons to 4–10 vertices with minimal shape loss is proposed to improve the efficiency of mapping objects. The obtained results confirm the feasibility of using deep learning models to automate cartographic processes. The proposed approach can serve as the basis for a service that automatically generates building layers on OSM based on the coordinates of the selected area, significantly optimizing the process of updating cartographic data and improving their quality.

References

Глазок О. М., Давидова О. М. Картографічний додаток відображення забруднення атмосфери: матеріали VII Всеукраїнської науково-практичної конференції за міжнародною участю (м. Київ, 27 квітня 2021 р.). Київ : Національний авіаційний університет, 2021. С. 101–106. DOI: 10.18372/2786-5487.1.15836 (дата звернення: 09.09.25)

Карпінський Ю. О., Кінь Д. О. Методичні рекомендації щодо діяльності органів місцевого самоврядування у сфері НІГД : практичний посібник. Київ : КНУБА, 2023. С. 108–109. URL: https://land.gov.ua/wp-content/uploads/2024/03/posibnyk-gromadam-nigd_fin.pdf (дата звернення: 09.09.25)

Що таке GSD: просторовa деталізація в аерозйомці. PortalGIS. Опубл. 15 2025. URL: https://portalgis.pro/bpla/shho-take-gsd-prostorova-detalizacziya-v-aerozjomczi/ (дата звернення: 09.09.25)

Calculate Google Map zoom attribute for defined range. CCW. URL: https://www.ccw.sk/sk/blog/calculate-googlemap-zoom-attribute-for-defined-range.html (дата звернення: 09.09.25)

Dakhno N., Barabash O., Shevchenko H., Leshchenko O., Dudnik A. (2021, October). Integro-differential models with a K-symmetric operator for controlling unmanned aerial vehicles using a improved gradient method. In 2021 IEEE 6th International Conference on Actual Problems of Unmanned Aerial Vehicles Development (APUAVD) (pp. 61–65). IEEE.

Esri. What is GeoAI? – Overview. URL: https://www.esri.com/en-us/capabilities/geoai/overview. (дата звернення: 09.10.25)

Kong L., Qian H., Xie L., Huang Z., Qiu Y., Bian C. Multilevel Regularization Method for Building Outlines Extracted from High-Resolution Remote Sensing Images. Applied Sciences. 2023. Vol. 13, No. 23. Article 12599. DOI: https://doi.org/10.3390/app132312599 (дата звернення: 09.09.25)

OpenGeoAI. Train segmentation model – Download sample data. URL: https://opengeoai.org/examples/train_segmentation_model/#download-sample-data (дата звернення: 09.09.25)

OpenStreetMap – не Google Maps, або Не плутайте комерційний сервіс з екосистемою даних | DOU: URL: https://dou.ua/forums/topic/44694/ (дата звернення: 09.09.25)

Trush O., Kravchenko I., Trush M., Pliushch O., Dudnik A., Shmat K. (2021, December). Model of the sensor network based on unmanned aerial vehicle. In 2021 IEEE 3rd International Conference on Advanced Trends in Information Theory (ATIT) (pp. 138–143). IEEE.

Ultralytics. YOLOv8 Performance Metrics. URL: https://docs.ultralytics.com/models/yolov8/#performancemetrics (дата звернення: 09.09.25)

Published

2025-12-30

How to Cite

KOLISNYK, I., DUDNIK, A., YAREMENKO, D., & FESENKO, V. (2025). AERIAL PHOTOGRAPH RECOGNITION AND SEGMENTATION METHODS FOR DIGITAL MAPS. Information Technology and Society, (4 (19), 214–220. https://doi.org/10.32689/maup.it.2025.4.32

Most read articles by the same author(s)

Obs.: This plugin requires at least one statistics/report plugin to be enabled. If your statistics plugins provide more than one metric then please also select a main metric on the admin's site settings page and/or on the journal manager's settings pages.

Similar Articles

<< < 1 2 3 4 5 6 7 8 9 10 > >> 

You may also start an advanced similarity search for this article.