Unsupervised Knowledge Extraction of Distinctive Landmarks from Earth Imagery Using Deep Feature Outliers for Robust UAV Geo-Localization
Machine Learning and Knowledge Extraction, Vol. 7, No. 3, Article 81
Editorial summary
A pretrained segmentation network encodes buildings from aerial images, and Isolation Forest selects visually distinctive outliers as landmarks. The method needs no additional network training. Retrieval experiments on VPAIR assess whether these landmarks are easier to match for navigation without GPS.
Bibliographic reference
Use the publisher’s record for the citation format required by your journal or organisation.
Zakhar Ostrovskyi, Oleksander Barmak, Pavlo Radiuk, Iurii Krak. "Unsupervised Knowledge Extraction of Distinctive Landmarks from Earth Imagery Using Deep Feature Outliers for Robust UAV Geo-Localization". Machine Learning and Knowledge Extraction, Vol. 7, No. 3, Article 81, 2025. https://doi.org/10.3390/make7030081
Test a method against your business task.
Describe the decision your system needs to support and the data you have. We assess whether a method fits, define a prototype and agree how to measure its quality, response time and operating cost.