Emin Argun Oral
Nida Kumbasar
Aslı Nur Ömeroğlu
İbrahim Yücel Özbek
Abstract
This study proposes a new dataset of high resolution satellite images of hangars located at airports. It contains one thousand pictures obtained from Google Earth with different heights and angles. The detection of hangars is a challenging problem as the dataset contains camouflaged and non-camouflaged targets in different sizes. Mask R-CNN, regional based, and YOLOv2, regression based, algorithms were used in the detection problem. Mask R-CNN enables instance segmentation with a bounding box. YOLOv2, on the other hand, is used in real-time applications and provides only a bounding box. The object detection accuracy obtained by using Mask R-CNN and YOLOv2 algorithms to detect different sized objects was obtained as %72 and% 74, respectively.
Keywords
YOLOv2; Mask R-CNN; Satellite Images; Camouflaged and Non-Camouflaged Hangar; Object Detection
