Hvlseg: An Ensemble Model For Instance Segmentation On Satellite Images

Automatic object detection and segmentation algorithms are essential for automatic analysis of high resolution remote sensing images. In order to make automatic analysis more healthy in remote sensing images, the sensitive boundaries of the objects detected in the images should also be segmented. In this study, an approach for automatic detection of aircrafts, ships, buildings/hangars, POL and heliports in electro-optic satellite images is proposed. The developed method is a hybrid approach in which multiple models are used together depending on the target type. The proposed approach won the General Classification first place in the automatic target detection competition (Innovative Software Competition) in satellite images organized by SSB and METU OGAM in 2019.