İNSANSIZ HAVA ARAÇLARI İÇİN U-NET TABANLI ARAÇ TESPİT YÖNTEMİ

Abstract

With the developments in computer hardware technology, studies in the fields of computer vision and artificial intelligence has accelerated. However, the number of areas where autonomous systems are used has also increased. Among these areas are unmanned aerial vehicles, which are one of the most important parameters of today's military technology. In this study, which includes two different scenarios, we aimed to improve the vision capabilities of unmanned aerial vehicles based on artificial intelligence. Within the scope of Scenario-1, the U-Net model suitable for binary semantic segmentation method was trained with the help of images taken by unmanned aerial vehicle camera. Within the scope of Scenario-2, which is designed for moving or stationary vehicle detection, the U-Net model is trained in accordance with multi-class semantic segmentation method. In all these training processes, a publicly available dataset was used. The model trained for Scenario-1 reached mean Intersection over Union (mIoU) value of 84.3%, while the model trained for Scenario-2 reached 79.7% mIoU. In this study, approaches were shared about the use of high-resolution images in model training and testing stages. Applying such studies in the field can help improve precision and reliability in arms industry.