Deep learning for object detection

Abstract

The aim of this dissertation is to compare three YOLO-based object detection algorithms on the MAR20 dataset, which includes annotated remote sensing images of military aircraft. The study focuses on the small versions of YOLOv8s, YOLOv9s, and YOLOv10s. Each model, after tuning, was trained and evaluated under the same settings. Their performance was measured using precision, recall, F1-score, and average precision (mAP). The results show that all YOLO versions achieve high detection accuracy, but YOLOv8s offers the best overall balance between performance and computational efficiency.