Enhancing Small Object Detection in Aerial Imagery: A Comparative Study of YOLO and RT-DETR Models Using Slicing Aided Hyper Inference

Rajendran, Ajaykkumar (2024) Enhancing Small Object Detection in Aerial Imagery: A Comparative Study of YOLO and RT-DETR Models Using Slicing Aided Hyper Inference. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (11MB) | Preview
Preview	PDF (Configuration Manual) Download (210kB) | Preview

Abstract

Detecting small objects in many aerial images caught by drones is now an important challenge that many experts face in the field of computer vision. Traditional object detection algorithms often struggle to identify small objects. Their low resolution and complicated aerial environments cause this issue. Traffic monitoring, urban planning, and disaster management will grow in relevance with this issue. The performance of deep learning models like YOLO and RT-DETR are investigated in this study. The Slicing Helped Hyper Inference technique is used to enhance these models for detection improvements. In this study, three models were trained on a VisDrone dataset for 200 epochs and their performance were compared with standard inference and SAHI techniques to determine their efficiency. The outcome of this study showed that with SAHI models were performing better, with higher mAP metrics, especially in detecting small objects. By increasing accuracy and reliability, many practical applications can greatly benefit from improved object detection for drone surveillance systems, which are important in many fields such as urban planning and emergency responses. SAHI technique and pre-trained models are trained and combined in this study, drawing attention to opportunities to solve problems in computer vision.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Singh, Jaswinder UNSPECIFIED
Uncontrolled Keywords:	Deep Learning; Object Detection; Small Objects; YOLO (You Only Look Once); RT-DETR (Real-Time Detection Transformer); SAHI (Slicing Aided Hyper Inference
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	04 Sep 2025 11:01
Last Modified:	04 Sep 2025 11:01
URI:	https://norma.ncirl.ie/id/eprint/8780

Actions (login required)

View Item