Abdulkarim Atrash, M.Sc.
Department of Scientific Computing
September 2025
Supervisor: Ömür Uğur (Institute of Applied Mathematics, Middle East Technical University, Ankara)
Co-Supervisor: Şeyda Ertekin Bolelli (Computer Engıneering, Middle East Technical University, Ankara)
Abstract
Infrared Small Target Detection (IRSTD) is a challenging problem with critical applications in defense and surveillance. It involves detecting tiny, low-contrast targets in cluttered infrared scenes. Single-frame IRSTD (SIRST) refers to a class of algorithms that rely solely on spatial features extracted from a single input frame. Learning-based SIRST methods leverage neural network architectures to address the IRSTD problem. Although these methods are relatively straightforward to design, they face several challenges that limit their applicability in real-world scenarios. In response to these challenges, this thesis investigates and presents innovative solutions. We first conduct an evaluation study of state-of-the-art YOLO algorithms to motivate the selection of our baseline model. Building upon this baseline, we propose TY-RIST, a unified framework based on the recently introduced YOLOv12n architecture, designed to address several critical challenges in the field of IRSTD. Extensive experiments on four benchmarks and across 20 different models demonstrate state-of-the-art performance. Cross-dataset validation on a fifth unseen dataset further confirms the strong generalization capability of our method.
Keywords: Single-Frame Infrared Small Target Detection, Moving Target Detection, Small Object Detection, SIRST, YOLO