Small Object Detection in High-Resolution Images: A Systematic Literature Review

Abstract

Detecting small objects in high-resolution imagery remains challenging due to extreme scale variation, feature degradation during down sampling, and complex background clutter. This study utilized a Systematic Literature Review (SLR) to analyze 55 deep learning studies on small object detection published between 2021 and 2026. The review aimed to identify dominant architectural approaches, methodological improvements, and remaining technical limitations in current detection frameworks. The analysis shows that YOLO-based architectures dominate the research landscape, accounting for 49.1% of the reviewed methods. The results indicate that multi-scale feature fusion and spatial-preservation techniques are essential for detecting objects smaller than 16 × 16 pixels. Methods such as Space-to-Depth down sampling, high-resolution P2 prediction heads, and coarse-to-fine detection strategies consistently improve feature retention and detection performance in high-resolution imagery. The review also finds that Transformer-based and hybrid CNN–Transformer architectures provide stronger contextual modeling in complex scenes; however, their computational complexity limits deployment in real-time edge environments. The findings highlight the need for more computationally efficient architectures and identify emerging directions such as Vision Mamba state-space models, temporal-aware detection using video data, and lightweight model distillation to improve scalability and cross-domain robustness.