Answers are grounded in this paper and cite their sources.
Infrared small target detection plays a vital role in ground-based and airborne remote sensing applications, including wide-area surveillance, early warning systems, and uncrewed aerial vehicle monitoring. However, existing methods demonstrate significant performance degradation under extreme environmental conditions, particularly in the presence of atmospheric obscurants such as fog and thermal clutter from complex terrestrial backgrounds, resulting in elevated false alarm rates and low detection rates. In this study, a gear coupling network (GearNet) is proposed with a novel architecture that leverages a mechanical gear-inspired interaction to establish a bidirectional closed-loop fusion mechanism for multiscale feature enhancement. The proposed framework incorporates a nested convolution module designed for efficient fine-grained feature extraction, substantially improving system robustness and detection accuracy, while a deeper feature extraction module (C3k8) and a normalized Wasserstein distance loss are further examined through ablation. The experimental results on the dim and small target detection dataset (DATDT) infrared dataset demonstrate that GearNet achieves a precision of 98.35% and an mAP@50 of 90.09% with only 2.69 M parameters, improving the baseline YOLOv11 by 7.58 and 1.13 percentage points, respectively. Extensive evaluations across diverse infrared backgrounds (sea, mountain, city, and sky) on the SIDD dataset and generalization experiments on the Det-Fly dataset and Drone-dataset further confirm the model’s robustness and practical applicability for real-world remote sensing scenarios.
No takes yet. Share an insight, caveat, or question.
Zhu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: