The detection of small targets in infrared imagery captured by unmanned aerial vehicles (UAVs) is critical for surveillance and monitoring applications. However, this task is challenged by the small target size, low signal-to-noise ratio, and the limited computational resources of UAV platforms. To address these issues, this paper proposes IFD-YOLO, a novel lightweight detector based on YOLOv11n, specifically designed for onboard infrared sensing systems. Our framework introduces three key improvements. First, a RepViT backbone enhances both global and local feature extraction. Second, a C3k2-DyGhost module performs dynamic and efficient feature fusion. Third, an Adaptive Fusion-IoU (AF-IoU) loss improves bounding-box regression accuracy for small targets. Extensive experiments on the HIT-UAV and IRSTD-1k datasets demonstrate that IFD-YOLO achieves a superior balance between accuracy and efficiency. Compared to YOLOv11n, our model improves mAP@50 and mAP@50:95 by 4.9% and 3.1%, respectively, while simultaneously reducing the number of parameters and GFLOPs by 23% and 21%. These results validate the strong potential of IFD-YOLO for real-time infrared sensing tasks on resource-constrained UAV platforms.
Li et al. (Sun,) studied this question.