IFD-YOLO: A Lightweight Infrared Sensor-Based Detector for Small UAV Targets

Area of Science:

• Computer Vision
• Artificial Intelligence
• Aerospace Engineering

Background:

• Small target detection in infrared imagery from UAVs is crucial for surveillance.
• Challenges include small target size, low signal-to-noise ratio, and limited UAV computational resources.

Purpose of the Study:

• To propose IFD-YOLO, a novel lightweight detector for onboard infrared sensing systems on UAVs.
• To enhance accuracy and efficiency in detecting small targets under resource constraints.

Main Methods:

• Developed IFD-YOLO based on YOLOv11n, incorporating a RepViT backbone for feature extraction.
• Introduced a C3k2-DyGhost module for dynamic feature fusion and Adaptive Fusion-IoU (AF-IoU) loss for improved bounding-box regression.
• Validated performance on HIT-UAV and IRSTD-1k datasets.

Main Results:

• IFD-YOLO achieved a superior balance between accuracy and efficiency compared to YOLOv11n.
• mAP@50 increased by 4.9% and mAP@50:95 by 3.1%.
• Model parameters and GFLOPs were reduced by 23% and 21%, respectively.

Conclusions:

• IFD-YOLO demonstrates strong potential for real-time infrared sensing on resource-constrained UAV platforms.
• The proposed model effectively addresses challenges in small target detection for aerial surveillance.