2025 5th International Conference on Neural Networks, Information and Communication Engineering (NNICE) YOLO-UIBi: A Lightweight Object Detection Model for UAV Infrared Thermal Imaging

With the rapid development of unmanned aerial vehicle (UAV) technology, infrared image-based object detection has garnered significant attention for applications in complex environments. However, infrared target detection still faces numerous challenges, including small target sizes, low signal-to-noise ratios, and interference from complex backgrounds. This study aims to enhance detection accuracy for infrared targets while reducing model complexity by proposing YOLO-UIBi, a lightweight object detection model specifically designed for infrared thermal imaging based on an improved YOLOv8. Fistly, the C2f module is replaced in the backbone network with the C2f_UIB module from MobileNetV4, significantly reducing computational complexity. Additionally, BiFPN is introduced in the neck network to enhance multi-scale detection performance through an efficient feature fusion mechanism. Finally, the CIoU loss function is replaced with WIoU, further improving model performance. Experimental results demonstrate that the YOLO-UIBi model effectively detects infrared targets in UAV environments, achieving an mAP of 0.793, which is a 3.8% improvement over the baseline model. Additionally, the model's parameters are reduced by 26.81%, computational complexity by 23.46%, and its size is only 5.3 MB. Compared to other mainstream detection models, the improved YOLOv8 model shows significant advantages in detection accuracy, parameter size, computational efficiency, and model size, making it highly practical for real-world applications.