IRSO-YOLO: An Improved Infrared Small Object Detection Algorithm Based on YOLOv8

Le résumé fourni par la source

Infrared images are characterized by low contrast and low signal-to-noise ratios, particularly for small objects, which lack distinct features in infrared images, making the detection of infrared small object a significant challenge. To alleviate these issues, based on YOLOv8 algorithm, this paper proposes an improved algorithm named IRSO-YOLO for infrared small object detection, which enhanced detection performance through optimizing the feature scales and integrating an SEAttention module. Experimental results demonstrate that the IRSO-YOLO outperforms other classic algorithms on the test set of the HIT-UAV dataset, with a 3.2% increase in mAP50 over the baseline, and increases of 2.3% and $13.5 \%$ in AP50 for Tiny and Small objects, respectively, while reducing the model’s Params by $69.4 \%$. Additionally, the improved algorithm demonstrates excellent detection performance on embedded platform, achieving real-time detection effect of 31.4 FPS, which proves its application potential and high efficiency in real environment.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.