Abstract: Infrared tiny object tracking plays a critical role in various unmanned aerial vehicle (UAV) applications, including aerial surveillance, wildlife protection, and security patrol, owing to its advantages such as low visibility, strong anti-interference capability, and all-weather operability. However, infrared images typically present challenges including small target scale, low contrast, blurred boundaries, and complex background interference, which severely hinder the performance of conventional tracking algorithms in terms of feature perception, localization accuracy, and real-time deployment. To address these issues, a lightweight Transformer-based tracking algorithm tailored for edge-side real-time infrared small object tracking was proposed. The proposed method designed an efficient backbone network based on MobileNetV3, which could significantly reduce the computational complexity while preserving feature representation capacity, thereby meeting the resource constraints of domestic embedded platforms. Furthermore, a dual-level thermal-aware feature interaction structure was constructed to enhance the discriminability of small target regions by integrating both local saliency and global thermal responses.In addition, the regression loss was formulated based on the normalized Wasserstein distance (NWD) to improve the spatial localization accuracy and geometric stability of small targets under scale variation and perspective change. Extensive experimental results conducted on the public BIRDSAI dataset demonstrate that the proposed method effectively addresses the challenges of infrared small object tracking and achieves stable real-time performance on domestic embedded platforms with an precision of up to 80.8% and a frame rate of 43 frame/s, indicating strong potential for practical deployment.