SiamGMF: an improved target tracking model based on UAV aerial images
Siyao Duan, Ting Wang, Tao Li, Wankou Yang · Engineering Research Express · 2025
Abstract UAV target tracking often involves the objects with various challenging attributes, such as background clutter, occlusion, and motion blur, which complicate the tracking process. In this paper, a target tracking model called SiamGMF is proposed, specifically based on UAV aerial images. First, to enhance the model’s ability to extract target features in complex environments, a global efficient attention (GEA) module is introduced into the backbone network, which enhances feature representations by refining channel importance through efficient local interactions and enriching them with position-aware multi-scale contextual information. Second, a structure of multi-scale feature fusion module (MFNet) is introduced, which effectively integrates deep semantic information with shallow positional information, thus enriching the feature representation of the targets. Furthermore, to demonstrate the capability of SiamGMF in tracking targets under challenging conditions, the targets with different attributes using the OTB100 and UAV123 datasets are evaluated and visualized. Finally, experimental results indicate that SiamGMF outperforms the baseline models in terms of precision, success rate, and frames per second (FPS). In conclusion, our model demonstrates superior tracking performance compared with existing methods, indicating that the proposed approach is well-suited for target tracking in the presence of diverse challenging attributes.