Abstract: ‍ ‍The photodetector is a critical component of the anti-unmanned aerial vehicle （anti-UAV） system， playing a pivotal role in the confirmation and forensics of UAV. The core technology of the photodetector is target tracking. Significant challenges exist in UAV tracking because of the compact size of the targets and the complexity of the background environment. Existing UAV tracking datasets are limited in terms of target size and attribute distribution characteristics and thus cannot fully represent complex real-world scenarios. To address these challenges， this paper proposes a Siamese network-based UAV tracking method that integrates a Regional Proposal Network （RPN） with motion constraints. The proposed method is based on the dual-template Siamese （SiamDT） network and involves the optimization of weight initialization strategies to enhance the stability and accuracy of model training. Furthermore， to address the challenges posed by false alarms and missed detections in UAV tracking， a dynamic adaptive filtering strategy that integrates target confidence thresholds and motion constraints is proposed. This strategy enhances the scene adaptive capability of the UAV tracking method by establishing multi-level confidence thresholds and dynamically adjusting the detection strategy. Finally， a high and low-frequency adaptive image enhancement strategy is proposed to solve the problems of missing UAV detail information and uneven brightness distribution in some scenes， thereby improving the performance of the tracking method in low-contrast scenes and high-dynamic-range images. To verify the performance of the proposed tracking method more comprehensively， the Anti-UAV500 dataset， which is more challenging and has stronger generalization performance， is constructed based on the Anti-UAV410 benchmark. The experimental results show that the SiamXC model proposed in this paper improves performance by 1.28% compared to the SiamDT model on the Anti-UAV410 dataset， performing significantly better than all existing state-of-the-art tracking methods on our dataset.