Satellite video object tracking has unique challenges, such as small target, low target-background contrast, similar targets interference and frequent occlusions. However, most current trackers rely on the sparse temporal relationship between template frames and search frames, ignoring the contextual information between the global frames in the video. In this paper, we propose SiamDTO, a siamese network tracking framework that fuses spatio-temporal attention with Mamba. Specifically, we utilizes state space model of Mamba to establish cross-frame feature continuity. By passing hidden state between video frames, the correlation of temporal context information is established. In addition, we introduces a coordinate space attention and a lightweight channel attention mechanism to improve separability between target and background region. We propose new threshold for detecting occlusion events to process occlusion. Experiments demonstrate that SiamDTO outperforms state-of-the-art trackers, e.g., it achieves 49.7% and 52% AUC on the SatSOT and SV248S, respectively.