Abstract: Skywave over-the-horizon radar （OTHR） has unique advantages in remote target detection and tracking； however， it also has the disadvantages of low measurement accuracy and difficult target tracking. To this end， this paper proposes a target-tracking approach in the OTHR measurement coordinate system， namely the dual-channel feature fusion transformer network （DcFFTNet）. DcFFTNet uses a dual-channel feature fusion module to extract target trajectory sequence information from different dimensions. It constructs feature vectors for each time step of the entire trajectory sequence， ensuring that the features from each dimension can be fully expressed. This enables a more comprehensive description of the target’s motion state and characteristics. Moreover， DcFFTNet uses a multi-head deformable attention mechanism to adaptively focus on crucial parts of the trajectory sequence， thereby better capturing critical measurement information during the tracking process. Consequently， this method enhances the feature representation of trajectory sequences and enables adaptive information extraction， thus making it effective for tracking targets detected by OTHR. To facilitate the training and testing of the network model， this study developed a target trajectory generator. Based on the state space model and OTHR measurement model， it creates a dataset of target trajectory samples with combined motion patterns of constant velocity and constant turn-rate. Simulations show that the approach in this paper outperforms models， such as the unscented Kalman filter and long short-term memory networks in tracking accuracy.