<p>Convolutional neural network based trackers have achieved excellent tracking performance in terms of accuracy and speed. The feature extraction network is an essential component of trackers, serving as the foundation for target representation and discrimination. Existing feature extraction sub-networks struggle to adequately extract discriminative target spatial and channel features, which limits their representation capability. Moreover, they fail to effectively suppress background redundancy, leading to potential interference from cluttered environments. To address the limitations of existing tracking methods, we design a novel feature extraction subnetwork that incorporates a refinement and rebuilding module, integrating a Spatial Feature Refinement Module (SFRM) and a Channel Feature Rebuilding Module (CFRM). Different from existing approaches, our approach fully exploits their complementary nature to enhance target representations. The SFRM focuses on refining spatial features to preserve structural and locational accuracy, while the CFRM adaptively rebuilds channel features to strengthen semantic consistency. This design enables more accurate extraction of target features from both the template and search region, effectively highlighting foreground targets while suppressing background. By reducing redundancy in the spatial domain and enhancing feature expressiveness, our method presents a novel and effective tracking framework which achieves more precise target localization. We evaluated the proposed tracker on LaSOT, TrackingNet, NFS, UAV123, GOT-10K, DTB70 and TNL2K benchmarks. Compared to EMAT, our tracker improves by 3.7% in AO.</p>