Unifying object detection and re-identification into a single network enables faster multi-object tracking. However, the competition between the detection and re-identification tasks within one-stage model may easily lead to inferior perfor-mance. In this paper, we propose a one-stage decoupling-based feature learning network for feature disentanglement and enhancement to solve the problem, termed as DFTracker. It can alleviate the competition between internal detection and re-identification tasks within the model, and promote their collaboration. Specifically, a new feature decoupling approach is designed to extract features required for detection and re-identification tasks separately, while two properly designed attention mechanism-based feature enhancement modules are introduced to strengthen the features of both parts. Finally, based on the obtained reinforced decoupling features, a novel lost target retrieval strategy is designed to further retrieve the lost targets, thereby enhancing the synergy between tasks and achieving more accurate tracking. By evaluating on the MOT17 and MOT20 benchmarks, we demonstrate that the design of our framework effectively enhances tracking per-formance, leading to clear improvements over existing one-stage approaches.