Abstract: The contradiction between the requirements for reliable, efficient, and precise target recognition performance and the challenges in constructing comprehensive target databases demands that radar target recognition systems possess dynamic learning capabilities. These capabilities enable dynamic updates of data and models, as well as continuous improvement in recognition performance. The realization of functions such as sample self-labeling and model self-updating serves as prerequisite for achieving this objective. To address the practical need for performance self-enhancement in radar target recognition applications, an online transfer learning framework by integrating concepts from online learning and transfer learning is proposed in this study. Featuring a closed-loop structure, the framework combines online learning with transfer learning technologies to achieve self-iterative model optimization through sample annotation and model fine-tuning, thereby automatically completing tasks such as sample labeling and model updating. Experimental results based on simulated data demonstrate that the proposed framework significantly enhances radar target recognition accuracy. With advantages including streamlined processes and rapid deployment, the framework exhibits strong engineering practicality.