Aerial person re-identification (AReID) remains a challenging task due to the small scale, low resolution, and drastic viewpoint variations of UAV-captured images. Traditional person ReID methods, optimized for ground-based scenarios, often fail to generalize to aerial scenes. In this paper, we propose SRSG-PReID, a novel framework specifically designed for AReID that integrates robust feature learning with self-rewarded attention regulation. First, we introduce Self-Rewarded Shuffle Grouping (SRSG), which applies shift, shuffle, and grouping operations over patch tokens, enforcing resilience to spatial distortions and encouraging the model to capture fine-grained discriminative cues beyond global structures. Second, we design an Self-Rewarded Attention (SRA) mechanism, where self-adaptive rewards dynamically modulate attention weights to Progressive Reward Scheduling (PRS) strategy to stabilize training by gradually increasing the contribution of reward-guided supervision in an easy-to-hard learning fashion. Extensive experiments on aerial, ground, and cross-view datasets demonstrate that SRSG-AReID consistently outperforms state-of-the-art methods, achieving superior robustness and accuracy under complex UAV-based ReID conditions.