<p>A hybrid framework is presented for blur classification and adaptive restoration of aerial imagery captured under variable illumination and motion conditions. The pipeline begins with a multi-scale entropy-guided classification module that identifies the underlying blur category and generates confidence cues for subsequent stages. For motion-dominated scenarios, a physics-informed kernel regression stage estimates blur trajectories consistent with the dynamics of the aerial platform. Restoration is then performed through an attention-driven Transformer coupled with a dual-domain fusion mechanism that combines spatial refinement and frequency-based enhancement to preserve fine structural details. Evaluation on diverse aerial datasets demonstrates improvements in perceptual clarity and structural similarity, particularly in regions exhibiting complex edge geometry. The framework further produces measurable gains in downstream object detection tasks, indicating enhanced operational utility in surveillance and inspection scenarios. Results confirm that the integrated design effectively mitigates degradation encountered during aerial acquisition at varying speeds and environmental contrasts. The approach contributes an application-aware quality index that reflects both pixel-level reconstruction improvements and task-level performance shifts. The results indicate that the restoration process strengthens both interpretability and analytic reliability for mission-critical aerial applications. The design principles established through this work support deployment in real-time imaging pipelines across diverse environmental contexts.</p>