Unlocking freeform structured surface denoising with small sample learning: Enhancing performance via physics-informed loss and detail-driven data augmentation.