Enhancing YOLO through Multi-Task Learning: Joint Detection, Reconstruction, and Classification of Distorted Text Images