Random rotation is one of the core methods of data augmentation. It rotates an image at a random angle within a preset angle range (such as 0-360 degrees) and maintains the image size through padding operations, enabling the model to learn the feature expression of targets at different rotation angles and enhancing the model's robustness.
