Railway Subgrade Defect Automatic Recognition Method Based on Improved Faster R-CNN

Table 2

Data augmentation methods applied in the improved model.

Data Augmentation Method	Method description
Flip	Flipping the image horizontally or vertically.
Contrast	Changing the saturation S and V luminance component in the HSV color space of the image, and keeping the tone H unchanged.
Zoom	Enlarging or compressing the image according to a certain proportion.
Random Cropping	Clipping the image randomly, including the scale jittering method and the enhancement transform of dimension and length-width ratio.
Rotation	Rotating the image at arbitrary angle to change the orientation.