Table 2.

Summary of real-time data augmentation methods used in the study

Method	Range	No. of synthetic images
rotate	−30° ≤ rotation angle ≤30°	61
resize	0.85 ≤ width < 1.0, 0.9 ≤ height < 1.0	150
shear	−5° ≤ x angle ≤5°, −5° ≤ y angle ≤5°	121
pixel transform	α* pixel +_β, (0.9 ≤ α ≤ 1.0, 0 < β ≤ 10)	100

Geometric (rotation, resizing, and shearing) and photometric transformations (contrast and brightness) were applied to input images prior to training the network to prevent overfitting