Table 2 Summary of the best hyperparameter used to train the segmentation model.

Architecture

U-Net²⁰

Backbones

ResNet-34²⁶

Batch size

16

Normalization

0 − 255 to 0 − 1 OptimizerAdam

Learning rate

0.0001

Data augmentation

Rotation, Horizontal/vertical flipping Epochs 100

Loss function

Weighted cross entropy (WCE)