Table 4 Sensitivity of patch-based CNN model to different loss functions based on U-Net.

From: Patch-Based convolutional neural networks for multiple microstructural features detection in FIB-SEM micrographs of irradiated nuclear fuel

Model

Dice

IOU

Accuracy

Mean absolute error (MAE)

0.11

0.10

 

Mean squared error (MSE)

0.84

0.74

0.95

Categorical crossentropy

0.83

0.73

0.88

Focal

0.83

0.73

0.88

Dice

0.84

0.74

0.97

Dice + MAE

0.11

0.10

 

Dice + MSE

0.84

0.74

0.97

Dice + Crossentropy

0.83

0.73

 

Crossentropy + MAE

0.11

0.10

 

Crossentropy + MSE

0.84

0.74

0.88

Crossentropy + Focal

0.83

0.73

 

Focal + MSE

0.82

0.84

 
Back to article page