Table 1 Confusion matrices: each column shows the reference, while numbers running diagonally show the percentage of correct classification for every class considered in the experimental study

VGG architecture

 1

0.930

0

0

0

0

 2

0.023

0.943

0.036

0.019

0.027

 3

0.022

0.029

0.932

0.022

0.027

 4

0.025

0.027

0.032

0.957

0.028

 > 4

0

0

0

0.001

0.917

AlexNet architecture

 1

0.910

0

0

0

0

 2

0.030

0.944

0.039

0.030

0.026

 3

0.029

0.029

0.920

0.032

0.026

 4

0.031

0.027

0.041

0.937

0.031

 > 4

0

0

0

0.001

0.917