461
Figure 5. Trends of RMSE of testing samples for the different model condition.
model performances results. The performance of the network is investigated separately for
the different number of hidden layers (1, 2, 3, 4, and 5) with the different number of nodes
in the hidden layers (2, 3, 4, 5, 10, 50, 100, and 500). In the present study, 40 Deep network models (M11, M12, …M15, M21,…M25, M31…M35, M41…M45, M51….M55,
M61…M65, M71….M75, M81…M85) are trained and tested. The performance of the developed model was evaluated using the four indices (RMSE, NMSE, R2, and bias) as explained
in Section 2.4. The model performance indices for different combinations of layer and node
are summarized in Table 2.
All the indices are determined using the predicted grades and observed grades of the testing samples. The results indicated that the NMSE value of the model M43 is lowest for
the 3-hidden layer and 5-nodes in each hidden layer. The lowest NMSE indicates that the
observed grade values are more closely matched with the predicted values. The trend of
RMSE and R
2 values are also represented in Figure 5 and Figure 6 respectively. It is observed
that the RMSE value is lowest for model M43.The trend of RMSE clearly indicates that the
performance of the M43 model is best out of all. Furthermore, the trend of R
2 shows that
the regression coefficient of the M43 model (= 0.8701) is nearly equal to the highest R
2 value
observed for model M51 (= 0.8758). The R
2 value indicates that the correlation between the
observed and predicted values is good for the M43 model. The biases for all the models are
found to be relatively high and negative. A negative bias indicates that the model performs
with over prediction and thus the models need to be trained and tested using more number
of image samples.
Figure 6. Trends of R
2 of testing samples for the different model condition.
"'
11.00
10.50
.; 10.00
;>
~ 9.50
"' :E 9.00
~ 8.50
8.00
90
85
"'
.E! 80
'"
;>
' il:: 75
70
65
2
t=
2
~ Ll ~ L2 ~ L3 ~ L4 ~ L5
. .
3
4
5
lO
50
100
500
Number of Nodes in Each Hidden Layer
~LI ~ L2 ~ L3 ~ L4 ~ L5
I
4
5
10
50
100
500
Number of Nodes in Each Hidden Layer
Figure 5. Trends of RMSE of testing samples for the different model condition.
model performances results. The performance of the network is investigated separately for
the different number of hidden layers (1, 2, 3, 4, and 5) with the different number of nodes
in the hidden layers (2, 3, 4, 5, 10, 50, 100, and 500). In the present study, 40 Deep network models (M11, M12, …M15, M21,…M25, M31…M35, M41…M45, M51….M55,
M61…M65, M71….M75, M81…M85) are trained and tested. The performance of the developed model was evaluated using the four indices (RMSE, NMSE, R2, and bias) as explained
in Section 2.4. The model performance indices for different combinations of layer and node
are summarized in Table 2.
All the indices are determined using the predicted grades and observed grades of the testing samples. The results indicated that the NMSE value of the model M43 is lowest for
the 3-hidden layer and 5-nodes in each hidden layer. The lowest NMSE indicates that the
observed grade values are more closely matched with the predicted values. The trend of
RMSE and R
2 values are also represented in Figure 5 and Figure 6 respectively. It is observed
that the RMSE value is lowest for model M43.The trend of RMSE clearly indicates that the
performance of the M43 model is best out of all. Furthermore, the trend of R
2 shows that
the regression coefficient of the M43 model (= 0.8701) is nearly equal to the highest R
2 value
observed for model M51 (= 0.8758). The R
2 value indicates that the correlation between the
observed and predicted values is good for the M43 model. The biases for all the models are
found to be relatively high and negative. A negative bias indicates that the model performs
with over prediction and thus the models need to be trained and tested using more number
of image samples.
Figure 6. Trends of R
2 of testing samples for the different model condition.
"'
11.00
10.50
.; 10.00
;>
~ 9.50
"' :E 9.00
~ 8.50
8.00
90
85
"'
.E! 80
'"
;>
' il:: 75
70
65
2
t=
2
~ Ll ~ L2 ~ L3 ~ L4 ~ L5
. .
3
4
5
lO
50
100
500
Number of Nodes in Each Hidden Layer
~LI ~ L2 ~ L3 ~ L4 ~ L5
I
4
5
10
50
100
500
Number of Nodes in Each Hidden Layer