interest (
AOI
) of this study. Conductivity measurements were
made with an EM38-MK2 instrument along the transects
shown in Plate 1. Kriging interpolation was applied to the
conductivity data to generate the conductivity map as shown
in Plate 1, from which the training and validation values were
taken. This figure shows a range of conductivity from 63 to
110 mS/m where the higher conductivity values are in
AOI
4.
We applied the algorithms and scenarios described above to
the masked areas by using the conductivity data for training
and testing the
BPNN
and
WBNN
as estimators. For each feature
group scenario the
BPNN
and
WBNN
results were analyzed.
A summary of the correlation and root mean squared er-
ror (
RMSE
) of the estimates and target data for the scenarios
are shown in Tables 2 and 3 for
BPNN
and
WBNN
algorithms
with 100 iterations. As we can see, there is not much differ-
ence between the
WBNN
and
BPNN
results. Our aim in testing
WBNN
was to determine whether there was any advantage
from using wavelet based activation functions in the neural
network, especially since we are applying scaling and wavelet
functions to the input data. Thus, the high and low frequency
Figure 7. Background image of areas of study (National Agricul-
ture Imagery Program - NAIP 2009).
Plate 1. Areas of interest masks and EC data collected along the black lines, and interpolation of the conductivity using kriging method
for AOIs (Background image: National Agriculture Imagery Program - NAIP 2009).
T
able
2. RMSE
and
C
orrelation
(A
verages
of
100 T
rials
)
between
the
C
onductivity
E
stimate
and
T
arget
for
S
cenario
1, 2,
and
3 U
sing
BPNN
Scenario 1
Scenario 2
Scenario 3
Eval.
AOI
RMSE Correlation RMSE Correlation RMSE Correlation
1 7.40
0.10
7.21
0.20
7.54
0.37
2 4.57
0.30
3.75
0.48
3.86
0.53
3 5.50
0.49
4.53
0.63
4.84 0.64
4 3.44
0.15
3.39
0.23
3.42 0.26
T
able
3. RMSE
and
C
orrelation
(A
verages
of
100 T
rials
)
between
the
C
onductivity
E
stimate
and
T
arget
for
S
cenario
1, 2,
and
3 U
sing
WBNN.
Scenario 1
Scenario 2
Scenario 3
Eval.
AOI
RMSE Correlation RMSE Correlation RMSE Correlation
1 7.59
0.09
7.28
0.14
6.84
0.35
2 4.09
0.30
3.74
0.50
3.53
0.52
3 5.29
0.49
4.63
0.67
4.38
0.70
4 3.40
0.13
3.36
0.22
3.46
0.28
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2016
515
