XXIX-B3, 2012
1e correspondence
SS-III pixels (here
pixel of AWIFS)
essment. The flow
in Fig. 1. The six
orest, eucalyptus
crop, agriculture
t crop have been
collected with the
while taking 100
mly selected.
ried to find out the
or FCM and PCM
entation on noise
ere it has tried to
arameter (?) with
kappa coefficient.
n taken from 1 to
eighting exponent
racy, fuzzy kappa
meters have been
data sets. It has
r increases, fuzzy
coefficient also
also observed that
i given Fig. 4 and
be the appropriate
. noise clustering
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
5. RESULTS AND DISCUSSIONS
The uncertainty is a significant issue in the classification of
remote sensing data. The uncertainty estimation of the
classification results is important and necessary to evaluate
the classifier performance. In this paper, we addressed the
evaluation of FCM, PCM and Noise Clustering without
Entropy
based classifier, while estimating uncertainty in fuzzy overall
accuracy and fuzzy kappa coefficient with varying spatial
resolution of classification and reference sub-pixel outputs.
The uncertainty criteria have been estimated from SCM
matrix based on actual and desired outputs of classifier.
Therefore, these criteria are dependent on the error of the
results and sensitive to error variations. So it has also been
tried to estimate entropy, based on actual outputs of classifier
and hence is sensitive to uncertain variations.
In this research work performance of each classifier was
estimated based on overall accuracy, fuzzy kappa coefficient,
uncertainty in overall accuracy and fuzzy kappa coefficient
and entropy mentioned in Fig 3, 5, 7 respectively. In this
paper, we addressed the evaluation of FCM classifier, PCM
classifier and noise clustering without entropy while
estimating uncertainty and overall accuracy from SCM and
fuzzy kappa coefficients shown in Table 1 and Fig. 3, 4, 5
and 6 for AWIFS with LISS-III. From the Fig. 7 it is clear
that the entropy is higher for AWIFS in case of PCM
classifier.
6. CONCLUSION
In this research work performance of each classifier was
estimated based on overall accuracy, fuzzy kappa coefficient,
uncertainty in overall accuracy and fuzzy kappa coefficient
and entropy. It has been tried to generate fraction outputs
from FCM, PCM, and noise clustering without entropy.
These outputs have been generated from AWIFS as well as
LISS-III images of IRS-P6 data. Fuzzy overall accuracy and
fuzzy kappa coefficient are relative accuracy assessment but
entropy is an absolute uncertainty indicator.
100
90 | &
80
70 -
60
50 -
Overall Accuracy (9o
40 - fi Be Gee Be e
30 +
Table 1: Overall maximum fuzzy accuracy from
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
different classifiers with optimum parameters
5 > »
9 ~~ N 7 N = >
; E |<E = E| 38, | £2 | €
o S 5 0 cs £5 es =
A 2 sn = 85 = 2 z- =
= en BT py M2 22 m
3 B =5 | E= RS = 2 N
= N NS ST NS Sc m
© S S58i558| 29 5E z
= & 3 3 2 |. 9S
Eb 8 £0 B <
o
rx
FCM 31 7758 | 1193 | 0.700 0.165 | 0.01
PCM 1.4 45.41 | 30.65 | 0.0919 | 0.623 0.14
Noise 20
Clusteri aod
ng 96.27 0.29 0.494 | 0.058 0.71
; delta
without zl
Entropy
The following legends are used in graphs.
seis
FOAM
oo. 535TH A
Pe
NU- ulus Egtson
ee BW a Bids op
wo BH -W ot Entropy
ss HON mbent Entgopy
m dede RÀ
Values of Regulanzing Parameter
Figure 3: Overall Accuracy for different c
v iwucel
cO NC-Wubent Entopy Bor Ce Ht
pw CMI
y Dor 93M:
tar 2-13
E
14 15 1617 18 19.2 21 22 23 24 25 26 2.7 283 29 3 31 32
lassifiers of AWIFS with LISS-III
