2004
———
ation
n be
ia
able.
the
each
ation
> that
oads,
ation
ation
was
May
prior
ding
luced
ation.
on of
luced
r and
over
ding
asked
class
Crops
, and
ugust
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
After completing the masking classification procedure, the
classified outputs were combined to make a single classified
image. À field-based analysis was then performed on the
final classified output by computing the class percentages
within each field and applying the entire field the label of the
modal class. The result of the multi-temporal masking
classification is illustrated in figure 3. The validation of the
masking classification was carried out by comparing the
reference data, field-by-field, to the classified image. The
individual class accuracies and the overall accuracy are
illustrated in an error matrix in table 3.
Figure 3. The result of sequential masking classification.
Reference
Cr Rs | Tm Sb Cl Pa Pp | Wn l| Bs Re | Cw
Cr 181 0 10 l 0 0 14 2 0 l ] | 210
Rs 0| 162 0 0 0 0 0 0 0 0 0 | 162
Tm 33 S 1:252 15 0 0 11 2 3 0 5132
Sb 0 0 0 84 0 0 0 0 0 0 0 84
CI 0 0 0 0 2 0 0 0 0 0 0 2
Pa 0 4 | 0 0 14 0 0 0 0 2:45:21
Pp 18 0 21 4 0 0 28 | 0 0 ] 73
Wm 0 0 | | 0 0 0 9 0 | 0 12
Bs 2 2 3 0 0 0 0 0 4 0 6] 17
Re 0 0 0 0 0 0 0 0 0 8 0 8
Cw 0 1 3 0 0 0 0 0 0 0 1 21
234 | 174 | 291 | 105 2 14 53 14 7 10 32 | 936
PA | 77.4 | 93.1 | 86.6 | 80.0 | 100 | 100 | 52.8 | 64.3 | 57.1 | 80.0 | 53.1
UA | 86.2 | 100 | 77.3 | 100 | 100 | 66.7 | 38.4 | 75.0 | 23.5 | 100 | 81.0
Overall: 81.3 %
Table 3. Error matrix for the sequential masking
classification.
5. RESULTS AND DISCUSSION
The producer's and user's accuracies and the overall
accuracy for the classification of all bands of the May, July,
and August images are summarized in table 1. The overall
accuracy of the May image (88.9%) was the highest. The
overall accuracies for the July and August images were
68.8% and 70.8% respectively. On May image, bare soil had
a producer’s accuracy of 94.2% and a user’s accuracy of
97.9%. The producer’s and user’s accuracies of wheat were
found to be 86.8% and 99.4% respectively. Among the
remaining classes, onion had the lowest user’s accuracy. Of
the two clover fields, one was omitted. All pasture fields
were correctly classified. However, some confusion is
evident between pea and pasture. Both the producer’s
accuracy (50%) and the user’s accuracy (62.5%) of rice were
below the overall accuracy. The user's and producer's
accuracies of pea were 81.8% and 60% respectively. Some
195
confusion is evident between the pea class and the onion and
pasture classes.
The overall accuracies of the July image (66.8%) and the
August image (70.8%) were lower than the May image. A
large number of fields (446) that were classified as bare soil
on May image contain crops at their active growth phase on
July and August images. Tomato, pepper, sugar beet, and
corn present an important vegetative development in the
June-August period. On July image, residue had the highest
producer’s accuracy of 93.1% and a very high user’s
accuracy of 92.6%. Both rice and sugar beet also exhibit
significantly high accuracies. Pepper, uncultivated land,
onion, and watermelon exhibit a significant amount of errors
of commission. The producer’s and user’s accuracies of
tomato and pasture were above the overall accuracy. The
user’s accuracy of clover was computed to be 100% but the
producer’s accuracy (50%) was significantly low. Similarly,
corn exhibits a relatively high user’s accuracy (76.1%) and a
rather low producer’s accuracy (56.1%). The classification of
the August image indicates that the producer’s and user’s
accuracies of corn, tomato, pepper and watermelon increased.
The user’s accuracy of residue and uncultivated land
improved but their producer’s accuracies decreased. The
reverse trend is observed for pasture, rice, and cauliflower
which exhibit higher producer’s accuracy and lower user’s
accuracy in the August image classification than the July
image classification. The August image classification
revealed no difference in the accuracies of clover.
The results of the first 4 PCs classification (Table 2) reveal
that the overall accuracy of the May image (88.4%) is the
highest. Similar to all bands classification (Table 1) the
overall accuracies of both the July and August images are
remarkably lower than that of May image. When we compare
the overall accuracies of the PCs classification to that of all
bands classification it is evident that the PCs classification
accuracies are slightly lower. The July image shows the
maximum decrease of 5.4%.
The multi-temporal masking classification yields better
results (Table 3) than both the all bands classification and the
first 4 PCs classification of the July and August images
alone. The overall accuracy of the masking classification was
computed as 81.3%. When compared to the classification of
the July and August images alone the improvement in overall
accuracy of more than 10% is evident. This improvement in
accuracy with the masking procedure is due to the use of
different spectral responses of the crops over a period of time
according to phenological evolution. The accuracy of the
clover, residue, sugar beet, rice, tomato, and corn classes
were quite high. The producer’s and user’s accuracies of
clover were 100%. Residue had a user’s accuracy of 100%
and a producer’s accuracy of 93.1%. It was found that of the
174 reference fields, only 12 were omitted from the residue
category. The user’s and producer’s accuracies of sugar beet
were 100% and 80% respectively. It is evident that sugar beet
does not exhibit commission error. However, the exclusion of
21 fields out of 105 from this class resulted in 20% omission
error. The user’s and producer’s accuracies of rice were also
computed as 100% and 80% respectively. Of the total ten
fields, eight were correctly classified through masking
classification procedure and two were omitted from rice.
Tomato had a producer’s accuracy of 86.6% and a user’s
accuracy of 77.3%. Of the 291 reference tomato fields, 252
were correctly classified. Tomato is most often confused with
corn and pepper, and additional imagery can improve
