International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
  
May July August 
PA | UA PA UA PA UA 
(%) | (7) (e) | (%) (%) | (%) 
  
Bs. 1942|979]C€r 156.11 76.1 [Cr | 76.0 | 87.4 
wt [86.8|994|Rs |93.1| 92.6 |Rs | 70.2 | 98.9 
ti [509.0] 100 | Tm | 71.9] 65.3 | Tm | 854 | 72.3 
Pa 100 |64.7|Sb |78.3] 95.4 |Sb | 60.4 | 95.3 
Re 1300/6272. 5101 15007100 [CI | 500 1100 
Pe 160.0|818[Pa |66.7| 83.3 | Pa 100 |40.0 
On |60.0|10.0|Pp [16.4] 18.0 [Pp | 50.9 | 32.9 
wml 77 | 250 | Wm! 37.1 | 800 
Bs.1800] 98 |Bs | 224 | 429 
Rc 80.0 100 | Rc 90.0 | 75.0 
Cw | 3443] 480 [Cw | 57.1 [270 
On 50.0 7.0 
Overall: 66.8 
  
  
  
  
  
  
  
  
  
  
Overall: 88.9 
  
  
  
  
  
  
  
  
  
Overall: 70.8 
  
  
  
  
Tablel. Producer’s and user’s accuracies of individual classes 
for the classification of all bands. 
Bs: Bare Soil; Wt: Wheat; Cl: Clover; Pa: Pasture; Re: Rice; 
Pc: Pea; On: Onion; Cr: Corn; Rs: Residue; Tm: Tomato; 
Sb: Sugar Beet; Pp: Pepper; Wm: Watermelon; Cw: 
Cauliflower; PA: Producer’s Accuracy; UA: User’s Accuracy 
  
  
May July August 
PA UA PA UA PA UA 
(%) | (%) (%) | (%) (%) | (%) 
  
Bs | 925 | 933 | Cr 43.9: | 67.7: {Cr 76.4 | 87.4 
Wt | 85.8 | 100 | Rs 92.0 | 93.0 | Rs 67.2 [98.9 
CI 100 | 100 | Tm 661] 6135 | Tm 83.7 | 70.7 
Pa 100 | 61.1 | Sb 80.2 | 95.5 | Sb 54.5 | 96.5 
Re | 60.0 | 54.5 | CI 50.0 | 100 | CI 50.0 | 50.0 
Pe 622 | 84.8 | Pa 60.0 | 69.2 | Pa 100 
On | 70.0 | 10.6 | Pp 14.5 | 13.6 | Pp 49.1 
Wm 7.7 10.0 | Wm | 643 
Bs 40.01 3.8 | Bs 23.9 
Rc 70.0 | 50.0 | Rc 80.0 
Cw 28.6 | 34.5 | Cw 51.4 
On 500 | 85 
Overall: 61.2 
  
  
  
  
  
La 
> 
do 
  
LI 
  
  
  
I | ON 
$^ |$5|A3 $|- 
NOW 
  
Overall: 88.4 
  
  
oc 
  
N 
  
  
  
  
  
Overall: 69.2 
  
  
  
  
  
Table 2. Producer’s and user’s accuracies of individual 
classes for the classification of the first 4 PCs. 
Next, the sequential masking classification was performed. 
To. design the classification procedure, we analyzed the 
producer's and user's accuracies of individual crops and 
determined those classes with high accuracy. The thresholds 
of 80% and 90% were defined for producer’s and user’s 
accuracies respectively. The threshold for the user’s accuracy 
was kept higher for avoiding wrong masking of the fields 
belonging to other classess due to commission error. The 
accuracies of residue and rice were higher than the thresholds 
on the all bands classification of the July image. The 
producer’s and user’s accuracies of residue were 93.1% and 
92.6% respectively. Rice had a user’s accuracy of 100% and 
a producer’s accuracy of 80%. The other classes that meet the 
preset accuracy criteria are clover and sugar beet. While the 
producer's and user's accuracies of sugar beet were 80.2% 
and 95.5% respectively on the classification of the first 4 PCs 
of the July image, clover had the highest producer’s and 
user’s accuracies of 100% on the classification of the first 4 
PCs of the May image. The considerably high classification 
accuracies of residue, rice, clover, and sugar beet can be 
attributed to their phenological evolution. Therefore, a 
sequential-masking classification seems to be more suitable. 
It allows us to perform step-by-step classification of the 
classes using the multi-date images, excluding after each 
classification the class properly classified. 
The main steps of the proposed sequential classification 
procedure is illustrated in figure 2. We should indicate that 
the areas other than agricultural fields, such as roads, 
channels, and villages were not included in the classification 
and were eliminated by masking them out. The classification 
procedure was carried out as follows: First, clover was 
masked out on the classification of the first 4 PCs of the May 
image. Those fields classified as clover were excluded prior 
to further classification and the class training corresponding 
to clover was taken out. After this operation, the reduced 
class tranings were used to perform further classification. 
Next, those fields labeled as residue on the classification of 
the all bands of the July image were masked out. The reduced 
class training-set will now contain all crops except clover and 
residue. Further classification will now be performed over 
unmasked fields only. After that, those fields corresponding 
to rice on the classification of the July image were masked 
before proceeding the classification procedure. The last class 
to be masked was sugar beet. Finally, the remaining crops 
(corn, tomato, pasture, pepper, watermelon, bare soil, and 
cauliflower) were classified using all the bands of the August 
image through their training statistics only. 
  
Classification of the first 4 
PCs of the May Image 
Y 
Clover 
Mask 
Y Clover 
All bands classification 
of the July Image 
* 
Residue and Rice 
Classification of the 
first 4 PCs of the July 
Sugar Beet 
Mask Sugar 
y Beet 
All bands classification 
of the August Image 
Y 
Classified Output 
Y 
Field-Based 
Analysis 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Mask Residue 
and Rice 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Figure 2. The sequential masking classification 
procedure. 
Internation 
AE 
After comp 
classified o 
image. À 1 
final classi 
within each 
modal clas 
classificatic 
masking cl 
reference d 
individual 
illustrated ii 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Figure 3. 
Cr R 
Cr 181 
Rs 0 | 1€ 
Im 33 
Sb 0 
CI 0 
Pa 0 
Pp 18 
Wm 0 
Bs 2 
Rc 0 
Cw 0 
234 | 17 
PA 77.4 | 93 
UA | 86.2 | 10 
Table 
The produ 
accuracy foi 
and August 
accuracy of 
overall acci 
68.8% and 
a producer” 
97.9%. The 
found to b 
remaining c 
the two clo 
were corre: 
evident bet 
accuracy (5( 
below the 
accuracies c