International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
4. CLASSIFICATION RESULTS 
4.1 Classification Results for SIR-C Data 
Using SIR-C Data, three amplitude images for HH, HV, and 
VV polarization can be obtained for two wavelength band: L- 
band and C-band. In this section, supervised classification 
results are shown for two different feature vectors. One of them 
has six elements that are pixel values of six amplitude images, 
three kinds of polarization for two kinds of wavelength band. 
Another of them has twelve elements that are six elements from 
amplitude images and other elements are obtained Rajski 
distance images. As the classification technique, Euclid 
distance method and maximum likelihood method. 
The classification score matrix for Sarobetsu site is shown in 
Table 1. And the matrix for Kashima site is shown in Table 2. 
In these table, table (a) and (b) are the case of six elements, 
table (c) and (d) are the case of twelve elements. And table (a) 
and (c) are the results of Euclid distance method, table (b) and 
(d) are the results of maximum likelihood method. There 
numbers shows the score after otherwise class was excluded 
from test areas. 
In these tables, category names are defined under the tables. 
Category names of column direction, Cl, C2, and so on, mean 
the categories before classification and ones the names of row 
direction, Cl, C2 , and so on, mean the categories after 
classification. And the numbers in ( ) mean pixel numbers 
classified to the categories. The sum of scores to row direction 
will be 100%. 
In Table 1, average classification accuracy was 70.92% for (a), 
75.31% for (b), 78.86% for (c), and 88.28% for (d), respectively. 
Table 1. Classification score matrix for Sarobetsu site 
(Unit: 96) 
(a) Euclid distance method (six elements) 
  
  
  
  
  
  
  
  
C1 C2 C3 C4 CS C6 
cil 746 4i 210 0.1 0 0.1 
(546) (30) (184) (1) (0) (1) 
en 0. 864 0.9 02 117 0.8 
(0) (67985) (70) (18) (925) (61) 
e 8S1 427 435 23 1.5 1.5 
7 | (430) 0261) (2305) (445) (82) 077) 
e Ü 381 01 583 0 3.5 
(0) (444) (1) (680) (0) (41) 
C 0 0 0 0 100.0 0 
(0) (0) (0) (0) (144) (0) 
(6 0 :343 0 0.3 01 3534 
(0) (472) (0) (4) (D (804) 
(b) Maximum likelihood method (six elements) 
C1 C2 C3 C4 C5 C6 
al 768 0 232 0 0 0 
(562) (0) (170) (0) (0) (0) 
à 01 764 160 0.1 0 74 
(4) (6018) (1258) (8) (0) (584) 
51 Bi MT — 709 0 0 2.8 
7^ | (694) (748) (3709) (0) (0) (149) 
C41 0 4.9 07 .— 941 0 0.3 
(0) (57) (8) (1097) (0) (4) 
cs 0422 0 0. 528 0 
(0 (68) (0) (0) (76) (0) 
; 0 — 147 22 0 0: 761 
C6| © (2035) Q0 (09 (0 (104) 
  
  
24 
(c) Euclid distance method (twelve elements) 
  
  
  
  
  
  
  
C1 €? €3 C4 C5 C6 
Cl 70.7 1.0 28.3 0 0 0 
(442) (6) (177) (0) (0) (0) 
© 0 97.4 1.8 0 0 0.8 
(0) (6486) (118) (0) (0) (55) 
C3 24 32.1 64.4 0.3 0 1.1 
(108) (1640) (3285) (13) (0) (54) 
CA 0 10.1 0 89.9 0 0 
(0) (115) (0) (1025) (0) (0) 
CS 0 14.3 0 0 85.7 0 
(0) (20) (0) (0) (120) (0) 
C6 0 34.8 0 0.2 0 65.0 
(0) (410) (0) (2) (0) (766) 
(d) Maximum likelihood method (twelve elements) 
C1 C2 C3 C4 Cs C6 
Cl 82.1 0 17.9 0 0 0 
(513) 0) 112) (0) (0) (0) 
C2 0 90.3 5.8 0 0 3.9 
(0) (6012) (387) (0) (0) (260) 
C3 1.5 12.7 85.5 0 0 0.3 
(76) (647) (4363) (0) (0) (14) 
CA 0 1.4 0 98.6 0 0 
(0) (16) (0) (1124) (0) (0) 
CS 0 15.0 0 0 85.0 0 
(0) Q1) (0) (0) (119) (0) 
C6 0 11.3 0.5 0 0 88.2 
(0) (133) (6) (0) (0) (1039) 
  
  
Cl: urban, C2: grass, C3: forest, C4: sea, 
C5: opened lake, C6: frozen lake 
In Table 2, average classification accuracy was 73.91% for (a), 
66.79% for (b), 87.57% for (c), and 84.47% for (d), respectively. 
In the both sites, classification accuracy was improved by 
increasing the elements of feature vector. 
Table 2. Classification score matrix for Kashima site 
(Unit: %) 
(a) Euclid distance method (six elements) 
  
  
  
  
  
Ci Co C3 (4^ CS. C6 C7 
al? 03 (Qa 01 31.5 0 0 
(374) Q) (83) (D- 212) (0) (0) 
co 0:785 02 213 0 0 0 
(0) (439) (1)... (119) (0) (0) (0) 
cl 5 Ll] 379 28 08 0 0 
(290) (41) (3421) (108) GO.  Q) (0 
ul 02 45 25 $0 06 Ü 59 
(4j (09 (68) Cle (4) (0 (22 
c MO $7. 49 Ü 784 0 0 
(54) | (28) (24) (0) (385) (0) (0) 
C5 0 0 0 0 Ü 664. 336 
(0) (0) (0) (0) (0) (3255) (1645) 
c 0 0 0 0 0 366 634 
(0) (0) ©) ©) 0) OSO) (440 
  
Intern 
CS 
C6 
C7 
  
Cl 
C2 
C3 
C4 
CS 
C6 
(7 
  
Ci 
C 
C3 
C4 
C5 
C6 
€7 
  
  
  
4.2 € 
Pi-SA 
multi- 
Howe 
only t 
data, | 
been : 
for tw 
classi