The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
289 
3 
(4) Image of the forth feature 
Figure 2. The feature images extracted by RBF-GDA 
( cr 2 = 10 7 ) 
We evaluated the classification precision with the testing 
samples, using the minimum distance classifier, and the result 
was shown in Table 2. The classification result with the feature 
extracted by RBF-GDA was shown in Figure 3. 
Figure 4. Samples distribution in this PHI image 
We assigned the samples each class randomly as the training 
samples and testing samples equally. The feature was extracted 
by different feature extraction methods. In the feature space, the 
distribution of samples was shown in Figure 5. 
Figure 3. The classification result with feature extracted by 
RBF-GDA (CT 2 = 10 7 ) 
Feature extracted Methods 
Miss classification (%) 
All bands 
23.87 
PCA 
25.7 
LDA 
23.84 
Ploy-KPCA d = l,p = 0 
40.1 
RBF-KPCA a 2 = 10 7 
19.22 
Ploy-GDA d = 2,p = 0 
7.53 
RBF-GDA <j 2 =10 7 
3.75 
RBF-GDA cr 2 =10 8 
4.83 
(1) PCA 
(2) Ploy-KPCA (d = l,p = 0) 
Figure 5. Samples distribution in different feature space 
We assigned the samples each class randomly as the training 
samples and testing samples equally. The feature was extracted 
by different feature extraction methods. In the feature space, the 
distribution of samples was shown in Figure 5. 
Table 2. The precision of classification with features extracted 
with different methods. 
3.3 Experiment 2 
Experiment Data: The PHI instrument, created in Shanghai 
Institute of Technology and Physics, acquired data over 
Changzhou, Jiangsu, China, (E119°22'll", N31°41'44"). PHI 
acquires data in 80 bands width with centre wavelengths from 
0.42-0.85pm, and the size of the image is 346X512. 
Six kinds of objects exist in the image: (Colour-Class of the 
target-Amount of sample): 1-house-221, 2-water-222, 3-soil- 
205, 4-tree-228, 5-vegetation-266, 6-road-238, the results are 
visualized in figure 4. 
Figure 6. The classification result with feature extracted by 
RBF-GDA (cr 2 = 10 3 )