The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 200H 
290 
Feature extracted Methods 
Miss classification (%) 
All bands 
8.77 
PCA 
9.04 
LDA 
8.05 
Ploy-KPCA d = l,p = 0 
11.84 
RBF-KPCA <j 2 = 10 7 
16.81 
Ploy-GDA d = 2,p = 0 
7.46 
RBF-GDA cr 2 =10 7 
1.46 
RBF-GDA cr 2 =10 8 
3.51 
Table 3. The precision of classification with features extracted 
with different methods. 
We evaluated the classification precision with the testing 
samples, using the minimum distance classifier. The 
classification result with the feature extracted by RBF-GDA 
was shown in Figure 6. The classification result of different 
feature extraction methods is shown in Table 3. 
4. CONCLUSION 
Through the experiments of feature extraction with AVIRIS and 
PHI images we made some conclusions. 
The PCA is to find project directions, which can make the 
samples variance maximized. The KPCA, using the kernel 
function, can realize the information compression to a great 
extent, but it is not good for classification. 
When the kernel function and its parameters are correctly 
selected, in the feature space extracted by GDA, the samples of 
the same class are near with each other; the samples of the 
different classes are far away. The GDA is a feature extracting 
method which is more suitable to classification than the LDA. 
When the kernel function and its parameters are correctly 
selected, the classification precision is much better with the 
features extracted by GDA, than the features extracted by other 
methods. How to select the kernel function and find suitable 
parameter is our further research. 
REFERENCES 
Tong, Q.X., Zhang, B., Zheng, L.F., 2006. Hyperspectral 
Remote Sensing. Higher Education Press, Beijing, pp. 363-470. 
Hoffbeck J.P., Landgrebe D.A., 1996, Classification of Remote 
Sensing Images having High Spectral Resolution. Remote 
Sensing of Environment, 57(3), pp.l 19-126. 
Zhang, L.P., 2003. Study of Feature Extraction and 
Classification of Hyperspectral Remote Sensing Image Based 
on Projection Pursuit and Nonlinear Principle Curves. 
Shandong University of Science and Technology, pp.1-8. 
John S.T., Nello C., 2004. Kernel Methods for Pattern Analysis. 
Cambridge University Press. Cambridge, pp.l-24. 
Lu, W., 2005. A Research of Feature Extraction and 
Classification Techniques for Target Detection in 
Hyperspectral image. Information engineering University. 
Zhengzhou, pp.l21-137. 
Baudat, G., Anouar, F. 2000.Generalized discriminant analysis 
using a kernel approach. Neural Computation, 12(10), pp. 
2385-2404. 
Gao X.M., Yang, J.Y., Jin, Z., 2004, Kernel-Based Foley- 
Sammon Discriminant Analysis and Face Recognition. Journal 
of Computer-Aided Design & Computer Graphics, pp. 962-967. 
Li W.H., 2003, Mechanical Fault Feature Extraction and 
Classification Based on Kernel Methods. Huazhong University 
of Science & Technology, pp.30-40.