|
Hyperspectral & Data Acquisition Systems
|
Vegetation Spectral Feature Extraction Model
Table 2 lists correlation matrix of 8 feature positions , from Table 2 , we can get priority order for 8 feature positions : first , select two positions with minimum correlation coefficient (absolute value) , remove these two bands and those bands with highest correlation coefficient with them , then , select repeatedly for the rest bands until we get to band number we predetermine.
Table 2 correlation coefficient of 8 feature positions (bands)
Corres
ponding feature position |
M |
B |
G |
Y |
R |
V |
I1 |
I |
Wave
length(nm) |
404.7 |
522.4 |
544.7 |
573.4 |
673.1 |
723.4 |
759.0 |
890.1 |
| 404.7 |
|
0.8890 |
0.8952 |
0.8372 |
0.4575 |
0.8408 |
0.6561 |
0.6117 |
| 522.4 |
0.8890 |
|
0.9895 |
0.9925 |
0.7071 |
0.8265 |
0.5520 |
0.5185 |
| 544.7 |
0.8952 |
0.9895 |
|
0.9812 |
0.6161 |
0.8785 |
0.6364 |
0.6035 |
| 573.4 |
0.8372 |
0.9925 |
0.9812 |
|
0.7455 |
0.7960 |
0.5053 |
0.4758 |
| 673.1 |
0.4575 |
0.7071 |
0.6161 |
0.7455 |
|
0.3313 |
-0.0230 |
-0.0366 |
| 723.4 |
0.8408 |
0.8265 |
0.8785 |
0.7960 |
0.3313 |
|
0.9020 |
0.8806 |
| 759.0 |
0.6561 |
0.5520 |
0.6364 |
0.5053 |
-0.0230 |
0.9020 |
|
0.9947 |
| 890.1 |
0.6117 |
0.5185 |
0.6035 |
0.4758 |
-0.0366 |
0.8806 |
0.9947 |
|
Fig. 2 The simulated images of the correlation coefficient matrix among 187
bands(400-950nm) of Se590
from Table 2 , we can get priority order for 8 feature positions : first , select two positions with minimum correlation coefficient (absolute value) , remove these two bands and those bands with highest correlation coefficient with them , then , select repeatedly for the rest bands until we get to band number we predetermine.
4. Conclusion
According to above analysis , we find out that:
-
Through lots of analysis on field vegetation spectral data, A new spectral feature selection and extraction model(for vegetation only!)--- Vegetation Spectral Feature Extraction Model (VSFEM) is presented , with which 8 spectral feature positions are suggested to control vegetation spectral curves . Those feature positions are 404(M), 525(B), 556(G),573(Y),671(R),723(V),758(I1),900nm(I) .
- It is reasonable to select feature positions as best bands. However, the feature positions and bandwidths for different feature are not the same. As a result, the accurate positions of G, R, M, I are not necessary. They should change in a proper range, such as I and M which could use a broader band.
- According to the correlationship among bands, the bands could be removed are: I, M, B, Y and G. As for band R ¢ V ¢ I1, they should not be removed . If subdivision allowed , it's proper to subdivide red edge(R-V-I1 ) , moreover , we can subdivide around B and Y position .
Anyway , above results are acquired through spectral data by Se590 and some other conditions , we cannot reject absolutely new results under higher spectral resolution and other conditions .
References
-
Boardman, J. W., and Kruse, F. A., 1994,"Automated spectral analysis: a geological example using AVIRIS data, north Grapevine Mountains, Nevada.", Proceedings, ERIM Tenth Thematic Conference on Geologic Remote Sensing, Environmental Research Institute of Michigan, Ann Arbor, MI, p. I-407 - I-418.
- Fukunaga,K., Koontz,W.L.G., 1970, "Application of the Karhunen_Loeve expansion to feature selection and ordering." , IEEE Trans. Comput., Vol. C-19,No.4,pp311-318,Apr.1970.
- Green, A. A., Berman, M., Switzer, P, and Craig, M. D., 1988, "A transformation for ordering multispectral data in terms of image quality with implications for noise removal", IEEE Transactions on Geoscience and Remote Sensing, v. 26, no. 1, p. 65-74.
- Jia,Xiuping, Richards,J. A.,1999, "Segmented principal components transformation for efficient hyperspectral remote-sensing image display and classification.", IEEE Trans. On Geosci. And Remote Sensing, Vol.37, No.1, 538-542, Jan. 1999.
- Kazakos,D.,1978, "On the optimal linear feature." , IEEE Trans. Inform. Theory, Vol. IT-24,No.5,pp651-652,Sept. 1978.
- Lee, C. and Landgrebe, D. A.,1993, "Feature extraction based on decision boundaries.", IEEE Trans on P.A.M.I. Vol. 15, No.4, April 1993.
- Richards,John A., Remote Sensing Digital Image Analysis, An Introduction, Spriger-Verlag,1986.
Acknowledgement
The authors wish to thank Laboratory of Remote Sensing Information Sciences , Institute of Remote Sensing Applications for providing the data and their sincere help .
|
|
|
|
|
