|
|
|
Hyper Spectral Image Processing
|
Extraction and Recognition of Urban Objects
by Hyperspectral Remote Sensing
4. Result and Discussion
By using the sketch expressed in the last section, the image was classified into 17 classes. Figure 4 shows the classification result. We can see that water body and coal area is discriminated pretty well. Freeway and asphalt road, although made with same material, are spectrally different because the freeway was built later than asphalt road, the dust and maintenance conditions are different, so they can be discriminated. The partition between two strips of freeway is very completed, it is the mixture of vegetation, soil, concrete and metals which can not be classified into any class. The major land cover and metals types are cement concrete surface and oil paper roofs, with several kind of distinct roofs such as red tile roof, white roof etc. as a comparison to the traditional one layer classifiers, maximum likelihood classifier (MLC) was also used. MLC is effective to discriminate classes with prominent Spectral differences, but is not so to classify the classes with minor Spectral differences, such as water soil and coal area red tile and grayish red tile roofs. The classification accuracy was assessed using error matrix (Congalton, 1991). The accuracy of classification is great than 73% for most of the classes.
Figure 4 Classification result
It must be accounted for that the data were analyzed only based on relative reflectance. If the data id radiatively corrected and convert into reflectance, we can use the spectral library to recognize the different cover types by spectral analysis. This is feasible because hyperspectrla data have detailed spectral information about the materials.
5. Conclusion
Cover types recognition and classification is essential in remote sensing application. In the urban studies by remote sensing, high accuracy classification of cover types is a elementary step for further studies. Hyperspectral data contains fire spectral information of ground objects, this enables the spectral recognition of urban land cover types. Spectral recognition of urban land cover types. Hierarchical and masking technique is an effective tool that can classify the complex urban objects by spectral analysis. The major advantages of the method lie in that, it allow classification procedure goes from easier to difficult, and for the particular class pair different exist method of feature extraction and enhancement can be adopted.
6. References
-
E. Ben-dor, N. Levin, H. Saaroni, 1999, Remote Sensing of an Urban environment Using Hyperspectral Technology, Proceedings of the Thirteenth International Conference on Applied Geological Remote Sensing. 1-3 March 1999, Vancouver, BC. Canada, 116-11.
-
Congalton, Result G., 1991, A review of Assessing the Accuracy of Classification of Remotely Sensing Data. Remote Sensing of Environment, 37:35-46(1991).
-
Xiuping Jia and J. A. Rechards, 1998, Progressive Two-class Decision Classifier for Optimization of Class Discriminations. Remote Sensing of Environment, 63:289-297 (1998).
-
Liu Jiangui, Recognition and Extraction of Urban and Man-Made Objects Using Hyperspectral Data: Unpublished Ph. D. Dissertation. Institution of Remote Sensing Application. CAS,S.R. China , 1999.
-
J.A. Richards , 1994, Remote Sensing Digital Image Analysis, Springer-Verlag, Berlin, p.340.
-
S.R. Safavian, D.A. Landgrebe, 1990, A Survey of decision tree classifier methodology. IEEE Trans. On system, Man, and Cybernetics, Vol. 21. 660-674 (1990).
|
|
|
|
|
|
|
