Urban Image Analusys Using Adaptive Resonance Theory
3. Experimental Resaults
Figure 3: Classification result of the area in Fig 1.
The classification was applied to a three-band image of Bangkok (Fig. 1). The image was recorded by the ADEOS satellite. We applied the previously presented neural classification algorithm to this data set by OPS image the result are completely satisfying. The neural classification starts by applying an ART network looking for spectral aggregation of the pixel. The three-band data are aggregate an input vector in order to match the ability of ART networks to discriminated against different pattern. Fig. 3 presents spectral classification result obtained by the ART2 algorithm and we found very difficult to tune the network parameters to obtain satisfying result.
4. Conclusion
This work simplifies and completes the ART approach to remote sensing data analysis introduced in (Silva, 1997). We apply the methodology to urban environments and multiband data introduce a clustering step to solve class redundancy and high lightening the advantages and disadvantages of data analysis. The results presented, corresponding to a part of data, are satisfying, and it can be using to analysis another image for classification by computer-aided.
5. Acknowledgement
The authors wish to thank the National Research Council of Thailand (NRCT) for providing the satellite image data
References:
-
Carpenter A. G., Grossberg S. and Rosen. B. D., 1991. ART 2-A:, an adaptive resonance algorithm for rapid category learning and recognition. neural network, (4), pp. 493-504.
-
Fausett L., 1994. Fundamentals of Neural Network Architecture Algorithm and Application. Prentice-Hall.
-
Silva S. and Caetano M., 1997. Using Artificial Recurrent Neural Nets to Identify Spectral and Spatial Pattern for Satellite Imagery Classification of Urban Area. In: Neurocomputation in Remote Sensing Data Analysis, I. Kanellopoulos et al. (eds.), Springer: Berlin, pp. 151-159.
