Decision Fusion of Multiple Classifiers for Vegetation Mapping and Monitoring Applications by Means of Hyperspectral Data
In this chapter, we introduce methodologies for fusion of multiple classifiers and a neural network architecture for mapping vegetation by means of remote sensing imagery. It is very normal that different classification schemes yield slightly different results for different classes. This effect is even more prominent in vegetation mapping applications due to the inconsistent spectral signatures of the vegetation classes. We study the possibility of combining the results of different classifiers by considering the best results for individual classes to produce an improved classification result. We propose two types of methodologies, one which uses only the classification result and the other which uses the class membership values produced by the weak classifiers. A comparison is also done with the simple majority voting scheme of the multiple classifiers. Our experiments clearly show the improvement of classification accuracy using the proposed fusion techniques.
KeywordsHyperspectral Classification Neural Network Decision tree Ensemble
The authors would like to acknowledge that this research was carried out within the Marie-Curie founded Hyper-I-Net Research Training Network. We also would like to thank Antonio J. Plaza and Alberto Villa the help and support which they provided.
- 1.Gamba, P., Plaza, A., Benediktsson, A.J., Chanussot, J.: European perspectives in hyperspectral data analysis. In: Proceedings of 2007 IEEE Geoscience and Remote Sensing Symposium (IGARSS 2007), pp. 4794–4797 (2007)Google Scholar
- 3.Bakos, K.L., Gamba, P.: Potential of hyperspectral remote sensing for vegetation mapping in high mountain ecosystems. In: Proceedings of 6th EarSEL SIG IS Workshop 2009, Unformatted CD-ROM (2009)Google Scholar
- 6.Pearson, K.: On lines and planes of closest fit to systems of points in space. Philos. Mag. 2, 559–572 (1901)Google Scholar
- 8.Richards, J.A.: Remote Sensing Digital Image Analysis. Springer-Verlag, Berlin (1999)Google Scholar
- 9.Tadjudin, S., Landgrebe, D.A.: Classification of high dimensional data with limited training samples. Technical Report TR-ECE 98-8, Purdue University (1998)Google Scholar
- 11.Kuncheva, L.I., Bezdek, J.C., Sutton, M.A.: On combining multiple classifiers by fuzzy templates. In: Proceedings of NAFIPS Conference EDS, pp. 193–197 (1998)Google Scholar
- 14.Liu, W.G., Gopal, S., Woodcock, C.E.: Uncertainty and confidence in land cover classification using a hybrid classifier approach. Photogramm. Eng. Remote Sens. 70, 963–971 (2004)Google Scholar
- 18.Roli, F., Giacinto, G., Vernazza, G.: Methods for designing multiple classifier systems. In: Proceedings of MCS, pp. 78–87 (2001)Google Scholar
- 22.Canty, M.J.: Image Analysis, Classification and Change Detection in Remote Sensing, with Algorithms for ENVI/IDL. CRC Press, Boca Raton (2007)Google Scholar
- 23.Marpu, P.R., Gamba, P., Niemeyer, I.: Hyperspectral data classification using an ensemble of class-dependent neural networks. In: Proceedings of IEEE GRSS Workshop on Hyperspectral Image and Signal Processing Conference (WHISPERS), Unformatted CD-ROM (2009)Google Scholar
- 25.Lillesand, T., Kiefer, R.: Remote Sensing and Image Processing. Wiley, New York (1978)Google Scholar