Classification of High Resolution Satellite Images Using Equivariant Robust Independent Component Analysis
Classification approach helps to extract the important information from satellite images, but it is quite effective on extracting the information from mixed classes. The problem of classes is well known in the area of satellite image processing due to similar spectral resolution among the few classes (objects). It is quite obvious in multispectral data which is having little variation in spectral resolution with heterogeneous classes. In earlier times, Neural network based classification has been used widely, but at a cost of high time and computation complexity. To resolve the problem of the mixed classes due to the spectral behavior in sufficient time, a novel method Equivariant Robust Independent Component Analysis (ERICA) is proposed. This algorithm separates the objects from mixed classes, which shows similar spectral behavior. It can easily predict the objects without using of pre-whitening technique. Therefore, pre-whitening is not playing an important role in convergence of the algorithm. Due to Quasi-Newton based iteration in this algorithm helps to converge to a saddle point with locally isotropic convergence, regardless of the spatial and spectral distributions of satellite images. Hence, this proposed ERICA gives major contribution for classification of satellite images in healthy trees, buildings and road areas. Another important one in the image is shadow information, which helps to show elevated factor due to high rising buildings and flyovers in emerging suburban areas. The experimental results of the remote sensing data clearly indicates that the proposed ERICA has better classification accuracy and convergence speed, and is also appropriate to solve the image classification problems.
Keywordsquivariant Robust Independent Component Analysis Information extraction Image classification Mixed class Performance index
Unable to display preview. Download preview PDF.
- 1.Almeida, L.B., Silva, F.M.: Adaptive Decorrelation. In: Aleksander, I., Taylor, J. (eds.) Artificial Neural Networks, vol. 2, pp. 149–156. Elsevier Science Publishers (1992)Google Scholar
- 7.Cichocki, A., Georgiev, P.: Blind Source Separation Algorithms with Matrix Constraints. IEICE Trans. on information and Systems, Special Section on Special issue on Blind Signal Processing E86-A(1), 522–531 (2003)Google Scholar
- 8.Cruces, S., Castedo, L., Cichocki, A.: Novel Blind Source Separation Algorithms Using Cumulants. In: IEEE International Conference on Acoustics, Speech, and Signal Processing, V, Istanbul, Turkey, pp. 3152–3155 (2000)Google Scholar
- 9.Kelley, C.T.: Iterative methods for linear and nonlinear equations. In: Frontiers in Applied Mathematics, vol. 16, pp. 71–78. SIAM, Philadelphia (1995)Google Scholar
- 12.Singh, P.P., Garg, R.D.: A Hybrid approach for Information Extraction from High Resolution Satellite Imagery. International Journal of Image and Graphics 13(2), 340007(1-16) (2013)Google Scholar
- 13.Yang, J.H., Liu, J., Zhong, J.C.: Anisotropic diffusion with morphological reconstruction and automatic seeded region growing for colour image segmentation. In: Yu, F. (ed.) Proceedings of the International Symposium on Information Science and Engineering, Shangai, China, pp. 591–595 (2008)Google Scholar
- 14.Zhang, L., Amari, S., Cichocki, A.: Natural Gradient Approach to Blind Separation of Over- and Under-complete Mixtures. In: Proc. of Independent Component Analysis and Signal Separation, Aussois, France, pp. 455–460 (1999)Google Scholar