Abstract
The application of principal component analysis (PCA) for enhancement and compression of multispectral images provided by remote sensing satellites involves the computation of the spectral image covariance matrix and its spectral decomposition to extract their eigenvalues and corresponding eigenvectors. When the size of the imaged scene and/or the number of spectral images grows significantly, the computation of the covariance matrix and its decomposition become practically inefficient and inaccurate due to round-off errors. These deficiencies make the conventional scheme of PCA inefficient for this application. We propose here a neural network model that performs the PCA directly from the original spectral images without any additional non-neuronal computations or preliminary matrix estimation. Since the end user is usually not a neural network specialist, the neural network model, as well as its execution, are carefully designed in order to be independent from the user as much as possible. This includes both the design of the network topology and the input/output representation as well as the design of the learning algorithms. The global convergence of the model is studied. Its application has been realized on real multispectral image provided by Landsat-Thematic Mapper satellite. The obtained results show that the model performs well. A comparative study between the proposed model and the standard method of PCA is undertaken. The study has shown that the proposed method is superior to the standard method of PCA.