Multispectral Image Classification Using Back-propagation Neural Network in Pca Domain

Recently, in classification of multispectral remote resensing image by using back-propagation neural network (BPNN), all bands of image must be used for training and classing. Disadvantage of the mentioned method not only requires more time for training and classing but also more complexity. In this paper, to decrease the mentioned disadvantage, principal component analysis (PCA) is applied to reduce dimensionality of multispectral remote sensing image as preprocessing. The first three principal components which contain information more than that of original images of 95 percents are then used for training and classing. Landsat 7 satellite TM image in visible bands of 6 is implemented to test results. We compare results of the classified multispectral remote sensing image as the proposed method with those of one as maximum likelihood classifier with principal component analysis (MLC-PCA) in term of accuracy percentage. Our results show that classification using the threelayer back-propagation neural network with principal component analysis (BPNN-PCA) is better than MLC-PCA and also it is lower complexity certainly.