Band priortization and feature extraction of remotely sensed image

Remotely sensed images are composite images made of spectral bands ranging over Visible, Near Infrared, Middle Infrared and Far Infrared region of electromagnetic spectrum. Pixel based classification of high dimension multispectral images are complex, as the process takes lot of time and dataset is large. In this paper, Supervised Classification of Multispectral image is performed after band selection and feature extraction. Band selection is achieved by Canonical Analysis which prioritizes the bands by computing their discriminating power. It uses the concept of simultaneously maximizing distance of pixels between the classes and minimizing the distance of pixels within a class. Once dimension of the image is reduced, textural and spectral features are extracted from the image. Textural features are extracted using GLCM. Supervised classification using Feed Forward Neural Network architecture is accomplished. The features extracted are used to train the neural network and predict the classes of testing samples.