Hyperspectral image classification is an emerging area of research in the remote sensing domain. Dimensionality reduction is an crucial step to increase efficiency of the learning algorithm significantly. This chapter first focuses on the application of Biogeography based optimization (BBO) technique to effectively reduce the dimensionality of hyperspectral images. To assess the goodness of the band selection technique, SVM is used to classify the images. Following that, extensive experiments have been conducted with different Convolutional Neural Network (CNN) models to do classification on the selected (through BBO) set of bands. Various CNN architectures are tested for the task like 2D CNN, 3D CNN, Multi-Dimensional CNN and Diverse Region Based CNN. Qualitative and quantitative analysis of these methods have been made, which give prominent results in terms of hyperspectral image classification.