Feature Reduction and Classification of Hyperspectral Image Based on Multiple Kernel PCA and Deep Learning

Abstract

In recent years, the classification of Hyper Spectral Image (HSI) is a big challenge for its multidimensional property. So it is burning question to reduce the dimension of HSIs. There are several ways to reduce the dimension of hyperspectral images like Principle Component Analysis (PCA), Kernel Principle Component Analysis (KPCA), Kernel Entropy Component Analysis (KECA) and so on. In this paper, we proposed a modified version of KPCA using multiple kernels like Linear, Radial Basis Function (RBF), Cosine, Sigmoid. Then fused their spectral and special properties by doing the classification of the HSIs using Hybrid Spectral Net (HybridSN) Model which is a recently trending modified deep neural network algorithm of Convolutional Neural Network (CNN). Finally, this paper demonstrates experimental results to show the effects and performance on classification of using different kernels of KPCA algorithm with other algorithms such as Non-negative Matrix Factorization(NMF), Independent Component Analysis (ICA) and Singular Value Decomposition(SVD) on well-known hyperspectral dataset.