Mixed Pixel Wise Characterization Based on HMM and Hyper Spectral Image Gradient Enhancement for Classification Using SVM-FSK.

Hyperspectral image classification is one of the most important complicated tasks in recent years mainly due to the diversified pixels of present data in which only limited information is known prior. To solve this problem, in this work a combined pixel-wise characterization framework is applied to hyperspectral image samples. In this work, a mixed pixel-wise characterization is presented for utilization of mixing both the spectral and the spatial information. The proposed work comprises of three major steps: Spectral gradient enhancement, spectral-spatial classification obtained from the Support vector machine-Fuzzy Sigmoid Kernel (SVM-FSK) and probability estimation based on Hidden Markov Model (HMM). To improve the gradient level of spatial information, this work uses Improved Empirical Mode Decomposition (IEMD) with PSO (IEMD-PSO) to increase the mixed pixel wise SVM -FSK classification accuracy. Apply EMD method to enhance the gradient level of spatial data to separately identifiable of intrinsic mode functions (IMFs) of every one of spectral band, weight values of IMFs are calculated by using Particle swarm optimization (PSO).The obtained spectral and spatial information learns the probability value from HMM and mixed pixel-wise SVM-FSK characterization methods to find number of mixed components in each pixel. The proposed mixed pixel-wise probabilistic characterization for SVM-FSK and spectral gradient enhancement method IEMD-PSO shows improved classification accuracy in terms of parameters like overall accuracy, standard deviation and mean. The proposed methods can be compared with existing methods such as an Empirical Mode Decomposition (EMD), Empirical Mode Decomposition -Genetic Algorithm (EMD-GA) methods and Support Vector Machine -Radial Basis function (SVM-RBF). [ABSTRACT FROM AUTHOR]