Hyperspectral Imagery Classification Based on Sparse Feature and Markov Random Field

摘要

Aiming at the deficiencies that some implicit information is not well reflected in the traditional hyperspectral image classification methods, this paper presents a new hyperspectral image classification method based on Sparse Feature and Markov Random Field (SFMRF). The core idea of the SFMRF is firstly to express the hyperspectral image with the sparse feature, and then to apply the Support Vector Machine (SVM) to estimate the class conditional probability density functions, also, the MRF method is used to estimate the context-based class priors. Finally, the final class labels are determined by the Maximum a Posteriori (MAP) method. Theory analysis and simulation results on the Indian Pines hyperspectral data show that the proposed SFMRF method can greatly improve the classification accuracy.