Spectral-Spatial Classification of Hyperspectral Imagery Using Support Vector and Fuzzy-MRF

摘要

Hyper-Spectral Image (HSI) classification is one of the essential problems in hyperspectral image processing. It has been researched extensively and has resulted in a variety of publications. A key approach investigated in recent years incorporates both spectral and spatial characteristics to analyze the hyperspectral data. In this paper we have presented our proposed approach to improve the accuracy of HSI classification. Support Vector Machines have been used to classify spectral characteristics of images in conjunction with Markov Random Fields that classify HSI using spatial means. However, this current technique of combining them does not enforce smoothness in spatial and spectral analyses. We ensure finer segmentations in the results by adding our innovative approach of including Fuzzy-Markov Random Field to spectral classification. The 'fuzziness' promotes smoother transitions among classified pixels while preserving region integrity. Results show the efficacy of our approach.