MRF-Based Multiple Classifier System for Hyperspectral Remote Sensing Image Classification

Hyperspectral remote sensing image (HRSI) classification is a challenging problem because of its large amounts of spectral channels. Meanwhile, labeled samples for supervised classifier is very limited. The above two reasons often lead to unstable classification result and poor generalization capacity. Recent research has demonstrated the potential of multiple classifier system (MCS) for producing more accurate classification result. In addition, another vital aspect of HRSI classification is spatial contents. Markov random field (MRF), which takes the spatial dependence among neighborhood pixels based on the intensity field from observed data into consideration, is always adopted as an effective way to integrate the spatial information. In this paper, we proposed an effective framework for classifying HRSI image, called MRF-based MCS, which are based on the aforementioned two powerful algorithms. The proposed model is validated by multinomial logistic regression (MLR) classifier. Experimental results with hyperspectral images collected by the NASA Jet Propulsion Laboratory’s Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS) demonstrate that MRF-based MCS is a promising strategy in the context of hyperspectral image classification.