Hyperspectral Image Classification Using Dynamic Classifier Selection with Multiple Feature Extractions

Abstract

Dynamic classifier selection is a strategy in multiple classifier system design. Feature extraction is one of the important procedures for mitigate Hughes phenomenon in hyperspectral image classification. Most papers have discussed the potential discriminatory information between different classifiers. In this paper, we try to exploit the discriminatory information extracted by different feature extractions for improving classification accuracy. Information is then combined by using a dynamic classifier selection strategy based on local information to make a consistency decision. This paper provides another thinking of constructing a multiple classifier system without additional classifier design by using multiple feature extraction.