A new ensemble approach for hyper-spectral image segmentation

Abstract

The ensemble is an universal machine learning method that is based on the divide-and-conquer principle. In data clustering, ensemble aims to improve performance in terms of processing speed and clustering quality. Most existing ensemble methods become more difficult due to the inherent complexities such as uncertainty, vagueness and overlapping. In this paper, we proposed a new ensemble method that improve the ability to identify uncertainty issues, deal with the noise, and accelerate hyperspectral image data clustering. We called fuzzy co-clustering ensemble algorithm (eFCoC). EFCOC uses fuzzy co-clustering algorithm (FCOC) to clustering data and silhouette-based assessment of custer tendency algorithm (SACT) to ensemble the final clustering result. Experiments were conducted on synthetic data sets and hyperspectral images. Experimental results demonstrated the key properties, rationality, and practicality of the proposed method. Index Terms-Fuzzy co-clustering, clustering ensemble, assessment of cluster tendency, hyper-spectral image, image segmentation