Classifier Ensemble by Semi-supervised Learning: Local Aggregation Methodology

Abstract

A novel approach for automatic mine detection using SONAR data is proposed in this paper relying on a probabilistic based fusion method to classify SONAR instances as mine or mine-like object. The proposed semi-supervised algorithm minimizes some target functions, which fuse context identification, multi-algorithm fusion criteria and a semi-supervised learning term. Our optimization purpose is to learn contexts as compact clusters in subspaces of the high-dimensional feature space through probabilistic feature discrimination and semi-supervised learning. The semi-supervised clustering component appoints degree of typicality to each data sample in order to identify and reduce the influence of noise points and outliers. Then, the approach yields optimal fusion parameters for each context. The experiments on synthetic datasets and standard SONAR dataset illustrate that our semi-supervised local fusion outperforms individual classifiers and unsupervised local fusion.