Combining classifiers for multisensor data fusion

Learn++ was recently introduced as an ensemble of classifiers based incremental learning algorithm, capable of retaining formerly acquired knowledge while learning novel information content from new datasets without requiring access to any of the previously seen data. In this contribution, we discuss the conceptual similarity between incremental learning and data fusion, the latter also requiring learning from new data, albeit composed of a different set of features. Following the technical description of the algorithm, we present our recent promising results on a realworld data fusion application of non-destructive evaluation for pipeline defect identification.