Approaching Semantically-Mediated Acoustic Data

Our primary hypothesis is that it should be possible to enrich data fusion by semantic processing, with wide potential application. In order to achieve our aim we need to represent the semantic data and enable reasoning about it in a framework that can be aligned with data fusion. Ontologies are most suited to this task as they allow for appropriate representation of data structure; some approaches include probabilistic representation. These can be aligned with data fusion approaches, such as Bayesian, which can fuse by including estimates of uncertainty. We shall describe our initial approaches towards establishing our hypothesis. We shall survey the enabling technologies, showing how they can contribute to our goal. We shall describe our selection of application data which derives from an acoustic sensor (military) scenario. We shall show how feature subset selection can reduce information-redundancy and improve efficiency in these domains, prior to fusion to enhance performance further. We shall explore the semantic attributes and the representations that can be deployed for enrichment purposes, showing how ontologies can be used in this context. In these respects we are aiming to show how we can approach enrichment of data fusion by semantic technologies, how this can capitalise on the current stock of techniques, and illustrate the potential benefits associated with this new approach.