Abstract
The ever-increasing amount of data means that today's decision-makers and analysts are often faced with an overwhelming amount of information, arriving in a variety of formats from multiple sources. This work addresses three interrelated challenges within information fusion - context exploitation, explainability and situation awareness and demonstrates their application to maritime situational awareness.In uncertain reasoning, scenario context can be defined as information relevant to the reasoning task at hand but not directly involved with it. This context can be exploited in the construction of the model. The benefit of exploiting context in the construction of the model is two-fold. On the one hand, its explicit representation can provide better insight into the problem. On the other hand, it may improve the expressiveness of the model and result in an inference that better represents available knowledge. Context exploitation can also improve the interpretability of the model and can be used to generate better explanations.
A reasoning process built upon evidential networks is transparent by design. Still, it can be not easy to follow due to many variables and the complex information combination process. As such, tools for explanation generation and tracking the origin of beliefs, uncertainty and conflict in evidential reasoning systems are proposed. Finally, we must consider the problem of fusion on a higher level in the Joint Directors of Laboratories (JDL) framework, where uncertain relations between multiple entities need to be tracked. An extension of the conceptual graphs framework to allow uncertainty modelling is proposed, along with appropriate information fusion strategies. These three notions are put together, and their relationship is demonstrated in two maritime domain scenarios: situation awareness of multiple vessels involved in illegal fishing and underwater infrastructure threat assessment focused on the impact of partially reliable sources.
| Date of Award | 2 Dec 2021 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Trevor P Martin (Supervisor) & Jonathan Lawry (Supervisor) |