Evaluating sensor reliability in classification problems based on evidence theory.

This paper presents a new framework for sensor reliability evaluation in classification problems based on evidence theory (or the Dempster-Shafer theory of belief functions). The evaluation is treated as a two-stage training process. First, the authors assess the static reliability from a training set by comparing the sensor classification readings with the actual values of data, which are both represented by belief functions. Information content contained in the actual values of each target is extracted to determine its influence on the evaluation. Next, considering the ability of the sensor to understand a dynamic working environment, the dynamic reliability is evaluated by measuring the degree of consensus among a group of sensors. Finally, the authors discuss why and how to combine these two kinds of reliabilities. A significant improvement using the authors' method is observed in numerical simulations as compared with the recently proposed method.