Collaborative Fusion for Distributed Target Classification Using Evidence Theory in IOT Environment

Abstract

As an efficient strategy, collaborative fusion can promote the classification performance while decreasing data transmission energy consumption and bandwidth requirements. In practice, the appropriate reliability assessment plays an essential rule in the fusion process. In this paper, we mainly concentrate on the classification problem of distributed target in Internet of Things scenarios, and an effective collaborative fusion method in terms of internal reliability and relative reliability evaluation is proposed. The inner reliability reveals the potential classes of the target in accordance with the local hard decision made by distinct sensor. The relative reliability reflects the credibility of the soft decision represented by belief function. These two reliability measures are complementary with each other. In our proposed fusion method, the inner reliability is applied to transfer the local hard decision into rational soft decision, and the relative reliability is utilized to decrease the influence of conflicting soft decisions by making full use of the evidential discounting operation. The discounted soft decisions are related to the combination rule of Dempster–Shafer for the final target classification. Results of experiment show that compared with the traditional fusion method, this method has the better fusion performance.