An Approximate Optimal Chernoff Fusion Method Via Importance Sampling

This paper focuses on addressing the decentralized data fusion (DDF) problem in dynamic sensor networks based on Chernoff rule. Generally, the Chernoff rule is challenging to implement since the fused probability density functions (pdfs) that cannot be obtained in closed form. Besides, the existing works for implementing Chernoff rule are mostly confined to iterative fusion of two sensors. To address these issues, a novel importance sampling (IS) based Chernoff fusion method is proposed. In particular, by considering the multi-sensor cases, the two sensor Chernoff fusion is generalized to a multi-sensor Chernoff fusion, and the accompanying high-order optimization problem for calculating fusion exponent is addressed by particle swarm optimization (PSO) method. Additionally, to ensure accurate approximation of the Chernoff fusion pdf, an IS based procedure is incorporated, wherein the Chernoff fusion is no longer achieved by fusing (Gaussian or Gaussian mixture) parameters of the local sensors but particle samples that obtained from IS. Numerical results show the efficiency of our method.