Feature Selection for Bayesian Inference Network in Radar Jamming Effect Analysis

Bayesian inference network (BIN) is one of the most popular methods for radar jamming effect analysis (RJEA) and its success stems from its intelligibility and direction-free knowledge inference. However, its training efficiency is always limited by its size. When numerous features are exploited for BIN in RJEA, this issue will become more serious. To tackle this hurdle, we put forward a feature selection method in the premise of keeping or even improving its inference precision. To do so, we develop manifold learning to implement feature selection, where the intrinsic subspace of original features is first estimated by the modified locally linear embedding algorithm and then the features are evaluated according to their contributions to preserving original spatial structure. The results in numerical experiments can illustrate the feasibility of the proposed method.