Combat Identification of Synthetic Aperture Radar Images Using Contextual Features and Bayesian Belief Networks

摘要

Given the nearly infinite combination of modifications and configurations for weapon systems, no two targets are ever exactly the same. Synthetic Aperture Radar (SAR) imagery and associated High Range Resolution (HRR) profiles of the same target will have different signatures when viewed from different angles. To overcome this challenge, data from a wide range of aspect and depression angles must be used to train pattern recognition algorithms. Alternatively, features invariant to aspect and depression angle must be found. This research uses simple segmentation algorithms and multivariate analysis methods to extract contextual features from SAR imagery. These features used in conjunction with HRR features improve classification accuracy at similar or extended operating conditions. Classification accuracy improvements achieved through Bayesian Belief Networks and the direct use of the contextual features in a template matching algorithm are demonstrated using a General Dynamics Data Collection System SAR data set.