Contextual information in SAR target detection

Abstract

Established target detection schemes make a statistical decision as to whether the value of a test pixel is more likely to have arisen due to the presence of a target or of clutter, subject to prior probabilities for the occurrence of targets or clutter. However, such schemes do not take into account the contextual information which is available. For example, it will be known that military vehicles may be more likely to be located in fields close to hedges and woodland edges to provide cover and will often travel in groups. It is demonstrated how such contextual information may be incorporated into a SAR target detection scheme, while maintaining the statistical rigour of established techniques. A theoretical framework is developed in which the prior probability for the presence of a target is modified by the contextual influences via conditional dependencies. The resulting detection scheme is applied to simulated SAR images to obtain target detection rates for various combinations of contextual information. An improvement in performance of about 13% is observed when the context is fully exploited.