Use of Non-Ideal Training Data in SAR ATR for Targeting

This paper investigates a template-matching approach to automatic target recognition (ATR) for SAR for the case in which positive identification of a single known target type is required. ATR schemes are necessarily supported by databases of training imagery and of particular interest in this study was the impact on performance of using a database of simulated imagery to match against real in situ SAR imagery. Assessments have been made using 30cm resolution, X-band, spotlight imagery to provide the in situ test data and training databases firstly based on different examples from the same in situ data to provide an idealised baseline and secondly using simulated imagery. The impact on performance for both single channel and polarimetric data is reported and discussed.