Ballard, James Philip; (2002) Radar range profile ship signature classification. Doctoral thesis (Ph.D), UCL (University College London).

Text Radar_range_profile_ship_signa.pdf Download (11MB)

Abstract

Radar range profiles can be used to classify ships at long range, in all weathers, day and night. Pattern recognition techniques have been used to automatically classify ship range profiles with varying degrees of success. An emerging theme in this area is the issues associated with realising the performance that is achievable in a benign lab environment when using an automatic classifier in an operational environment. A number of these issues are selected and techniques to combat them are developed and tested on measured data. A classifier trained on range profiles recorded under benign high signal-to-noise ratio conditions will drop rapidly in performance when forced to classify range profiles collected under more demanding low signal-to-noise ratio conditions. It is suggested and then shown that a higher probability of correct classification can be achieved by temporarily adjusting the parameters of the database to match the lower signal-to-noise ratio signal rather than by adjusting the signal to match the database. A technique to make the maximum likelihood classifier robust to changes in the configuration of the target is proposed and investigated. The techniques of combining multiple looks at a target and using a confidence threshold are applied to the output of three classifiers and then combined to suggest that in multiple target scenarios it may be better to repeatedly scan over all targets rather than staring at each one in turn. Finally, the potential to ease the data collection task by using range profiles at one aspect angle to classify profiles from a different aspect angle is assessed.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item