A modular neural network for enhancement of cross-polar radar targets

-A polarimetric radar navigation system makes use o f polarization rotating twist-grid retroreflectors in order to navigate a confined waterway, even in inclement weather or after dark. Despite the polarization diversity offered by such a radar target, depolarization allows significant cross-polar clutter to obscure the reflector return. A novel modular neural network solution integrates an adaptive crass-polar interference canceller, a radial basis function network, and a conventional cell-averaging CFAR processor to successfully demonstrate the enhancement and detection o f a polarization target. The modular solution outperforms any one o f the aforementioned methods on their own. This is indicated subjectively through the display o f the resultant processed images, and objectively by the estimates o f target-to-clutter ratio and receiver operating curves. A post-detection processor uses a priori information about the reflector location along the water-land boundary o f the waterway. A fuzzy processor combines primary detection information with the output from a vision-based edge detector to effectively remove false alarms. Keywords---Radar, Detection, Modular, Neural network, Navigation, Polarization, Image processing, Vision. 1. I N T R O D U C T I O N 1.1. Precise Navigation Problem Inland confined waterways, such as St. Lawrence seaway and the Mississippi River, are of great importance in the transportation of goods by ship. The shipping community strongly desires to make best use of this resource. However, some factors exist that limit the time that the waterways may be used. During periods of low visibility, namely fog, heavy rain, or darkness, the ships cannot navigate by visual aids. The buoys which are used as visual aids in navigation are deployed at the beginning of the shipping season, and removed at the end of the season. To extend the season beyond the times when buoys are available, and to travel in times of poor visibility, another navigational aid is needed to Acknowledgements: Financial support provided by the Natural Sciences and Engineering Research Council (NSERC), Ottawa, for doing the work discussed herein is deeply appreciated. The authors are also grateful to an anonymous reviewer for many useful comments in an earlier version of the paper. Requests for reprints should be sent to A. M. Ukrainec, Communications Research Laboratory, McMaster University, Hamilton, Ontario, Canada; E-mail; [email protected] supplant, or replace, that of visual nagivation by the ship's pilot. It goes without saying that such a system must be robust and provide an accuracy of navigation comparable to that of the ship's pilot. It was judged that a ship's pilot could navigate visually within an accuracy of approximately 4-3 m from the channel centerline and 4-7 m from the channel limits. The use of a marine radar system was investigated as a means by which radar ranging to known targets could be used for triangulation to locate a ship within a waterway. The discriminants that are available to the radar systems designer for the identification of stationary targets are power, frequency, and polarization. In the context of noncoherent radar, polarimetric diversity is the only viable option. The polarization of an electromagnetic wave is defined as the direction of the electric field component. Most marine radars transmit with the electric field linearly polarized in the horizontal plane. A passive reflector target which is able to rotate the plane of polarization efficently is obviously needed. It is known that a dihedral reflector mounted on a 45 ° angle from the horizontal has the desirable property of rotating a linearly polarized field through 90 ° efficiently. Early studies showed that the dihedral exhibited the polarimetric characteristics that could be exploited in navigation. Unfortunately, the dihedral only