Maritime radar target detection using neural networks

Two neural network based maritime radar target detection systems are presented: one is based on a backpropagation (BP) network and the other on the combination of a principal component analysis (PCA) network and a BP network. Using maritime radar images as inputs, both systems include a pre-processing stage, a BP network classification stage and a post-processing stage. In the PCA+BP network based system, an extra PCA network is used for compacting test patterns. Experimental results show that: (a) for the BP network based system, the overall detection rate is 95.3%; (b) for the PCA+BP network based system, the same overall detection rate can be reached with a reduced number of false-alarms; (c) as the number of components of compacted data from the PCA network is reduced, the overall detection rate is maintained at nearly the same level; and (d) compared with a conventional mean-level constant false-alarm rate detector, the two neural network based systems can provide higher overall detection rates while maintaining a lower number of false-alarms