Image reconstruction and target identification based on neural network models.

摘要

In this dissertation, two approaches based on neural network models for target identification, with either high resolution images formed or label representations generated from partial information of targets, are developed and described. For those applications where enough information about a target can be acquired from different aspects of the target and forming an image of the target is possible, the neuromorphic processor described in this dissertation is able to reconstruct much higher resolution images than traditional approaches through adaptive processing. For applications where information about a target can not be acquired for a wide aspect range needed for forming an image, a learning neural net is developed which is able to perform real-time, robust identification. Both identification approaches are tested using realistic experimental microwave data. (Abstract shortened with permission of author.)