Improving maritime domain awareness using neural networks for target of interest classification

• Main Author: Schaus, Brian M.
• Other Authors: Scrofani, James
• Published: Monterey, California: Naval Postgraduate School 2015
• Online Access: http://hdl.handle.net/10945/45252

Approved for public release; distribution is unlimited === Techniques for classifying maritime domain targets-of-interest within images are explored in this thesis. Geometric and photometric features within each image are extracted from processed images and are used to train a neural network. The trained neural network is tested with features of a known object. In the binary classification case, the neural network is used to determine whether a ship is present or not present in the image. In the multi-class and multi-level classification cases, the neural network is used to determine if the object belongs to one of four classes specified: warship, cargo ship, small boat, or other. The Hough transformation is used to identify and characterize linear patterns exhibited by objects in images. As an alternative to geometric and photometric features to classify targets-of-interest, these linear patterns are used to train a neural network. The performance of the neural network is then tested for binary, multi-class, and multi-level classification schemes. The development of neural-network-based techniques for automated target-of-interest classification is a significant result of this thesis.