Automatic target recognition with Chebychev networks
Akira Namatame, N. Ueda · 1991
Summary form only given. Presents a connectionist automatic target recognition (ATR) system with a novel network architecture, Chebychev neural networks. The architecture of the connectionist ATR consists of a Chebychev network module, a conjunctive network module, and a classification network module. In the Chebychev network module, nonmonotonic Chebychev activation functions are used for the input units. A 2D silhouette is characterized by its complex shape. Two types of training examples-positive examples that represent the positions within the silhouette and negative examples that represent the positions outside of the silhouette-were used for training the Chebychev networks. After training, the Chebychev networks automatically extract features such as the area and complex shape of the silhouette. The recognition performance and the size of the networks of a connectionist ATR with Chebychev networks do not depend on the complexity of the target silhouette.>