Geometric error correction using hierarchical/hybrid artificial neural systems

A neural-network-based intelligent system is presented. It is capable of producing a reasonable output geometry from a noisy input geometry by recognizing the shape of the input and correcting the errors generated during the input stages. A scheme is introduced of categorizing and dividing system tasks for rapid convergence of the artificial neural networks and improved system performance on the geometry identification problems. The system consists of several artificial neural networks. All neural networks of this system are trained with a learning tool, the adaptive error backpropagation algorithm