A Xerion-based Perl program to train a neural network for grid pattern recognition

Abstract

Neural network training with the back propagation algorithm is an important artificial intelligence technique for grid pattern recognition. The training is time-consuming however and generally requires a trial-and-error procedure to configure the network. A Perl program executed with Xerion is presented to relieve the;training burden. Statistical reports such as computation time, learning performance, and validation performance are generated automatically by the program. A case study applying the program for training networks to determine a drainage pattern from Digital Elevation Model data is demonstrated and discussed. Manually determining drainage patterns from topographical maps for a grid-based model is tedious and subjective. The neural network has a self-learning capability that can replace human judgment involved in the conventional approach. Copyright (C) 1996 Elsevier Science Ltd.