Abstract: This article introduces a welding seam detecting algorithm based on the ART (Adaptive Resonance Theory) artificial neural network. The basic principle of the algorithm is to sum up the grey level distributions on the transversal direction of the seam into several typical space patterns and then match them with the patterns memorized in the LTM (Long Time Memory) of the ART neural network. The seam position can be determined according to the pattern distributions. It is a novel method for welding seam visual tracking which can inhibit strong noises and work correctly in a very noisy circums-umstance.

Key words: Artificial neural network, Pattern recognition, Welding seam tracking