Online variety recognition of auto rack girders based on combination of Fuzzy ART neural network with D-S evidence theory

Abstract

To address the difficulty of artificial recognition of hundreds of auto rack girders, this paper introduces an online automatic inspection method which synthesises machine vision, wavelet transformation theory, Fuzzy ART neural networks and D-S evidence theory on auto rack girders. First, local entropy, NMI and energy value of wavelet coefficients are used as input layers of a Fuzzy ART neural network, to gain the basic confidences of these three different characters. Next, D-S evidence theory is used to fuse the three basic confidences. Finally, total confidence in auto rack girder images, is obtained to determine a model for the inspected auto rack girders. This project of variety recognition for auto rack girders using D-S evidence theory and the Fuzzy ART neural network provides a new technology for use at home or overseas, which resolves the question of the lower recognition rate for a single character template and advances a method for multi-character fusion.