Research of vision recognition of auto rack girders based on adaptive neural network and D-S evidence theory

This paper puts forward an on-line automatic inspecting means, which covers ART2 neural network, wavelet transform and D-S evidence theory for types of auto rack girders. Firstly, for the real-time gathered auto rack girders top image on line, extracting wavelet decomposed coefficient of image with wavelet transform, energy value of wavelet coefficient is used as a character template; top images of auto rack girders are partitioned to 16 sub-images (4times4), which are processed with SUSAN edge inspecting algorithm, estimating numbers of edge pixels of every region respectively, which is used as a character template; the primary image is partitioned to 16 sub-images(4times4) in the same way, calculate barycenter character (distance between barycenter position and center) of every sub-region respectively, which is used as a character template. Secondly, in order to gain basal reliability of auto rack girders image, three character templates data which are energy value of wavelet coefficient, numbers of edge pixels and barycenter character are used as inputs of ART2 neural network. Finally, according to composition rule of D-S evidence theory, to gain total reliability and recognize types of auto rack girders. Before recognition, to gain standard character database study three character templates of various auto rack girders with ART2. a mass of experiments indicate on-line maximal recognition rate meets demands of production, which based on combination ART2, wavelet transform and D-S evidence theory to recognize kinds of auto rack girder, and possessed advantage of more rapid and more precise recognition etc.