Research of vision recognition on auto rack girders based on improved ART2 neural network and D-S evidence theory

For typepsilas recognition of camion rack girders, this paper puts forward a pattern recognition method based on improved ART2 neural network and D-S evidence theory. Firstly, for collected auto rack girders top images, region is partitioned to 16 equal regions ( 4times4 ) , which covers high-frequency wavelet coefficient with one-layer wavelet transform, and gained local variance of wavelet coefficient in every sub-region is used as a character template; in the same way, is partitioned to 16 sub-images (4times4), estimating numbers of gray value ldquo1rdquo in every sub-region, to gain the area character template. Secondly, data of two character templates are used as inputs of improved ART2 neural network, to gain data of joint weights of network, and the basal confidence m₂ , m₂ . Finally, gain the total confidence with composition rule of D-S evidence theory, according to the maximum of the total confidence, to recognize types of auto rack girders. Experiments indicate this algorithm may solve on-line vision recognition on hundreds of auto rack girders very well, and possesses advantage of more rapid , more precise and more reliable etc. Typepsilas recognition of camion rack girders has a more broad application future and higher practicality, based on neural network and D-S evidence theory.