Abstract: The basal model of directional basis probabilistic neural network (DBPNN) and the corresponding algorithm and theory applied in pattern recognition are investigated aiming at utilizing the advantage and overcoming the shortcomings of directional basis function neural network (DBFNN) and probabilistic neural network (PNN). Its application to pattern recognition is from the results obtained in classification of cracks and porosity in weld defect. It can be seen that DBPNN has greater improvement in computation speed and classification, compared with the DBFNN and PNN.

Key words: Directional basis function neural network, Directional basis probabilistic neural network, Fusion, Pattern recognition, Probabilistic neural network