A hybrid learning algorithm for pattern recognition

This paper propose a hybrid neuron model which construct various networks and are parallel to learn in each network. For pattern clustering, LVQ algorithm by Kohonen(1990) is used and for parallel learning, Adaptive Linear Neuron(ADALINE) by Widrow and Hoff(1990). The algorithm we propose consists of two parts. The first step is dividing N patterns into c clusters with LVQ clustering algorithm. The second step is recognizing c patterns with each pattern selected from each cluster using ADALINE algorithm. 250 patterns of ASCII characters normalized into 8 X 16 and 1124 data sampled from 9 vehicle types that passed Inductive Loop Detector(ILD) at 10 pointers were normalized as the data for this research. In the case of ASCII character recognition, 191(179) out of 250 patterns are recognized with 3%(5%) noise and 807 types are recognized showing 71.8% recognition ratio with 1124 vehicle types data. The result shows 10.2% improvement over the result obtained with back-propagation algorithm.