Improved algorithm of RBF neural networks and its application

Abstract

In order to improve the predictive accuracy of RBF neural network in function approximation, an improved RBF neural network was proposed. In this new model, human experience was added to the last layer as the activation function. The model of improved algorithm was built in Simulink, and was used to approximate a 2-dimensional function. The simulation result showed that the improved network performed well in function approximation. At last, a neural network system which was based on the improved algorithm was used in license plate recognition. In this system, the first two layers of the network were implemented in hardware, and the last layer was achieved in software. Experimental results show that the predictive accuracy of network is improved after joining human experience to the output layer.