A Solution to Dimensionality Curse of BP Network in Pattern Recognition Based on RS Theory

In order to solve the dimensionality curse of BP neural network in pattern recognition, this paper proposes a model of dimensionality reduction which based on rough set theory. While training network, the model first carries out attribute reduction based on rough set theory, and then picks up important characteristics of ideal samples to reduce input space dimensions. Hence the speed of network training is increased. During pattern recognition process, the model picks up important characteristics of practical samples and denoise, so the recognition rate is increased. For illustration, a letter recognition example is used to test the feasibility of this model. The Results of experiment show that the model can effectively solve the dimensionality curse of BP network in pattern recognition.