Solution to Noise Problem of BP Network for Pattern Recognition Based on RS Theory
Qin Hai-ou · Jiangnan daxue xuebao. Ziran kexue ban · 2010
In order to solve the noise problem of BP network for pattern recognition,the paper proposes a method to process the noisy samples based on the upper approximations,the lower approximations and the boundary region theories of rough sets.The method eliminates the noise of attribute values and changes them into ideal values when they are in the lower approximations;and those attribute values with noise will remain unchanged while they are in the boundary region.The sample will be refused to recognize if the percent of its attributes with their values in the boundary region is over a certain point.The results of experiment show that the method can effectively reduce the false recognition rate of BP network for pattern recognition.