Design of strong classifier based on adaboost M2 and back propagation network

摘要

The traditional back propagation neural network (BP neural network) often needs so many training samples. Its classification accuracy is often bad when it is used to solve the problem of fewer samples and multi-classification. In order to overcome the shortcoming, an Adaptive Booting algorithm method 2 (Adaboost algorithm M2) is proposed to improve the classification accuracy and the generalization ability of the traditional BP neural network. In this method, the BP neural network is treated as a weak classifier and the way of weighted voting is used to improve the performance. At the same time, the weighting factor and the sample weight threshold are also given to improve the learning ability of the Adaboost M2 algorithm for the wrongly classified samples and avoid over learning. The simulation results show that the classification results of the proposed strong classifier are much better than the single BP neural network weak classifier and enhance the generalization ability of the BP neural network.