A noise classification algorithm based on SAMME and BP neural network

Noise classification is a global extreme value solution problem for complex nonlinear functions. Based on SAMME and BP Neural Network, this paper proposes a noise classification algorithm-SAMME-NN. In SAMME-NN, multiple BP-NN weak classifiers are combined into a strong classifier. The weight of each sample is changed adaptively according to the classification results and the sampling method with unequal probability is used according to the weights of training set. A classification influence factor to each weak classifier is added to reduce the impact of BP-NN with high classification error rate on the overall result and the probability of falling into the local optimal solution. An experimental data set with 100,000 representative noise data is constructed in this paper, including birds, crowd, rain and cars. The results show that the classification accuracy of SAMME-NN algorithm is 98.62%, which is better than GMM and BP. Furthermore, experiments show that the convergence speed of SAMME-NN is fast and stable.