Classification results of ELM+RBM (solid lines) and PCA+ELM (dot-dashed lines) on data sets with different S/N. The results show that RBM+ELM performs better than PCA+ELM on data sets with S/N> 20 and 10 <S/N ≤ 20. For the data set with 0 <S/N ≤ 10, RBM+ELM performs worse than PCA+ELM when using less than 600 components and performs better than PCA+ELM when using more than 600 components.