Feature selection based on ReliefF and PCA for underwater sound classification

The performance of underwater noise classification system is highly related to the dimensions of the features and the size of the training set. However, underwater sound signals are difficult to obtain, the training sets are always in small size and the limited information are embedded in a few feature subspace. In this paper, MFCC features are extracted firstly, and then a feature selection method based on PCA and ReliefF is presented to find the most discriminating feature subset. PCA is used to project the original feature to a new feature space by maximizing the variance matrix. ReliefF method is applied to find the proper feature subset which has the maximum score. Experimental results show that our method performs well and achieves higher recognition accuracy than that of the original features in most cases.