A novel hybrid feature extraction approach of marine vessel signal via improved empirical mode decomposition and measuring complexity

Abstract

The feature extraction of marine vessel-radiated noise (MVRN) under the complex ocean background is explored. To this end, a hybrid approach is presented based on the analysis of MVRN in subspaces of intrinsic mode functions (IMF) extracted using the improved empirical mode decomposition (IEMD) and measuring complexity. The restriction of the end-effect is an important problem when employing the EMD algorithm. In this study, first, to reduce the end-effects, an IEMD algorithm based on the correlation expansion model is proposed. Then, a comparative study of IEMD, classic EMD (CEMD), and EMD by other expansion methods are conducted on several signals. Next, IEMD, CEMD, and variational mode decomposition (VMD) algorithms are utilized to extract a group of IMFs for three types of MVRN. Later, one obtained IMF from each method that contains the most dominant information is selected. Lastly, two statistical complexity measures (i.e., permutation entropy (PE) and slope entropy (SlopEn)) are used as the features of the chosen IMF to improve the underwater signal separability and stability. Experimental results indicate that the suggested approach (IEMD-PE/SlopEn) can effectively extract the feature information of underwater signals. Additionally, it has a better ability to discriminate between various kinds of MVRN.