Multiscale Sample Entropy-based Analysis of Ship Radiated Noise Signal for Surface Ship Recognition

Recently many signal processing and pattern recognition schemes have been developed to process ship radiated noise signals to improve the detection and recognition accuracy of surface ships. In this paper, we propose a new target recognition scheme for surface ship recognition that the contributions concentrate on feature selection and object classification. In the recognition scheme, first multiscale sample entropy (Multi-SampEn) method is applied to extract the discriminating features from ship radiated noise signals which has good performance in analysis of discrete signal of complexity. Then, in order to alleviate the parameter selection problem and enhance the generalization performance in Multi-SampEn, the two multilinear subspace learning (MSL) methods, i.e., multilinear principal component analysis (MPCA) and uncorrelated multilinear discriminant analysis (UMLDA) are respectively adopted for feature extraction and dimensionality reduction. Finally using the extracted features as the inputs, we construct two individual support vector machines (SVM) classifiers with different penalty constants for different classes, resulting in MPCA-SVM and UMLDA-SVM for surface ship recognition. The performance of the proposed scheme is demonstrated on real data which was collected by a towed array sonar on East China Sea in 2013. Experimental results show that Multi-SampEn for the analysis of ship radiated noise signals outperforms the other methods