Adaptive classification system of ship-radiated noise based on hybrid multi-algorithm

• Published in: Ocean engineering Vol. 310; p. 118633
• Main Authors: Yang, Hong, Wang, Chao, Li, Guohui
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.10.2024
• Subjects: Classification; Decomposition; Entropy; Feature extraction; Maximum information coefficient; Ship-radiated noise; Support vector machine; Maximum information coefficient; Ship-radiated noise; Classification; Feature extraction; Decomposition; Entropy; Support vector machine
• ISSN: 0029-8018
• DOI: 10.1016/j.oceaneng.2024.118633

As the main source of ship features, ship-radiated noise plays a key role in recognizing different types of ships. Therefore, to effectively extract the features from ship-radiated noise for accurate recognition, a novel adaptive classification system of ship-radiated noise based on modified singular spectrum decomposition using improved Cao algorithm and reverse amplitude aware permutation entropy (RICSSD), weighted fractional-order ensemble fluctuation dispersion entropy (WEFDEα) and improved support vector machine using snake optimization algorithm (SO-SVM) is proposed. Aiming at the shortcomings of subjective selection of embedding dimension and poor anti-noise performance in singular spectrum decomposition, RICSSD is proposed. Aiming at the disadvantages of fluctuation dispersion entropy, such as easy loss of statistical data and poor anti-interference ability, the concepts of weighting, new fractional-order, and ensemble are innovatively introduced into fluctuation dispersion entropy, and WEFDEα is proposed. To obtain the most suitable kernel function parameter and penalty coefficient of SVM, SO-SVM is proposed. Preprocessing: The ship-radiated noise is decomposed into a series of singular spectrum components (SSC) by RICSSD, and a new maximum information coefficient (ChiMIC) is calculated to measure the information that the SSC contains the original ship-radiated noise. Feature extraction: The SSC with the max ChiMIC is selected as the feature vector, and 50 samples are randomly selected from the feature vector to calculate their WEFDEα features. Classification: Each sample feature is input into SO-SVM, and the final recognition rate is obtained. The experimental results show that the classification accuracy of the system in ShipsEar and Deepship datasets reaches 97.5% and 98.75%, respectively. Therefore, it provides a new perspective for the classification of ship-radiated noise. •Modified SSD using ICao and RAAPE is proposed, which enhances the decomposition performance of SSD.•WEFDEα is proposed, which enhances the ability of FDE to characterize time series.•SO-SVM is proposed, which solves the difficulty of parameter selection of SVM.•An adaptive classification system of ship-radiated noise is proposed. Its effectiveness is verified.