Dual feature extraction system for ship-radiated noise and its application extension

Abstract

The complexity of maritime noise and the time-variability of hydroacoustic channels make it more and more difficult to extract feature from ship-radiated noise effectively. To accurately recognize ship-radiated noise, a dual feature extraction and recognition system based on pre-processing-dual feature extraction-recognition is proposed. Pre-processing: The primary decomposition and secondary decomposition are performed for the first time using variational mode decomposition optimized by giant trevally optimizer, in which the threshold is adaptively determined. Dual feature extraction: Grey relational analysis is calculated for the intrinsic mode function component after primary decomposition and the high-intrinsic mode function component after secondary decomposition, and two eigenvectors are selected separately. Then entropy of the combination of weighted fluctuation-based dispersion entropy and Lempel-Ziv complexity is proposed to calculate the eigenvalues for classification. Recognition: The feature values are recognized using random forest recognizer optimized by tuna swarm optimization to obtain the final recognition rate. The experimental results show that the proposed entropy has an overwhelming advantage over the others, such as permutation entropy, and dispersion entropy. It is proved that the proposed system for ship-radiated noise has good classification performance and the recognition rate is up to 97.5%. In addition, it is successfully applied to simulated chaotic signals as well as measured aquatic life signals.