Water entry sound detection in strong noise by using the spectrogram matrix decomposition method. (April 2020)

• Record Type: Journal Article
• Authors: Huang, Shichun; Yu, Liang; Jiang, Weikang
• Is Part Of: Applied acoustics. Volume 161(2020)
• Publication Date: 2020-04
• Subjects: Water entry sound -- Transient sound detection -- Noise reduction -- Robust Principal Component Analysis -- Non-Negative Matrix Factorization; Acoustical engineering -- Periodicals; Periodicals; 620.2
• DOI: 10.1016/j.apacoust.2019.107171
• Languages: English
• ISSNs: 0003-682X

Abstract

Abstract: The water entry transient sound of a projectile can be used for early target detection in passive sonar. Two detectors for the transient sound with low Signal to Noise Ratios (SNRs) have been proposed depending on the spectrum matrix decomposition. One detector is the Robust Principal Component Analysis (RPCA)-based detector, and the other one is the Non-negative Matrix Factorization (NMF)-based detector. In the RPCA-based detector, the measured spectrogram is divided into a low-rank matrix and a sparse matrix, and the sparse matrix representing the transient sound is used to construct the detection function. In the NMF-based detector, the measured spectrogram is decomposed into a series of rank-1 matrices, and the temporal patterns with large kurtoses are used to construct the detection function. The detectors are data-driven approaches based on a single channel, and no prior information is needed. The performances of the proposed algorithms are determined, and the experimental results demonstrated that the proposed algorithms have good performance with more than 2 dB relative detection gain.