Detection of Sound Field Aberrations Caused by Forward Scattering from Underwater Intruders Using Unsupervised Machine Learning

Abstract

Forward scattered waves are produced by underwater intruders that cross a source-receiver line. Strong direct blasts lead to a difficult detection of sound field aberrations caused by forward scattered waves. An unsupervised detection scheme that processes repeatedly transmitted pulses on a receiver array is proposed. For detection under strong blasts, the scheme performs unsupervised learning on spectra of normalized envelopes on an array output, which has the advantage of robustness for weak field aberrations and real-time detection after effective training. An experiment was carried out on the lake; the results show that the method has yielded reliable results in comparison with approximately 1-dB aberrations on the received pulse strengths caused by forward scattering from an intruder. Furthermore, the relationship between strength aberrations caused by forward scattering and the location of the intruder through the baseline is discussed further, and the capabilities of the scheme are further discussed with noise-added experimental data.