Neyman-Pearson detection of underwater bioacoustic signals - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2016

Neyman-Pearson detection of underwater bioacoustic signals

Résumé

The use of passive acoustic for the classification, localization, and density estimation of populations of marine mammals is a current area of interest. It is a cheap and an efficient alternative to visual surveys. However, the lack of an efficient automatic detector for unknown marine mammal calls greatly undermines the feasibility of those tasks, especially when dealing with species showing a great variability of calls. This study adds one more step toward the fully automatic detection of unknown bioacoustic signals in impulsive, non-stationary, and colored ocean noise. The detection procedure is a two-steps fully statistical method solely based on the knowledge of the background noise in the spectrogram. The first step models the noise power as a chi-squared distribution, which parameter is estimated. The signal is then detected using a Neyman-Pearson approach, providing a binary spectrogram that contains false and true detections. The second step removes a significant amount of false detections from the binary spectrogram. The time-frequency distribution of false detections is fitted with a correlated binomial distribution, which is used to discriminate patches of detections (signal) from uniformly distributed detections (false alarms). Examples showing the applicability of this method on several real underwater sounds are presented.
Fichier non déposé

Dates et versions

hal-01343782 , version 1 (09-07-2016)

Identifiants

  • HAL Id : hal-01343782 , version 1

Citer

Florian Dadouchi, Julien Huillery, Cedric Gervaise, Jerome I. Mars. Neyman-Pearson detection of underwater bioacoustic signals. Acoustics 2016 - 171st Meeting of The Acoustical Society of America, May 2016, Salt Lake City, United States. pp.2063. ⟨hal-01343782⟩
140 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More