Deep Learning for Marine Bioacoustics and Fish Classification Using Underwater Sounds
Jean-François Laplante, Moulay A. Akhloufi, Cédric Gervaise
Abstract
The migration of species is an important factor in the analysis of ecological systems. Changes in migratory patterns of a species or a specific group in an ecosystem often follow changes in the environment - many animals are sensitive to small changes that may not initially be thought of as significant for the health of an ecosystem. The presence of many species can be detected by the sounds they produce, and as such, environmental conservation efforts have much to gain from the automation of the analysis of Bioacoustics. Deep Learning shows promise for this type of task. This work evaluates the performance of different deep learning methods when performing the task of detecting the presence of brown meagre sounds in spectrograms with different window lengths and achieves an F1-score of 0.94 for brown meagre vocalization detection.