Research Detection and classification of fish sounds
Abstract
Fish sound detection and classification is a hot topic in the ocean acoustics community. The state of the art in automatic classification of fish sounds is very limited and need to be addressed. Analyzing fish sounds is a relevant choice for monitoring the biodiversity and vitality of a given coastal area. Here, we propose deep learning methods to detect and classify fish sounds recorded in the East coast of the United States. (work in progress).
Between 1950 and 1970 Marie Poland Fish and William H. Mowbray recorded sounds made by marine fish. Their collected studies were documented in the printed volume Sounds of Western North Atlantic Fishes: A Reference file of Biological Underwater Sounds published by the Johns Hopkins Press in 1970. Fish and Mowbray made recordings of sounds produced by 153 species of the 220 species of fish they studied. The collection of sounds corresponds directly to the printed sonograms and descriptive materials for each species found in the printed volume.
This catalog with 153 spices recorded was published by John Hopkins Press this year. Each track is voice-labeled at the beginning and has the same information as in the PDF. Fish sounds collection by Graduate School of Oceanograph, University of Rhode Island. A spectrogram sample is shown below.
An experiment conducted by the University of Delaware in Delaware Bay in 2013 yielded unlabeled data recorded with a sample rate of 192000 and two channels with 14 days duration. The following spectrogram presents recorded data from July 25, 2013 using channel 1.
In addition to the audio files collected during the experiment. Videos of fish in the area were recorded underwater in Delaware Bay. The video below shows some of the fish recorded in the experiment.
Related publications
-
Work in progress
