Rapid detection of fish calls within diverse coral reef soundscapes using a convolutional neural network

The Journal of the Acoustical Society of America, Journal Year: 2025, Volume and Issue: 157(3), P. 1665 - 1683

Published: March 1, 2025

The quantity of passive acoustic data collected in marine environments is rapidly expanding; however, the software developments required to meaningfully process large volumes soundscape have lagged behind. A significant bottleneck analysis biological patterns datasets human effort identify and annotate individual events, such as diverse abundant fish sounds. This paper addresses this problem by training a YOLOv5 convolutional neural network (CNN) automate detection tonal pulsed calls spectrogram from five tropical coral reefs U.S. Virgin Islands, building over 22 h annotated with 55 015 calls. identified mean average precision up 0.633, while processing 25× faster than it recorded. We compare CNN annotators on datasets, including three used for two untrained reefs. CNN-detected call rates reflected baseline reef cover observations; both expected (e.g., crepuscular choruses) novel were identified. Given importance reef-fish communities, their bioacoustic patterns, impending biodiversity crisis, these results provide vital scalable means assess community health.

Language: Английский

Soundscape and fish passive acoustic monitoring around a North Sea gas-production platform in the Dogger Bank

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(4), P. e0319536 - e0319536

Published: April 2, 2025

This study provides temporo-spatial characterisation of the underwater soundscape in proximity a relatively newly installed offshore gas-production platform North Sea’s Dogger Bank Special Area Conservation, recorded by Static Acoustic Monitoring at different distances from wellhead (70 m, 5 Km and 10 km). Long-Term Spectrogram Analysis percentile Power Spectral Densities demonstrated strong acoustic similarity between sites; no biophonic acoustic-mass phenomena were present. All locations characterized Underwater Radiated Noise, concentrated < 2 kHz, which dominated soundscape. Fish community analysis was performed to explore occurrence, richness, abundance, diel, seasonal patterns putative fish sounds. Principal Component used infer potential sound-emitting species, on Sea sounds downloaded Global Inventory known ( https://fishsounds.net/ ), analyzed for same features characterize during this study. The low levels diversity (acoustic richness ranging 1 2) abundance (never above min -1 ). sound type ‘Pulse Series’ (PS), emitted 70 m km station September ca. 19:00 23:00, with closest linear combination those typifying Eutrigla gurnardus . ‘Low-frequency Down-Sweep’ (LF-DS) all stations grunts Gadus morhua represents first application context environmental management an operational gas production platform.

Language: Английский