Neural Computing and Applications, Год журнала: 2024, Номер 36(32), С. 20513 - 20526

Опубликована: Авг. 13, 2024

Abstract Passive acoustic monitoring (PAM) is an effective, non-intrusive method for studying ecosystems, but obtaining meaningful ecological information from its large number of audio files challenging. In this study, we take advantage the expected animal behavior at different times day (e.g., higher activity dawn) and develop a novel approach to use these time-based patterns. We organize PAM data into 24-hour temporal blocks formed with sound features pretrained VGGish network. These feed 1D convolutional neural network class activation mapping technique that gives interpretability outcomes. As result, diel-cycle offer more accurate robust hour-by-hour than using traditional indices as features, effectively recognizing key ecosystem

Язык: Английский