Latest developments in acoustic research suggest that using surveying methods based on artificial intelligence (AI) could improve the effectiveness of underwater monitoring. Passive acoustic monitoring (PAM) has proven to be a cost-effective approach for gathering information about the acoustic behavior of dolphins and plays a crucial role in studying their vocalizations, particularly whistles. This study investigates the efficiency of a binary convolutional neural network (CNN) in detecting dolphin whistles amidst high-density vocalizations in an aquatic environment. Specifically, this analysis intends to determine whether a properly trained CNN can recognize a single whistle even in challenging condition, including situations where multiple dolphins vocalize simultaneously, resulting in overlapping whistles that may have different shapes and durations. To this aim, experimental trials were conducted at Oltremare marine park, Riccione, Italy, where underwater recordings of seven-dolphin vocalizations were collected over 22 consecutive hours. The CNN was trained on labeled whistle spectrograms. The model, comprising three convolutional layers followed by max pooling layers and rectified linear unit (ReLU) activation functions, was evaluated using a 10-fold cross-validation approach. Confusion matrix and performance metrics indicate that the proposed approach achieves results comparable to those reported in the literature, despite the more challenging working conditions. The study supports the potential of AI models in enhancing passive acoustic monitoring techniques.

Convolutional Neural Networks for Enhancing Detection of Dolphin Whistles in a Dense Acoustic Environment / Scaradozzi, D.; De Marco, R.; Li Veli, D.; Lucchetti, A.; Screpanti, L.; Di Nardo, F.. - In: IEEE ACCESS. - ISSN 2169-3536. - 12:(2024), pp. 127141-127148. [10.1109/ACCESS.2024.3454815]

Convolutional Neural Networks for Enhancing Detection of Dolphin Whistles in a Dense Acoustic Environment

Scaradozzi D.;Screpanti L.;Di Nardo F.
2024-01-01

Abstract

Latest developments in acoustic research suggest that using surveying methods based on artificial intelligence (AI) could improve the effectiveness of underwater monitoring. Passive acoustic monitoring (PAM) has proven to be a cost-effective approach for gathering information about the acoustic behavior of dolphins and plays a crucial role in studying their vocalizations, particularly whistles. This study investigates the efficiency of a binary convolutional neural network (CNN) in detecting dolphin whistles amidst high-density vocalizations in an aquatic environment. Specifically, this analysis intends to determine whether a properly trained CNN can recognize a single whistle even in challenging condition, including situations where multiple dolphins vocalize simultaneously, resulting in overlapping whistles that may have different shapes and durations. To this aim, experimental trials were conducted at Oltremare marine park, Riccione, Italy, where underwater recordings of seven-dolphin vocalizations were collected over 22 consecutive hours. The CNN was trained on labeled whistle spectrograms. The model, comprising three convolutional layers followed by max pooling layers and rectified linear unit (ReLU) activation functions, was evaluated using a 10-fold cross-validation approach. Confusion matrix and performance metrics indicate that the proposed approach achieves results comparable to those reported in the literature, despite the more challenging working conditions. The study supports the potential of AI models in enhancing passive acoustic monitoring techniques.
2024
File in questo prodotto:
File Dimensione Formato  
Convolutional_Neural_Networks_for_Enhancing_Detection_of_Dolphin_Whistles_in_a_Dense_Acoustic_Environ.pdf

accesso aperto

Tipologia: Versione editoriale (versione pubblicata con il layout dell'editore)
Licenza d'uso: Creative commons
Dimensione 5.98 MB
Formato Adobe PDF
5.98 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11566/337253
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 0
social impact