Gardado en:
| Autor Principal: | |
|---|---|
| Formato: | Recurso digital |
| Idioma: | |
| Publicado: |
Zenodo
2026
|
| Acceso en liña: | https://doi.org/10.5281/zenodo.18910541 |
| Tags: |
Engadir etiqueta
Sen Etiquetas, Sexa o primeiro en etiquetar este rexistro!
|
Table of Contents:
- <p><span lang="EN-GB">Whale and dolphin classification contributes to protecting the eco-friendly system, biodiversity conservation, and marine sustainability. This paper presents a new machine learning model based on the extraction of seven sound features and a random forest algorithm for automatically detecting and classifying the sounds of killer whales, long-finned pilot whales, and harp seals, whose living areas widely overlap. These features are MFCCs, delta features, Spectral features (spectral_centroid, spectral_bandwidth, spectral_contrast), Chroma features (chroma_stft, chroma_cqt, chroma_cens, Zero-crossing rate, RMS Energy, and Tonnetz (for harmonic signals). A new machine learning classifier is designed to classify between 49 types of whales and dolphins. The dataset includes 2000 record files, split into an 80% training dataset and a 20% validation dataset, balancing the dataset. Random Forest is considered a significant model for sound classification, achieving robustness and interpretability with seven sound features. The experimental result is applied to the Watkins Marine Mammal Sound Database, achieving a sound classification rate of 92.97%.</span></p>