Scholarlink

Biodiversity plays an important role in all ecosystems. Systems with a higher bio-diversity are more resilient to changes in the environment and generally represent a higher level of ecological health. However, monitoring biodiversity is not always manageable, especially without disturbing the environment. Collecting samples, tagging animals can generally be disruptive and negatively impact an ecosystem. This work explores the potential of utilizing audio analysis for estimating the biodiversity in a non-invasive manner. By combining Mel frequency cepstral coefficient (MFCC) and convolutional neural networks (CNN) a unified approach is proposed that ca be used to identify avian species based on the sounds of their calls in recordings. To ensure favorable model performance, hyperparameter tuning is carried out using a modified version of the recently proposed chimp optimization algorithm (ChOA). The best performing optimized models showcase favorable outcomes evaluate on a real world dataset attaining an accuracy of 0.838710 suggesting viability for real world use.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Classifying Birds of South America via Audio Analysis with Convolutional Networks Optimized by Metaheuristics

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ninoslava Jankovic,
  • Branislav Radomirovic,
  • Luka Jovanovic,
  • Stefan Ivanovic,
  • Smiljana Tedic,
  • Miodrag Zivkovic,
  • Nebojsa Bacanin

摘要

Biodiversity plays an important role in all ecosystems. Systems with a higher bio-diversity are more resilient to changes in the environment and generally represent a higher level of ecological health. However, monitoring biodiversity is not always manageable, especially without disturbing the environment. Collecting samples, tagging animals can generally be disruptive and negatively impact an ecosystem. This work explores the potential of utilizing audio analysis for estimating the biodiversity in a non-invasive manner. By combining Mel frequency cepstral coefficient (MFCC) and convolutional neural networks (CNN) a unified approach is proposed that ca be used to identify avian species based on the sounds of their calls in recordings. To ensure favorable model performance, hyperparameter tuning is carried out using a modified version of the recently proposed chimp optimization algorithm (ChOA). The best performing optimized models showcase favorable outcomes evaluate on a real world dataset attaining an accuracy of 0.838710 suggesting viability for real world use.