Scholarlink

<p>Animal identification is pivotal for ecological studies, yet automated recognition tools for bee species remain underexplored. Here, we present a machine learning approach using a Random Forest algorithm to identify five bee species representing three phylogenetically diverse families within Apoidea based on their flight and floral buzz sounds. Acoustic parameters were extracted from recordings, with the fundamental frequency emerging as the most relevant feature for species classification. Machine learning models achieved 90.94% using flight buzz and 82.22% with floral buzz. Combining both sound types increased accuracy to 95.04%. Among all bee species, <i>B. pauloensis</i> showed the lowest classification performance, likely due to intraspecific variation in body size, leading to acoustic overlap with other species. Despite this, the proposed method demonstrates high performance and suggests that acoustic features can be reliably used for species-level identification. This approach holds potential for non-invasive monitoring of bee richness and abundance in diverse communities, contributing to the development of automated tools for ecological research and biodiversity assessment.</p>

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

Flight and Floral Acoustic Signals for Bee Species Identification

  • 期刊论文

  • 发表日期：

    2025年10月8日

查看全文
  • César Augusto Arvelos,
  • Caique Rocha Resende,
  • João Pedro Santos Pereira,
  • Lucas Costa Brito,
  • Marcus Antonio Viana Duarte,
  • Vinícius Lourenço Garcia de Brito

摘要

Animal identification is pivotal for ecological studies, yet automated recognition tools for bee species remain underexplored. Here, we present a machine learning approach using a Random Forest algorithm to identify five bee species representing three phylogenetically diverse families within Apoidea based on their flight and floral buzz sounds. Acoustic parameters were extracted from recordings, with the fundamental frequency emerging as the most relevant feature for species classification. Machine learning models achieved 90.94% using flight buzz and 82.22% with floral buzz. Combining both sound types increased accuracy to 95.04%. Among all bee species, B. pauloensis showed the lowest classification performance, likely due to intraspecific variation in body size, leading to acoustic overlap with other species. Despite this, the proposed method demonstrates high performance and suggests that acoustic features can be reliably used for species-level identification. This approach holds potential for non-invasive monitoring of bee richness and abundance in diverse communities, contributing to the development of automated tools for ecological research and biodiversity assessment.