Scholarlink

<p><UnorderedList Mark="Bullet"> <ItemContent> <p>Key morphological traits can be successfully extracted by FaunaAIM.</p> </ItemContent> <ItemContent> <p>Automated measurements agree with manual assessments and are transferable across taxa.</p> </ItemContent> <ItemContent> <p>This approach facilitates large-scale, high-throughput trait-based ecology of soil fauna.</p> </ItemContent> </UnorderedList></p><p>Body size is a master functional trait of soil fauna, reflecting interactions among developmental, life-history, physiological, and ecological processes. Though recognized as a critical parameter, traditional manual measurement remains a major bottleneck due to its low efficiency and subjective error from different labors, hindering progress in large-scale, trait-based soil ecology. To address this gap, we developed FaunaAIM, a novel automated tool for high-throughput extraction of key morphological characteristics of soil fauna from images using machine learning. The workflow introduces an attention-based U-Net model integrated with Convolutional Block Attention Modules (CBAM) for individual segmentation, followed by morphological feature calculation using a series of morphological methods. The model achieved high segmentation accuracy (97.3% precision and a mean Intersection over Union (mIoU) of 0.951) on the test set (<i>n</i>=6000). Automated measurements of body length, body width, and area showed strong agreement with manual assessments, with concordance correlation coefficients (CCC) exceeding 0.97 and no significant systematic bias. Notably, the model was trained exclusively on Collembola data, while accurately segmenting mites, demonstrating strong cross-taxon transferability suitable for community-wide analyses. Overall, FaunaAIM substantially improves measurement efficiency and demonstrates excellent transferability, enabling high-throughput morphological characterization of soil fauna and potentially other biological organisms.</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

Automating identification and morphometric measurement of soil fauna with machine learning

  • 期刊论文

  • 发表日期：

    2026年1月20日

查看全文
  • Chenrui Ni,
  • Meng Pan,
  • Wenao Wu,
  • Xudong Wang,
  • Ming Bai,
  • Biao Zhu

摘要

Key morphological traits can be successfully extracted by FaunaAIM.

Automated measurements agree with manual assessments and are transferable across taxa.

This approach facilitates large-scale, high-throughput trait-based ecology of soil fauna.

Body size is a master functional trait of soil fauna, reflecting interactions among developmental, life-history, physiological, and ecological processes. Though recognized as a critical parameter, traditional manual measurement remains a major bottleneck due to its low efficiency and subjective error from different labors, hindering progress in large-scale, trait-based soil ecology. To address this gap, we developed FaunaAIM, a novel automated tool for high-throughput extraction of key morphological characteristics of soil fauna from images using machine learning. The workflow introduces an attention-based U-Net model integrated with Convolutional Block Attention Modules (CBAM) for individual segmentation, followed by morphological feature calculation using a series of morphological methods. The model achieved high segmentation accuracy (97.3% precision and a mean Intersection over Union (mIoU) of 0.951) on the test set (n=6000). Automated measurements of body length, body width, and area showed strong agreement with manual assessments, with concordance correlation coefficients (CCC) exceeding 0.97 and no significant systematic bias. Notably, the model was trained exclusively on Collembola data, while accurately segmenting mites, demonstrating strong cross-taxon transferability suitable for community-wide analyses. Overall, FaunaAIM substantially improves measurement efficiency and demonstrates excellent transferability, enabling high-throughput morphological characterization of soil fauna and potentially other biological organisms.