Scholarlink

<p>Macroalgae play a key role in the structure of benthic communities and provide essential ecological services; their responsiveness to stress positions them as indicators of ecosystem health. Traditional manual monitoring methods are resource-demanding and impractical for large areas, prompting the pursuit of more efficient techniques. Remote sensing, particularly with Unmanned Aerial Vehicles (UAVs), has emerged as a valuable solution. To overcome limitations in data accessibility, open-access datasets have become essential for training machine learning algorithms. However, existing datasets do not focus on macroalgae images. This study aims to fill this gap by providing a high-resolution dataset comprising UAV and <i>in situ</i> RGB imagery of 33 intertidal macroalgae from the NE Atlantic, facilitating the development of robust machine-learning models for classification and semantic segmentation of macroalgae. Three sub-datasets are shared: photoquadrats (507 images), orthoimages (7,954 manual polygons), and individual labels (7,685 images). The feasibility of this approach was demonstrated through training a Convolutional Neural Network (CNN) on the dataset created, yielding a test accuracy of 86.72% for 11 classes.</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

Dataset of High-Resolution Aerial Images for Intertidal Macroalgae

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月20日

查看全文
  • Andrea Martínez-Movilla,
  • Marta Román,
  • Gabriel Fontenla-Carrera,
  • Juan Luis Rodríguez-Somoza,
  • Celia Olabarria,
  • Joaquín Martínez-Sánchez

摘要

Macroalgae play a key role in the structure of benthic communities and provide essential ecological services; their responsiveness to stress positions them as indicators of ecosystem health. Traditional manual monitoring methods are resource-demanding and impractical for large areas, prompting the pursuit of more efficient techniques. Remote sensing, particularly with Unmanned Aerial Vehicles (UAVs), has emerged as a valuable solution. To overcome limitations in data accessibility, open-access datasets have become essential for training machine learning algorithms. However, existing datasets do not focus on macroalgae images. This study aims to fill this gap by providing a high-resolution dataset comprising UAV and in situ RGB imagery of 33 intertidal macroalgae from the NE Atlantic, facilitating the development of robust machine-learning models for classification and semantic segmentation of macroalgae. Three sub-datasets are shared: photoquadrats (507 images), orthoimages (7,954 manual polygons), and individual labels (7,685 images). The feasibility of this approach was demonstrated through training a Convolutional Neural Network (CNN) on the dataset created, yielding a test accuracy of 86.72% for 11 classes.