Scholarlink

<p>This study highlights the vital role of high-resolution (HR), open-source land cover maps for food security, land use planning, and environmental protection. The scarcity of freely available HR datasets underscores the importance of multi-spectral HR aerial images. We used unmanned aerial vehicle (UAV) to capture images for a centimeter-level orthomosaics, facilitating advanced remote sensing and spatial analysis. Our method compares the efficacy and accuracy of object-based image analysis (OBIA) combined with random forest and convolutional neural networks (CNN) for land cover classification. We produced detailed land cover maps for 27 varied landscapes across Serbia, identifying nine unique land cover classes and assessing human impact on natural habitats. This resulted in a valuable dataset of HR multi-spectral orthomosaics across ecological zones, alongside land cover classification with extensive metrics and training data for each site. This dataset is a valuable resource for researchers working on habitats mapping and assessment for biodiversity monitoring studies on one side and researchers working on novel machine learning methods for land cover classification.</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

Comprehensive dataset from high resolution UAV land cover mapping of diverse natural environments in Serbia

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年1月14日

查看全文
  • Bojana Ivošević,
  • Nina Pajević,
  • Sanja Brdar,
  • Rana Waqar,
  • Maryam Khan,
  • João Valente

摘要

This study highlights the vital role of high-resolution (HR), open-source land cover maps for food security, land use planning, and environmental protection. The scarcity of freely available HR datasets underscores the importance of multi-spectral HR aerial images. We used unmanned aerial vehicle (UAV) to capture images for a centimeter-level orthomosaics, facilitating advanced remote sensing and spatial analysis. Our method compares the efficacy and accuracy of object-based image analysis (OBIA) combined with random forest and convolutional neural networks (CNN) for land cover classification. We produced detailed land cover maps for 27 varied landscapes across Serbia, identifying nine unique land cover classes and assessing human impact on natural habitats. This resulted in a valuable dataset of HR multi-spectral orthomosaics across ecological zones, alongside land cover classification with extensive metrics and training data for each site. This dataset is a valuable resource for researchers working on habitats mapping and assessment for biodiversity monitoring studies on one side and researchers working on novel machine learning methods for land cover classification.