Scholarlink

<p>Ecological monitoring increasingly relies on accurate plant species segmentation using high-resolution remote sensing. This study evaluates the performance and generalization capability of a DeepLabv3Plus-U-Net architecture for segmenting four ecologically important Mediterranean species in the El Ksiba region of the High Atlas, Morocco: <i>Pistacia lentiscus</i>, <i>Tetraclinis articulata</i>, <i>Quercus ilex</i>, and <i>Juniperus oxycedrus</i>. High-resolution UAV imagery was acquired using a DJI Phantom 4 Pro, and 50 representative images were manually annotated to generate training and validation datasets. The proposed model achieved a mean Intersection over Union (MIoU) of 95.75% and a mean pixel accuracy (MPA) of 88.68% on the validation set. These results demonstrate the model’s ability to distinguish morphologically similar species under complex field conditions. Compared with architectures previously tested in another High Atlas region, DeepLabv3Plus-U-Net showed improved generalisation when applied to a different ecological context containing additional species. This study provides evidence that combining UAV imagery with advanced deep learning segmentation can support species-level vegetation mapping in heterogeneous Mediterranean landscapes. The findings contribute to the development of reproducible methodologies for biodiversity assessment and can inform future ecological monitoring and conservation strategies.</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

Deep Learning Techniques for Precise Segmentation of Plant Species: Using Deeplabv3Plus-U-Net and UAV Imagery

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Sara Badrouss,
  • Mohamed Jibril Daiaeddine,
  • El Mostafa Bachaoui,
  • Mohamed Biniz,
  • Hicham Mouncif,
  • Atika Mouaddine,
  • Abdessamad Hilali,
  • Ilias Jennaoui,
  • Abderrazak El Harti,
  • Abdelali Boulli,
  • Abderrahmene El Ghmari

摘要

Ecological monitoring increasingly relies on accurate plant species segmentation using high-resolution remote sensing. This study evaluates the performance and generalization capability of a DeepLabv3Plus-U-Net architecture for segmenting four ecologically important Mediterranean species in the El Ksiba region of the High Atlas, Morocco: Pistacia lentiscus, Tetraclinis articulata, Quercus ilex, and Juniperus oxycedrus. High-resolution UAV imagery was acquired using a DJI Phantom 4 Pro, and 50 representative images were manually annotated to generate training and validation datasets. The proposed model achieved a mean Intersection over Union (MIoU) of 95.75% and a mean pixel accuracy (MPA) of 88.68% on the validation set. These results demonstrate the model’s ability to distinguish morphologically similar species under complex field conditions. Compared with architectures previously tested in another High Atlas region, DeepLabv3Plus-U-Net showed improved generalisation when applied to a different ecological context containing additional species. This study provides evidence that combining UAV imagery with advanced deep learning segmentation can support species-level vegetation mapping in heterogeneous Mediterranean landscapes. The findings contribute to the development of reproducible methodologies for biodiversity assessment and can inform future ecological monitoring and conservation strategies.