Scholarlink

<p>Building extraction is a critical area of research with diverse applications in urban planning and development. This study focuses on extracting building roofs from aerial imagery of Ahvaz city to facilitate the updating and refinement of urban maps. To achieve this, variants of several deep learning architectures, such as U-Net, U-Net++, LinkNet, DeepLabv3+, UNETR, and Attention U-Net, were evaluated. Furthermore, a non-local attention embedded ResNet-UNet (ResUnet-NL) was proposed to enhance performance which was further improved through the implementation of an advanced training strategy. The accuracy of patch connections in test images was also refined using Seamless Patch Integration with Overlapping Patches (SPIOP). The experimental results indicate that the proposed architecture achieves superior performance, with an Intersection over Union (IoU) of 86% for building extraction and an average IoU of 92%. Additionally, it shows improved performance on blurry and low-quality images. The findings suggest that the extracted buildings can be effectively employed to ensure the precise updating of urban maps in Ahvaz.</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

Building roof extraction from aerial imagery using resUNet- nonlocal attention block and seamless patch integration: a case study in Ahvaz, Iran

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Mehrtash Manafifard

摘要

Building extraction is a critical area of research with diverse applications in urban planning and development. This study focuses on extracting building roofs from aerial imagery of Ahvaz city to facilitate the updating and refinement of urban maps. To achieve this, variants of several deep learning architectures, such as U-Net, U-Net++, LinkNet, DeepLabv3+, UNETR, and Attention U-Net, were evaluated. Furthermore, a non-local attention embedded ResNet-UNet (ResUnet-NL) was proposed to enhance performance which was further improved through the implementation of an advanced training strategy. The accuracy of patch connections in test images was also refined using Seamless Patch Integration with Overlapping Patches (SPIOP). The experimental results indicate that the proposed architecture achieves superior performance, with an Intersection over Union (IoU) of 86% for building extraction and an average IoU of 92%. Additionally, it shows improved performance on blurry and low-quality images. The findings suggest that the extracted buildings can be effectively employed to ensure the precise updating of urban maps in Ahvaz.