Scholarlink

<p>Road segmentation in remote sensing images has a wide range of applications in scenarios such as urban planning and autonomous driving. However, due to the complexity and diversity of remote sensing images, deep learning-based methods for road segmentation still entail many challenges: 1) Obstacles such as building shadows and trees often disrupting the road connectivity; 2) Complex environments and diverse road structures make capturing road details difficult. To address these issues, we propose a Multi-Scale Receptive Field Convolutional Road Extraction Network (MG-RoadNet) to retain road details and preserve road connectivity. Firstly, we design an Adaptive Connectivity Field Extractor (ACFE) that assigns an appropriately sized receptive field to each point to extract local connectivity features. Secondly, we design the Global Topological Feature Extractor (GTFE) that captures global topological features for the accurate road localization. Experimental results demonstrate that our method outperforms existing methods in capturing subtle road structures and handling occluded areas. Code is available at: <a href="https://github.com/Kenvily/MG-RoadNet.git">https://github.com/Kenvily/MG-RoadNet.git</a>.</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

MG-RoadNet: Road Segmentation Network for Remote Sensing Images Based on Multi-Receptive Field Graph Convolution

  • 期刊论文

  • 发表日期：

    2025年5月31日

查看全文
  • Runtian Song,
  • Fan Shi,
  • Guikang Du,
  • Xinpeng Zhang,
  • Cheng Jiang

摘要

Road segmentation in remote sensing images has a wide range of applications in scenarios such as urban planning and autonomous driving. However, due to the complexity and diversity of remote sensing images, deep learning-based methods for road segmentation still entail many challenges: 1) Obstacles such as building shadows and trees often disrupting the road connectivity; 2) Complex environments and diverse road structures make capturing road details difficult. To address these issues, we propose a Multi-Scale Receptive Field Convolutional Road Extraction Network (MG-RoadNet) to retain road details and preserve road connectivity. Firstly, we design an Adaptive Connectivity Field Extractor (ACFE) that assigns an appropriately sized receptive field to each point to extract local connectivity features. Secondly, we design the Global Topological Feature Extractor (GTFE) that captures global topological features for the accurate road localization. Experimental results demonstrate that our method outperforms existing methods in capturing subtle road structures and handling occluded areas. Code is available at: https://github.com/Kenvily/MG-RoadNet.git.