Scholarlink

<p>The rapid development of remote sensing sensors and their platforms has dramatically increased the ways to acquire multi-source remote sensing data. However, how to make full use of the complementary characteristics of multi-source data to make up for the similarity between classes has become a challenge for remote sensing intelligent interpretation. Combining the redundancy and complementarity characteristics of multi-source data, firstly, we the Multi-path Connection and Context Module Network (MCM-Net) in this paper. Second, we construct a multi-path cascade mechanism to effectively fuse multi-source data’s complementary features, thereby suppressing noise and improving boundary accuracy. Then, we adopt the context enhancement module to aggregate feature maps of different scales and extract global information to better characterize the category’s discriminative features. Finally, extensive experiments on the US3D dataset demonstrate the effectiveness of our method. Specifically, MCM-Net achieves a mean Intersection over Union (mIoU) of 65.35%, outperforming the baseline U-Net by 3.5% and surpassing other state-of-the-art semantic segmentation methods.</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

Semantic segmentation of multi-source remote sensing image based on multi-path context module

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月9日

查看全文
  • Di Zhang,
  • Yuhang Yan

摘要

The rapid development of remote sensing sensors and their platforms has dramatically increased the ways to acquire multi-source remote sensing data. However, how to make full use of the complementary characteristics of multi-source data to make up for the similarity between classes has become a challenge for remote sensing intelligent interpretation. Combining the redundancy and complementarity characteristics of multi-source data, firstly, we the Multi-path Connection and Context Module Network (MCM-Net) in this paper. Second, we construct a multi-path cascade mechanism to effectively fuse multi-source data’s complementary features, thereby suppressing noise and improving boundary accuracy. Then, we adopt the context enhancement module to aggregate feature maps of different scales and extract global information to better characterize the category’s discriminative features. Finally, extensive experiments on the US3D dataset demonstrate the effectiveness of our method. Specifically, MCM-Net achieves a mean Intersection over Union (mIoU) of 65.35%, outperforming the baseline U-Net by 3.5% and surpassing other state-of-the-art semantic segmentation methods.