Scholarlink

<p>Semantic segmentation of remote sensing images plays a crucial role in both geographic information systems and remote sensing. Complex backgrounds and varying target scales are significant challenges to this task. Current mainstream methods, extracting global context, overlook category information’s importance for semantic consistency. Meanwhile, inadequate fusion of deep and shallow features further affects segmentation accuracy. To address these issues, this paper proposes the Class-Aware Cross Fusion Network for remote sensing images. Firstly, we design the Semantic-Aware Global Enhancement Module. Based on learning features with intra-class consistency and inter-class variability, this module employs a cross-attention mechanism to model the relationships between pixels and classes from a global perspective, ensuring accurate capture of global information. To effectively capture multi-scale local information, we present the Multi-Scale Feature Enhancement Module, which leverages Blueprint Separable Convolution and incorporates a channel attention mechanism, improving the recognition capability for targets of different scales. Additionally, we design a Cross-Fusion Segmentation Head with a bidirectional attention mechanism, which can effectively reduce the semantic discrepancy between deep and shallow features and enhance the feature representation ability. Extensive experimental results demonstrate that the proposed model achieves significant performance improvements compared to current popular methods. The codes will be available at <a href="https://github.com/insertall/CACFNet">https://github.com/insertall/CACFNet</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

Semantic segmentation network for remote sensing images based on category-aware cross-fusion

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Zhongwei Zhang,
  • Zhaoyang Zhu,
  • Shudong Liu,
  • Haijie Qin

摘要

Semantic segmentation of remote sensing images plays a crucial role in both geographic information systems and remote sensing. Complex backgrounds and varying target scales are significant challenges to this task. Current mainstream methods, extracting global context, overlook category information’s importance for semantic consistency. Meanwhile, inadequate fusion of deep and shallow features further affects segmentation accuracy. To address these issues, this paper proposes the Class-Aware Cross Fusion Network for remote sensing images. Firstly, we design the Semantic-Aware Global Enhancement Module. Based on learning features with intra-class consistency and inter-class variability, this module employs a cross-attention mechanism to model the relationships between pixels and classes from a global perspective, ensuring accurate capture of global information. To effectively capture multi-scale local information, we present the Multi-Scale Feature Enhancement Module, which leverages Blueprint Separable Convolution and incorporates a channel attention mechanism, improving the recognition capability for targets of different scales. Additionally, we design a Cross-Fusion Segmentation Head with a bidirectional attention mechanism, which can effectively reduce the semantic discrepancy between deep and shallow features and enhance the feature representation ability. Extensive experimental results demonstrate that the proposed model achieves significant performance improvements compared to current popular methods. The codes will be available at https://github.com/insertall/CACFNet.