Scholarlink

<p>Aquatic vegetation plays a crucial role in providing ecosystem services for lakes. Remote sensing can effectively reveal the temporal dynamics of lake aquatic vegetation. Traditional dual-branch CNNs rely on static fusion, lacking dynamic weighting and higher-order interactions, thereby limiting representational capacity and generalization in complex aquatic vegetation classification scenarios.Here, taking Shengjin Lake in China as the study area, a novel spatial-spectral attention fusion network frame (SSAF-Net) that integrates spectral and texture image features was developed to classify aquatic vegetation groups. Additionally, the temporal dynamics of aquatic vegetation groups were investigated. The results show that SSAF-Net model can achieve adaptive interaction between spectral and spatial image features through a spatial-spectral cross-attention fusion module, which substantially improves the classification performance. The classification accuracies for Landsat and Sentinel-2 images both exceeded 85%, with kappa coefficients &gt; 0.82, demonstrating transferability and applicability to other lakes, such as Caizi Lake. Since 2000, the emergent and submerged vegetation in Shengjin Lake has exhibited fluctuating upward trends, whereas floating-leaved vegetation tended to decrease. In summary, we have developed a new deep learning classification method suitable for lake aquatic vegetation groups. These findings can provide scientific support for the resource management and ecological restoration of lakes.</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

SSAF-Net Deep Learning Classification Algorithm for Monitoring Aquatic Vegetation Groups in Lakes: A Case Study of Shengjin Lake, China

  • 期刊论文

  • 发表日期：

    2026年7月10日

查看全文
  • Yile Meng,
  • Jie Wang,
  • Zhen Liu,
  • Yuhuan Cui,
  • Yonghui Li

摘要

Aquatic vegetation plays a crucial role in providing ecosystem services for lakes. Remote sensing can effectively reveal the temporal dynamics of lake aquatic vegetation. Traditional dual-branch CNNs rely on static fusion, lacking dynamic weighting and higher-order interactions, thereby limiting representational capacity and generalization in complex aquatic vegetation classification scenarios.Here, taking Shengjin Lake in China as the study area, a novel spatial-spectral attention fusion network frame (SSAF-Net) that integrates spectral and texture image features was developed to classify aquatic vegetation groups. Additionally, the temporal dynamics of aquatic vegetation groups were investigated. The results show that SSAF-Net model can achieve adaptive interaction between spectral and spatial image features through a spatial-spectral cross-attention fusion module, which substantially improves the classification performance. The classification accuracies for Landsat and Sentinel-2 images both exceeded 85%, with kappa coefficients > 0.82, demonstrating transferability and applicability to other lakes, such as Caizi Lake. Since 2000, the emergent and submerged vegetation in Shengjin Lake has exhibited fluctuating upward trends, whereas floating-leaved vegetation tended to decrease. In summary, we have developed a new deep learning classification method suitable for lake aquatic vegetation groups. These findings can provide scientific support for the resource management and ecological restoration of lakes.