Scholarlink

<p>Semantic segmentation is crucial to remote sensing imaging (RSI) analysis of varied landscapes and objects. The Hybrid Feature Adaptive UNet (HFA-UNet) improves RSI semantic segmentation by combining spatial and channel attention modules into a convolutional network. This work intends to improve Land cover, classification in environmental protection, natural resource monitoring, urban land management, and object surveillance. The methodology entails creating a cutting-edge deep learning model trained on a novel dataset targeting the Wular Lake region. The dataset includes labeled semantic segmentation images from Landsat 8 satellites, which have a spectral resolution of 30&#xa0;m. The HFA-UNet architecture incorporates spatial attention modules to capture spatial information and channel attention modules to highlight feature map relationships. The experimental results showcase the impressive performance of the HFA-UNet model, with an accuracy of 99.61%, a Dice Similarity Coefficient (DSC) of 0.952, and a Hausdorff Distance (HD) of 0.0367. The model’s performance is assessed by measuring various metrics, including accuracy, binary accuracy, specificity, AUC, DSC, and HD. Ultimately, the HFA-UNet model greatly enhances the precision of semantic segmentation in RSI, specifically when detecting water coverage in Wular Lake. Future work will prioritize addressing class imbalance issues in RSI segmentation datasets to enhance model performance further.</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

Multiattention network with HFA-UNet for remote sensing semantic segmentation of Wular Lake in India

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月24日

查看全文
  • Monia Digra,
  • Renu Dhir,
  • Arzoo Miglani,
  • Rohit Singh,
  • Suman Bhakar

摘要

Semantic segmentation is crucial to remote sensing imaging (RSI) analysis of varied landscapes and objects. The Hybrid Feature Adaptive UNet (HFA-UNet) improves RSI semantic segmentation by combining spatial and channel attention modules into a convolutional network. This work intends to improve Land cover, classification in environmental protection, natural resource monitoring, urban land management, and object surveillance. The methodology entails creating a cutting-edge deep learning model trained on a novel dataset targeting the Wular Lake region. The dataset includes labeled semantic segmentation images from Landsat 8 satellites, which have a spectral resolution of 30 m. The HFA-UNet architecture incorporates spatial attention modules to capture spatial information and channel attention modules to highlight feature map relationships. The experimental results showcase the impressive performance of the HFA-UNet model, with an accuracy of 99.61%, a Dice Similarity Coefficient (DSC) of 0.952, and a Hausdorff Distance (HD) of 0.0367. The model’s performance is assessed by measuring various metrics, including accuracy, binary accuracy, specificity, AUC, DSC, and HD. Ultimately, the HFA-UNet model greatly enhances the precision of semantic segmentation in RSI, specifically when detecting water coverage in Wular Lake. Future work will prioritize addressing class imbalance issues in RSI segmentation datasets to enhance model performance further.