Scholarlink

<p>Land cover classification in remote sensing is a challenging area of research because of the mixed pixels, noise, and vague class boundaries. To address these challenges, this study focuses on hybrid methods such as Fuzzy C-Mean (FCM), which is a soft classification method that works on mixed pixel membership values. Its performance is enhanced with the combination of the learnable Gaussian filter (LGF) and the learnable bilateral filter (LBF), which added spatial regularization and also added the ensemble method for better accuracy. By focusing on the uncertainty, uncertainty-weighted random forest (UWRF) and uncertainty-weighted XGBoost (UWXG) are used for classification. The learnable filters optimize their parameters to preserve edge information while reducing the noise in fuzzy membership maps. Shannon entropy is used to measure the classification uncertainty, and it is used in sample weighting in ensemble classifiers. Multispectral Landsat 8 images of the Nainital region are used to examine the performance of hybrid methods. The proposed method FCM + LBF + UWXG shows better performance with an accuracy of 95.4% and a kappa coefficient of 0.94 in comparison to traditional methods. The enhanced classification precision and spatial consistency render this method appropriate for practical usage in land use planning, environmental observation, and resource control in heterogeneous landscapes.</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

Soft classification with learnable spatial filters using bayesian optimization and uncertainty-guided XGBoost for enhanced land cover mapping

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Swati Vishnoi,
  • Meenakshi Pareek

摘要

Land cover classification in remote sensing is a challenging area of research because of the mixed pixels, noise, and vague class boundaries. To address these challenges, this study focuses on hybrid methods such as Fuzzy C-Mean (FCM), which is a soft classification method that works on mixed pixel membership values. Its performance is enhanced with the combination of the learnable Gaussian filter (LGF) and the learnable bilateral filter (LBF), which added spatial regularization and also added the ensemble method for better accuracy. By focusing on the uncertainty, uncertainty-weighted random forest (UWRF) and uncertainty-weighted XGBoost (UWXG) are used for classification. The learnable filters optimize their parameters to preserve edge information while reducing the noise in fuzzy membership maps. Shannon entropy is used to measure the classification uncertainty, and it is used in sample weighting in ensemble classifiers. Multispectral Landsat 8 images of the Nainital region are used to examine the performance of hybrid methods. The proposed method FCM + LBF + UWXG shows better performance with an accuracy of 95.4% and a kappa coefficient of 0.94 in comparison to traditional methods. The enhanced classification precision and spatial consistency render this method appropriate for practical usage in land use planning, environmental observation, and resource control in heterogeneous landscapes.