Scholarlink

<p>Classification methods for Land Cover (LC) typically rely on spectral analysis, which often overlooks pixel spatial relationships and struggles to distinguish classes under varying illumination and humidity conditions. We propose a contextual classification approach utilizing a Visual Vocabulary (VV) of texture descriptors to improve LC inference in satellite images. Unlike traditional spectral models, the incorporation of texture information can address challenges, such as mislabeling due to terrain or illumination changes. This method extracts information from two texture descriptors, the Gray-Level Co-occurrence Matrix (GLCM) and the Local Binary Pattern. These descriptors are then analyzed to identify the most relevant and consistent features to support LC identification. LC patches representing various categories, such as vegetation and urban areas, are generated from Sentinel-2 images. A visual dictionary model constructs a VV for each texture descriptor. The resulting VV vectors are used as input for a Support Vector Machine model to classify Land Cover (LC). The results obtained are then compared with other models, such as Random Forest and Minimum Distance. Our results achieved 93% accuracy in LC classification, demonstrating GLCM texture description the potential for this task. These findings are promising, as they surpass the results of spectral models and contribute to improving the robustness of LC classification to variations in illumination and humidity conditions. Overall, this approach represents a significant advancement in LC classification methodology for satellite imagery.</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

Land cover classification using contextual texture information in sentinel-2 images of southern Mexico

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • B. Flores-Rojas,
  • H. Peregrina-Barreto,
  • S. Camacho-Lara

摘要

Classification methods for Land Cover (LC) typically rely on spectral analysis, which often overlooks pixel spatial relationships and struggles to distinguish classes under varying illumination and humidity conditions. We propose a contextual classification approach utilizing a Visual Vocabulary (VV) of texture descriptors to improve LC inference in satellite images. Unlike traditional spectral models, the incorporation of texture information can address challenges, such as mislabeling due to terrain or illumination changes. This method extracts information from two texture descriptors, the Gray-Level Co-occurrence Matrix (GLCM) and the Local Binary Pattern. These descriptors are then analyzed to identify the most relevant and consistent features to support LC identification. LC patches representing various categories, such as vegetation and urban areas, are generated from Sentinel-2 images. A visual dictionary model constructs a VV for each texture descriptor. The resulting VV vectors are used as input for a Support Vector Machine model to classify Land Cover (LC). The results obtained are then compared with other models, such as Random Forest and Minimum Distance. Our results achieved 93% accuracy in LC classification, demonstrating GLCM texture description the potential for this task. These findings are promising, as they surpass the results of spectral models and contribute to improving the robustness of LC classification to variations in illumination and humidity conditions. Overall, this approach represents a significant advancement in LC classification methodology for satellite imagery.