Scholarlink

This study analyzes texture features in Sentinel-2 Level 2A data from the post-cyclone phase (01-01-2021 to 31-12-2022), focusing on key bands such as green (B3), near infrared (B8), and shortwave infrared (B11). Several properties of texture are captured using the Gray-Level Co-occurrence Matrix (GLCM) and Haralick features, including contrast, dissimilarity, homogeneity, energy, and correlation, with GLCM computed using a sliding window method and features normalized using z-score standardization. The standardized feature vectors are clustered using the K-means method, and the ideal number of clusters is established by utilizing the silhouette score and elbow method. Results show high contrast at \(\theta \) = 90 \(^{\circ }\) and \(\theta \) = 135 \(^{\circ }\) , decreasing with distance, with homogeneity peaking at \(\theta \) = 0 \(^{\circ }\) and \(\theta \) = 45 \(^{\circ }\) . Dissimilarity patterns mirror contrast results, energy values are high at \(\theta \) = 0 \(^{\circ }\) and \(\theta \) = 90 \(^{\circ }\) , and correlation is strong at \(\theta \) = 45 \(^{\circ }\) and \(\theta \) = 135 \(^{\circ }\) . The ideal number of clusters is identified as three, balancing interpretability and clustering quality. This approach effectively distinguishes texture variations of our study area in the Saptamukhi Reserve Forest in the Sundarbans. Field observations revealed six different species predominantly falling within three major communities, supporting the clustering results and suggesting a hierarchical structure where species are grouped into broader ecological categories. This empirical observation reinforces the effectiveness of the three-cluster solution in capturing major community structures while recognizing sub-variations within those communities, providing valuable insights for further analysis.

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

Texture Feature Analysis and Clustering of Remotely Sensed Data Using Gray-Level Co-occurrence Matrix and Haralick Features

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Anindita Das Bhattacharjee,
  • Nilanjan Ghosh

摘要

This study analyzes texture features in Sentinel-2 Level 2A data from the post-cyclone phase (01-01-2021 to 31-12-2022), focusing on key bands such as green (B3), near infrared (B8), and shortwave infrared (B11). Several properties of texture are captured using the Gray-Level Co-occurrence Matrix (GLCM) and Haralick features, including contrast, dissimilarity, homogeneity, energy, and correlation, with GLCM computed using a sliding window method and features normalized using z-score standardization. The standardized feature vectors are clustered using the K-means method, and the ideal number of clusters is established by utilizing the silhouette score and elbow method. Results show high contrast at \(\theta \) = 90 \(^{\circ }\) and \(\theta \) = 135 \(^{\circ }\) , decreasing with distance, with homogeneity peaking at \(\theta \) = 0 \(^{\circ }\) and \(\theta \) = 45 \(^{\circ }\) . Dissimilarity patterns mirror contrast results, energy values are high at \(\theta \) = 0 \(^{\circ }\) and \(\theta \) = 90 \(^{\circ }\) , and correlation is strong at \(\theta \) = 45 \(^{\circ }\) and \(\theta \) = 135 \(^{\circ }\) . The ideal number of clusters is identified as three, balancing interpretability and clustering quality. This approach effectively distinguishes texture variations of our study area in the Saptamukhi Reserve Forest in the Sundarbans. Field observations revealed six different species predominantly falling within three major communities, supporting the clustering results and suggesting a hierarchical structure where species are grouped into broader ecological categories. This empirical observation reinforces the effectiveness of the three-cluster solution in capturing major community structures while recognizing sub-variations within those communities, providing valuable insights for further analysis.