Scholarlink

<p>Land-use land-cover (LULC) classification and change detection are vital for environmental management, urban planning, and climate monitoring. This study introduces a multi-self-organizing map (Multi-SOM) model that overcomes the challenges of high dimensionality, spectral variability, and spatial complexity in satellite imagery analysis. The algorithm performs automated clustering using a Multi-SOM architecture that integrates independent SOMs to improve segmentation accuracy by averaging weight grids. Clusters are then labeled based on spectral statistics, minimizing manual interventions. For LULC classification, a multilayer perceptron (MLP) model is proposed, designed with a structured architecture and regularization techniques to outperform traditional classifiers such as support vector machine (SVM) and logistic regression. The methodology is applied to Landsat imagery of the Silchar region, India, for 1988 and 2024. Results demonstrate that the Multi-SOM achieves well-separated clusters with high compactness, while the proposed MLP model delivers superior classification performance with a weighted average F1 score of 0.99. Change detection analysis, based on the classified images, highlights an 11.15% and 6.06% increase in the built-up area of the Silchar and Karimganj regions, respectively. These technical contributions establish the framework as a robust and scalable solution for LULC analysis and sustainable land management.</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

Multi-SOM-Based Multispectral Image Segmentation for LULC Classification and Change Detection

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Somnath Mukhopadhyay,
  • Wangjam Niranjan Singh,
  • Sunita Sarkar,
  • Sheikh Wakie Masood,
  • Ajoy Kumar Khan

摘要

Land-use land-cover (LULC) classification and change detection are vital for environmental management, urban planning, and climate monitoring. This study introduces a multi-self-organizing map (Multi-SOM) model that overcomes the challenges of high dimensionality, spectral variability, and spatial complexity in satellite imagery analysis. The algorithm performs automated clustering using a Multi-SOM architecture that integrates independent SOMs to improve segmentation accuracy by averaging weight grids. Clusters are then labeled based on spectral statistics, minimizing manual interventions. For LULC classification, a multilayer perceptron (MLP) model is proposed, designed with a structured architecture and regularization techniques to outperform traditional classifiers such as support vector machine (SVM) and logistic regression. The methodology is applied to Landsat imagery of the Silchar region, India, for 1988 and 2024. Results demonstrate that the Multi-SOM achieves well-separated clusters with high compactness, while the proposed MLP model delivers superior classification performance with a weighted average F1 score of 0.99. Change detection analysis, based on the classified images, highlights an 11.15% and 6.06% increase in the built-up area of the Silchar and Karimganj regions, respectively. These technical contributions establish the framework as a robust and scalable solution for LULC analysis and sustainable land management.