Scholarlink

Geographic Information Systems (GIS) integrate hardware, software, and geospatial data for effective spatial analysis and decision-making. Recent advancements in satellite sensors and Unmanned Aerial Systems (UAS) have positioned GIS as a critical tool in diverse domains, including road networks, environmental management, and infrastructure planning. This chapter explores the integration of deep learning methodologies in geospatial image classification, addressing challenges in feature extraction and class boundary delineation for multi-scale satellite imagery and hyperspectral images (HSI). Three innovative models are presented: a hybrid ResNet152-Extreme Learning Machine (EML) achieving up to 99.83% accuracy through advanced image decomposition techniques; a CNN-GCNN framework enhancing HSI classification with accuracy improvements from 97.48 to 99.68%; and a novel GCNN-SWIN Transformer (G_SwTr) model achieving an overall accuracy of 99.7% and a kappa coefficient of 99.6%. These approaches significantly advance remote sensing analysis, delivering superior classification performance and enabling more precise geospatial data interpretation.

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

Deep Learning Approaches for Geospatial Data Classification

  • 章节

  • 发表日期：

    2025年

查看全文
  • Rahul K. Ghotekar,
  • Minakhi Rout,
  • Kailash Shaw

摘要

Geographic Information Systems (GIS) integrate hardware, software, and geospatial data for effective spatial analysis and decision-making. Recent advancements in satellite sensors and Unmanned Aerial Systems (UAS) have positioned GIS as a critical tool in diverse domains, including road networks, environmental management, and infrastructure planning. This chapter explores the integration of deep learning methodologies in geospatial image classification, addressing challenges in feature extraction and class boundary delineation for multi-scale satellite imagery and hyperspectral images (HSI). Three innovative models are presented: a hybrid ResNet152-Extreme Learning Machine (EML) achieving up to 99.83% accuracy through advanced image decomposition techniques; a CNN-GCNN framework enhancing HSI classification with accuracy improvements from 97.48 to 99.68%; and a novel GCNN-SWIN Transformer (G_SwTr) model achieving an overall accuracy of 99.7% and a kappa coefficient of 99.6%. These approaches significantly advance remote sensing analysis, delivering superior classification performance and enabling more precise geospatial data interpretation.