Scholarlink

<p>Traditional methods face challenges due to the complexity of high-resolution satellite images, which exhibit diverse features including spectral variations, intricate textures, irregular shapes, spatial relationships, and temporal dynamics. This study explores the efficacy of advanced deep learning models in enhancing satellite remote sensing data classification. We evaluate various Convolutional Neural Network (CNN) architectures, such as SqueezeNet, MobileNetV3, ResNeXt, and VGG19, utilizing three prominent datasets: DeepGlobe, RSI-CB, and Aerial Image Dataset (AID). These datasets encompass a wide range of image categories relevant to land use and land cover (LULC) classification. In our experiments, SqueezeNet, MobileNetV3, and ResNeXt achieved classification accuracies of 98.0%, 96.6%, and 97.2%, respectively, on the DeepGlobe dataset, with similarly strong performance exceeding 92% across the RSI-CB and AID datasets. The models also attained high F1-scores reaching up to 97.2%, demonstrating their robustness in capturing complex spatial and spectral patterns in high-resolution satellite imagery. The findings demonstrate the potential of deep learning techniques to effectively capture complex patterns in high-resolution satellite imagery, resulting in substantial improvements in classification performance. This research contributes to the field of satellite image analysis, with implications for environmental monitoring, urban planning, and disaster response, highlighting the transformative impact of deep learning in remote sensing applications.</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

Comparative analysis of deep learning methods for high-resolution satellite remote sensing image classification

  • 期刊论文

  • 发表日期：

    2026年4月16日

查看全文
  • Ejaz Ul Haq,
  • Fung Fung Ting,
  • Raphael C. -W. Phan,
  • Mansour Almazroui,
  • Rasha M. Abou Samra,
  • Chee-Ming Ting

摘要

Traditional methods face challenges due to the complexity of high-resolution satellite images, which exhibit diverse features including spectral variations, intricate textures, irregular shapes, spatial relationships, and temporal dynamics. This study explores the efficacy of advanced deep learning models in enhancing satellite remote sensing data classification. We evaluate various Convolutional Neural Network (CNN) architectures, such as SqueezeNet, MobileNetV3, ResNeXt, and VGG19, utilizing three prominent datasets: DeepGlobe, RSI-CB, and Aerial Image Dataset (AID). These datasets encompass a wide range of image categories relevant to land use and land cover (LULC) classification. In our experiments, SqueezeNet, MobileNetV3, and ResNeXt achieved classification accuracies of 98.0%, 96.6%, and 97.2%, respectively, on the DeepGlobe dataset, with similarly strong performance exceeding 92% across the RSI-CB and AID datasets. The models also attained high F1-scores reaching up to 97.2%, demonstrating their robustness in capturing complex spatial and spectral patterns in high-resolution satellite imagery. The findings demonstrate the potential of deep learning techniques to effectively capture complex patterns in high-resolution satellite imagery, resulting in substantial improvements in classification performance. This research contributes to the field of satellite image analysis, with implications for environmental monitoring, urban planning, and disaster response, highlighting the transformative impact of deep learning in remote sensing applications.