Scholarlink

<p>Growing cities and its extending urban built-up are required to be monitored timely for catering to&#xa0;the needs of sustainable development goals. Extraction of urban built-up areas has evolved with the advancement of computer vision and easily available satellite remote sensing data. However, there is still scope for improvement in classifying unplanned cities dominated with scattered development and cities in coastal regions where urban built-up areas are often mixed with other land cover classes. In this study, we target these cases and utilize multi-sensor and multi-channel data from sentinel-1 and sentinel-2 satellites. For classification, the deep learning model is proposed, the fine tuned VGG16 UNet-AP model with six bands satellite image network, it is a CNN based on transfer learning for multi-channel data utilizing the encoder-decoder network. The model achieved IOU and F1-score of 0.85 and 0.92 respectively which is significantly high when compared with other state-of-the-art networks.</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

Deep Learning Based Urban Built-Up Extraction for Scattered Development and Coastal Cities

  • 期刊论文

  • 发表日期：

    2025年6月16日

查看全文
  • Kriti Rastogi,
  • Shashikant A. Sharma

摘要

Growing cities and its extending urban built-up are required to be monitored timely for catering to the needs of sustainable development goals. Extraction of urban built-up areas has evolved with the advancement of computer vision and easily available satellite remote sensing data. However, there is still scope for improvement in classifying unplanned cities dominated with scattered development and cities in coastal regions where urban built-up areas are often mixed with other land cover classes. In this study, we target these cases and utilize multi-sensor and multi-channel data from sentinel-1 and sentinel-2 satellites. For classification, the deep learning model is proposed, the fine tuned VGG16 UNet-AP model with six bands satellite image network, it is a CNN based on transfer learning for multi-channel data utilizing the encoder-decoder network. The model achieved IOU and F1-score of 0.85 and 0.92 respectively which is significantly high when compared with other state-of-the-art networks.