Scholarlink

This paper proposes an integrated approach to automate disaster damage detection, addressing spatiotemporal misalignments, data imbalance problems, and real-time implementation. It combines Mutual Information-Based Image Registration (MI-IR), Synthetic Minority Oversampling Technique (SMOTE), with a customized Convolutional Neural Network (CNN) optimized for binary pixel classification on pre- and post-disaster satellite images. MI-IR ensures very accurate spatial alignment and minimizes registration errors, whereas SMOTE handles data imbalance so that classification performance improves. Custom CNN efficiently detects pattern damage with high accuracy of 99.84%. The results show better alignment of the images, structural integrity along with the accurate damage classification under different conditions. It shows its suitability for application in disaster response, urban planning, and environmental monitoring. This work advances the development of operational systems for real-time disaster management.

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

CNN-Based Automated Disaster Damage Assessment Using Satellite Imagery

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nandini Parashar,
  • Shelly Bhalla,
  • Riya Gupta,
  • Greeshma Arya

摘要

This paper proposes an integrated approach to automate disaster damage detection, addressing spatiotemporal misalignments, data imbalance problems, and real-time implementation. It combines Mutual Information-Based Image Registration (MI-IR), Synthetic Minority Oversampling Technique (SMOTE), with a customized Convolutional Neural Network (CNN) optimized for binary pixel classification on pre- and post-disaster satellite images. MI-IR ensures very accurate spatial alignment and minimizes registration errors, whereas SMOTE handles data imbalance so that classification performance improves. Custom CNN efficiently detects pattern damage with high accuracy of 99.84%. The results show better alignment of the images, structural integrity along with the accurate damage classification under different conditions. It shows its suitability for application in disaster response, urban planning, and environmental monitoring. This work advances the development of operational systems for real-time disaster management.