Scholarlink

Natural disasters pose global challenges, and integrating AI into disaster management offers significant improvements in preparedness, response, and recovery. This study evaluates the impact of various preprocessing techniques on AI models, focusing on numerical and image data. We specifically explore the CNN-LSTM model, which combines Convolutional Neural Networks (CNNs) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for sequence modeling. We compared preprocessing methods for numerical data, including Z-Score Normalization, PCA, and feature transformations like Logarithmic Transformation, and for image data, such as normalization, augmentation, and transfer learning with Fine-Tuning (ResNet). The CNN-LSTM model, leveraging both spatial and temporal data, showed improved performance, with Z-Score Normalization and Fine-Tuning enhancing model accuracy and robustness. This study highlights the importance of tailored preprocessing techniques and advanced model architectures in optimizing AI for disaster management, demonstrating the value of combining spatial and temporal insights to improve predictive accuracy.

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

An Analysis of Preprocessing Techniques for Disaster Management with CNN-LSTM Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jasmine Samraj,
  • B. Divya

摘要

Natural disasters pose global challenges, and integrating AI into disaster management offers significant improvements in preparedness, response, and recovery. This study evaluates the impact of various preprocessing techniques on AI models, focusing on numerical and image data. We specifically explore the CNN-LSTM model, which combines Convolutional Neural Networks (CNNs) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for sequence modeling. We compared preprocessing methods for numerical data, including Z-Score Normalization, PCA, and feature transformations like Logarithmic Transformation, and for image data, such as normalization, augmentation, and transfer learning with Fine-Tuning (ResNet). The CNN-LSTM model, leveraging both spatial and temporal data, showed improved performance, with Z-Score Normalization and Fine-Tuning enhancing model accuracy and robustness. This study highlights the importance of tailored preprocessing techniques and advanced model architectures in optimizing AI for disaster management, demonstrating the value of combining spatial and temporal insights to improve predictive accuracy.