Scholarlink

<p>Floods are among the most devastating natural disasters, posing serious threats to human life, infrastructure, and ecosystems. Accurate and timely classification of flooded areas is essential for effective disaster response. This study proposes Flood-FireNet, a novel hybrid model that integrates a transformer-based neural network with the adaptive firefly algorithm (AFA) to enhance flood detection from satellite imagery. The AFA is used to optimize feature selection by identifying the most informative high-level features, while the transformer efficiently captures spatial patterns and long-range dependencies for precise classification. This combined architecture represents the core innovation of the study, leveraging the strengths of both evolutionary optimization and deep learning. Experimental results demonstrate that Flood-FireNet achieves superior performance, with an accuracy of 97.85%, precision of 98.12%, recall of 95.73%, and an F1-score of 96.92%, outperforming several state-of-the-art methods. An ablation study highlights the individual and joint contributions of AFA and the transformer model. Additionally, statistical validation using the paired T-test and ANOVA (p &lt; 0.05) confirms the model’s effectiveness. Overall, Flood-FireNet offers a robust and scalable solution for flood classification tasks, supporting faster and more reliable decision-making in disaster management systems.</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

Flood-FireNet: a novel flood image classification architecture using the adaptive firefly algorithm

  • 期刊论文

  • 发表日期：

    2025年5月27日

查看全文
  • Vinay Dubey,
  • Rahul Katarya

摘要

Floods are among the most devastating natural disasters, posing serious threats to human life, infrastructure, and ecosystems. Accurate and timely classification of flooded areas is essential for effective disaster response. This study proposes Flood-FireNet, a novel hybrid model that integrates a transformer-based neural network with the adaptive firefly algorithm (AFA) to enhance flood detection from satellite imagery. The AFA is used to optimize feature selection by identifying the most informative high-level features, while the transformer efficiently captures spatial patterns and long-range dependencies for precise classification. This combined architecture represents the core innovation of the study, leveraging the strengths of both evolutionary optimization and deep learning. Experimental results demonstrate that Flood-FireNet achieves superior performance, with an accuracy of 97.85%, precision of 98.12%, recall of 95.73%, and an F1-score of 96.92%, outperforming several state-of-the-art methods. An ablation study highlights the individual and joint contributions of AFA and the transformer model. Additionally, statistical validation using the paired T-test and ANOVA (p < 0.05) confirms the model’s effectiveness. Overall, Flood-FireNet offers a robust and scalable solution for flood classification tasks, supporting faster and more reliable decision-making in disaster management systems.