Scholarlink

<p>This paper proposes a deep learning-based intelligent solution for monitoring underpass waterlogging. Due to substandard infrastructure design and inefficient drainage systems, underpass flooding poses a major urban transportation hazard. Further, this causes traffic disruption, safety risks, with infrastructure-related losses. The proposed solution uses convolutional neural networks (CNNs) and transfer learning for automated detection of underpass waterlogging conditions. A customized dataset of 1,200 underpass images, comprising flooded and normal scenarios, is constructed and used to train a binary classification model. Among multiple pre-trained CNN architectures, a fine tuned ResNet-50 model achieves the best performance for underpass flood detection. Due to its low inference latency, the proposed solution can be considered suitable for real-time deployment in intelligent transportation systems. Furthermore, the integration of early warning mechanisms and optimized traffic rerouting strategies can significantly mitigate risks during monsoon seasons and extreme weather events. Experimental results demonstrate that the proposed model achieves 92.00% test accuracy, with 92.57% precision, 91.33% recall, and 91.95% F1-score, outperforming some of the baseline approaches and demonstrating strong practical potential for scalable smart transportation 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

Deep convolutional neural networks for underpass flood detection

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Shuvabrata Bandopadhaya,
  • Amarjit Roy,
  • Soumya Ranjan Samal,
  • Nilanjan Dey,
  • Ameya Mudgal

摘要

This paper proposes a deep learning-based intelligent solution for monitoring underpass waterlogging. Due to substandard infrastructure design and inefficient drainage systems, underpass flooding poses a major urban transportation hazard. Further, this causes traffic disruption, safety risks, with infrastructure-related losses. The proposed solution uses convolutional neural networks (CNNs) and transfer learning for automated detection of underpass waterlogging conditions. A customized dataset of 1,200 underpass images, comprising flooded and normal scenarios, is constructed and used to train a binary classification model. Among multiple pre-trained CNN architectures, a fine tuned ResNet-50 model achieves the best performance for underpass flood detection. Due to its low inference latency, the proposed solution can be considered suitable for real-time deployment in intelligent transportation systems. Furthermore, the integration of early warning mechanisms and optimized traffic rerouting strategies can significantly mitigate risks during monsoon seasons and extreme weather events. Experimental results demonstrate that the proposed model achieves 92.00% test accuracy, with 92.57% precision, 91.33% recall, and 91.95% F1-score, outperforming some of the baseline approaches and demonstrating strong practical potential for scalable smart transportation systems.