Scholarlink

<p>With the rapid advancement of urbanization, road waterlogging has become a critical issue affecting urban operations and public safety. Under extreme weather conditions, it not only threatens traffic safety but also causes economic losses and cascading system risks. Traditional detection methods based on manual inspection and ground sensors suffer from low efficiency, high cost, and poor adaptability to complex environments. To address these challenges, this study proposes a SAM-Adapter-based method for road waterlogging detection. The approach integrates lightweight adapter modules into the Segment Anything Model (SAM), enabling efficient task-specific adaptation while preserving the strong representation capability of the pre-trained model. This design enhances fine-grained feature extraction and improves segmentation performance in complex urban scenes. Experimental results demonstrate that the proposed method achieves accurate and robust performance across different weather conditions and road types. Compared with conventional methods, it significantly improves detection accuracy and robustness. The main contributions of this study are summarized as follows: (1) A SAM-Adapter-based framework is proposed for urban road waterlogging detection, enabling efficient adaptation of large-scale pre-trained models to task-specific scenarios; (2) A high-quality dataset for urban road waterlogging detection is constructed, covering diverse environmental conditions and road scenes to support robust model training and evaluation; (3) The proposed method demonstrates strong robustness and generalization capability across diverse environmental conditions, offering a practical solution for real-world urban waterlogging monitoring.</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

Detection of Urban Road Waterlogging Using Deep Learning Methods

  • 期刊论文

  • 发表日期：

    2026年5月23日

查看全文
  • Yuncan Gao,
  • Ran Pan,
  • Guoqiang Li,
  • Zhifeng Hu,
  • Wenjun Hu,
  • Qingshan Liu,
  • Ying Zang

摘要

With the rapid advancement of urbanization, road waterlogging has become a critical issue affecting urban operations and public safety. Under extreme weather conditions, it not only threatens traffic safety but also causes economic losses and cascading system risks. Traditional detection methods based on manual inspection and ground sensors suffer from low efficiency, high cost, and poor adaptability to complex environments. To address these challenges, this study proposes a SAM-Adapter-based method for road waterlogging detection. The approach integrates lightweight adapter modules into the Segment Anything Model (SAM), enabling efficient task-specific adaptation while preserving the strong representation capability of the pre-trained model. This design enhances fine-grained feature extraction and improves segmentation performance in complex urban scenes. Experimental results demonstrate that the proposed method achieves accurate and robust performance across different weather conditions and road types. Compared with conventional methods, it significantly improves detection accuracy and robustness. The main contributions of this study are summarized as follows: (1) A SAM-Adapter-based framework is proposed for urban road waterlogging detection, enabling efficient adaptation of large-scale pre-trained models to task-specific scenarios; (2) A high-quality dataset for urban road waterlogging detection is constructed, covering diverse environmental conditions and road scenes to support robust model training and evaluation; (3) The proposed method demonstrates strong robustness and generalization capability across diverse environmental conditions, offering a practical solution for real-world urban waterlogging monitoring.