Scholarlink

<p>Urban flooding states have been intelligently detected in numerous studies via deep learning algorithms to identify objects associated with floods. However, knowledge gaps remain regarding the performance of detection models across different novel and advanced deep learning models that require different computational powers. Therefore, this study aimed to evaluate the performance of several state-of-the-art deep learning models in detecting urban flood levels based on a dataset of flooded vehicles, highlighting the novelty of applying the latest Non-Maximum Suppression (NMS)-free YOLOv10 architecture to complex urban flood scenarios. Through rigorous benchmark experiments, we evaluated precision, recall, mAP50, and computational latency across multiple models. The findings reveal that the YOLOv10 family provides a strictly superior speed-accuracy trade-off compared to traditional baselines (Faster R-CNN, SSD, and DETR), thereby establishing it as the optimal paradigm for real-time urban flood risk early warning. Specifically, YOLOv10x delivers maximum precision for fixed water-level surveillance infrastructure, while the lightweight YOLOv10n enables highly efficient deployment on edge devices with minimal computational overhead. Conversely, despite its high recall, the prohibitive inference latency of Faster R-CNN renders it impractical for rapid flood response. Ultimately, integrating these highly efficient deep learning detectors into urban surveillance networks can supply critical real-time boundary conditions for hydrodynamic modeling, significantly advancing smart water resources management and urban disaster resilience.</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

Evaluating the Performance of Automatic Detection for Urban Flood Levels Using Different Deep Learning Approaches

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • Zhonglin Zhao,
  • Baohong Lu,
  • Shuo Zhang,
  • Daoli Wang,
  • Jiaquan Wan,
  • Ranyu Liu,
  • Huang Li,
  • Zhensong Wang

摘要

Urban flooding states have been intelligently detected in numerous studies via deep learning algorithms to identify objects associated with floods. However, knowledge gaps remain regarding the performance of detection models across different novel and advanced deep learning models that require different computational powers. Therefore, this study aimed to evaluate the performance of several state-of-the-art deep learning models in detecting urban flood levels based on a dataset of flooded vehicles, highlighting the novelty of applying the latest Non-Maximum Suppression (NMS)-free YOLOv10 architecture to complex urban flood scenarios. Through rigorous benchmark experiments, we evaluated precision, recall, mAP50, and computational latency across multiple models. The findings reveal that the YOLOv10 family provides a strictly superior speed-accuracy trade-off compared to traditional baselines (Faster R-CNN, SSD, and DETR), thereby establishing it as the optimal paradigm for real-time urban flood risk early warning. Specifically, YOLOv10x delivers maximum precision for fixed water-level surveillance infrastructure, while the lightweight YOLOv10n enables highly efficient deployment on edge devices with minimal computational overhead. Conversely, despite its high recall, the prohibitive inference latency of Faster R-CNN renders it impractical for rapid flood response. Ultimately, integrating these highly efficient deep learning detectors into urban surveillance networks can supply critical real-time boundary conditions for hydrodynamic modeling, significantly advancing smart water resources management and urban disaster resilience.