Scholarlink

In the critical context of enhancing post-disaster assessments in regions frequently affected by earthquakes and typhoons, this study utilizes the YOLOv8 algorithm to innovate the classification of building damage, focusing on the Noto area in Japan. By employing a meticulously augmented dataset and optimizing the model over 200 epochs with the NAdam optimizer, our approach distinguishes itself through its efficiency and accuracy. The exceptional performance of the model is underscored by an overall mAP score of 0.952 and precision nearing 1.00 at higher confidence thresholds, markedly outperforming conventional methods of damage assessment. Moreover, with a recall rate of 0.8, YOLOv8 exhibits strong detection capabilities across various types of structural damage. This research not only introduces an innovative application of YOLOv8 for disaster response but also establishes a new benchmark in emergency management practices, underscoring the potential of advanced deep learning techniques in lessening the impact of natural disasters worldwide.

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

Using YOLOv8 for Building Damage Identification in Japan's Noto Region Following Earthquakes: A Deep Learning-Based Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chan Gao,
  • Genfeng Zhao,
  • Sen Gao,
  • Eunyoung Kim

摘要

In the critical context of enhancing post-disaster assessments in regions frequently affected by earthquakes and typhoons, this study utilizes the YOLOv8 algorithm to innovate the classification of building damage, focusing on the Noto area in Japan. By employing a meticulously augmented dataset and optimizing the model over 200 epochs with the NAdam optimizer, our approach distinguishes itself through its efficiency and accuracy. The exceptional performance of the model is underscored by an overall mAP score of 0.952 and precision nearing 1.00 at higher confidence thresholds, markedly outperforming conventional methods of damage assessment. Moreover, with a recall rate of 0.8, YOLOv8 exhibits strong detection capabilities across various types of structural damage. This research not only introduces an innovative application of YOLOv8 for disaster response but also establishes a new benchmark in emergency management practices, underscoring the potential of advanced deep learning techniques in lessening the impact of natural disasters worldwide.