Scholarlink

<p>Flood is the most alarming and devastating natural disasters, which poses significant threats to communities around the world. Flood causes significant destruction to critical infrastructure, human lives, and properties. Thus, it is important to detect humans in flood-related disasters. Nowadays, various techniques are employed for human object classification in a flooded environment, but these did not maintain high classification accuracy and often misclassified objects as humans. This paper presents a Fast Gated Recurrent Neural Network-based Mean Binary cross entropy Taylor concept (FastGRNN-MBT) for classifying human objects in flooded environments. The input image is pre-processed initially by utilizing an adaptive median filter, and the image enhancement of pre-processed image is performed by utilizing Retinex algorithm. Then, the objects are detected using You Only Look Once version 12 nano (YOLOv12n), where the detection performance is improved by fine-tuning YOLOv12n utilizing CAO algorithm. Following this, the different features are extracted and the human objects are finally classified using FastGRNN-MBT. Further, the experimental investigations are performed to validate the supremacy of FastGRNN-MBT in classifying human objects. The FastGRNN-MBT outperforms existing schemes with recall, precision, accuracy, execution time, inference time and F1-score of 96.979%, 97.763%, 96.226%, 147.236&#xa0;s, 28.178&#xa0;s, and 96.988% which indicates that it is better suitable for the classification task as compared to other state of the art classification methods.</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

A novel framework for human object classification in flooded environments using YOLOv12n and FastGRNN-MBT

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月28日

查看全文
  • Pallavi Nehete,
  • Madhuri Dharrao,
  • Anupkumar M. Bongale,
  • Deepak Dharrao

摘要

Flood is the most alarming and devastating natural disasters, which poses significant threats to communities around the world. Flood causes significant destruction to critical infrastructure, human lives, and properties. Thus, it is important to detect humans in flood-related disasters. Nowadays, various techniques are employed for human object classification in a flooded environment, but these did not maintain high classification accuracy and often misclassified objects as humans. This paper presents a Fast Gated Recurrent Neural Network-based Mean Binary cross entropy Taylor concept (FastGRNN-MBT) for classifying human objects in flooded environments. The input image is pre-processed initially by utilizing an adaptive median filter, and the image enhancement of pre-processed image is performed by utilizing Retinex algorithm. Then, the objects are detected using You Only Look Once version 12 nano (YOLOv12n), where the detection performance is improved by fine-tuning YOLOv12n utilizing CAO algorithm. Following this, the different features are extracted and the human objects are finally classified using FastGRNN-MBT. Further, the experimental investigations are performed to validate the supremacy of FastGRNN-MBT in classifying human objects. The FastGRNN-MBT outperforms existing schemes with recall, precision, accuracy, execution time, inference time and F1-score of 96.979%, 97.763%, 96.226%, 147.236 s, 28.178 s, and 96.988% which indicates that it is better suitable for the classification task as compared to other state of the art classification methods.