Scholarlink

In India, disasters can be caused by naturally occurring events, such as earthquakes, tsunamis, cyclones, landslides, floods, wildfires, and droughts. Globally, around 45,000 people per year are killed by natural disasters. According to the report from the United Nations Office for Disaster Risk Reduction (UNDRR), the total number of people who died between 2000 and 2019 was 79,732. In India, earthquakes (resulting in more than 35,400 deaths since 1950), floods (with an average of 1,600 lives lost per year), and cyclones have a more significant impact compared to landslides and droughts. Therefore, an automated reasoning system for disaster management analysis is mandatorily required. In this research, an efficient model for classifying earthquake, flood, cyclone, and wildfire disaster images using transfer learning techniques has been implemented and analyzed. The customized dataset has four classes of images with varying sizes, which are collected from different sources on Google. Subsequently, these images were pre-processed, divided into training and test set, and used for model creation, validation, and analysis. The pre-trained models chosen for training are ResNet50, EfficientNet-B0, and VGG19Net. Based on the results, the EfficientNet-B0 model outperformed the other models when trained with the Stochastic Gradient Descent (SGD) optimization algorithm. This accuracy of 96.93% highlights the effectiveness of the EfficientNet-B0 architecture in handling the given task and reinforces its potential for similar machine learning challenges in the future.

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

Classification of Disaster Images Using Transfer Learning Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • A. Sivakumar,
  • Apurva Narayan,
  • S. Kavitha

摘要

In India, disasters can be caused by naturally occurring events, such as earthquakes, tsunamis, cyclones, landslides, floods, wildfires, and droughts. Globally, around 45,000 people per year are killed by natural disasters. According to the report from the United Nations Office for Disaster Risk Reduction (UNDRR), the total number of people who died between 2000 and 2019 was 79,732. In India, earthquakes (resulting in more than 35,400 deaths since 1950), floods (with an average of 1,600 lives lost per year), and cyclones have a more significant impact compared to landslides and droughts. Therefore, an automated reasoning system for disaster management analysis is mandatorily required. In this research, an efficient model for classifying earthquake, flood, cyclone, and wildfire disaster images using transfer learning techniques has been implemented and analyzed. The customized dataset has four classes of images with varying sizes, which are collected from different sources on Google. Subsequently, these images were pre-processed, divided into training and test set, and used for model creation, validation, and analysis. The pre-trained models chosen for training are ResNet50, EfficientNet-B0, and VGG19Net. Based on the results, the EfficientNet-B0 model outperformed the other models when trained with the Stochastic Gradient Descent (SGD) optimization algorithm. This accuracy of 96.93% highlights the effectiveness of the EfficientNet-B0 architecture in handling the given task and reinforces its potential for similar machine learning challenges in the future.