Scholarlink

<p>Wildfires are a major threat to ecosystems, infrastructure, and lives; therefore, fast and accurate detection systems are crucial. In this paper, we present an intelligent real-time wildfire detection framework, which combines image  processing techniques and Edge Artificial Intelligence (Edge-AI) to achieve improved on-site analysis. The framework uses RGB imagery captured by surveillance cameras and aerial platforms to distinguish fire from non-fire scenarios in different environmental conditions. We show more than one deep learning-based classifier to test the reliability of detection and robustness. The proposed model architecture is tailored to run on the edge with lightweight computing devices and guarantees low-latency inference in a more connectivity-independent manner, dependent on the cloud. Experiments show the promising detection performance achieved with a running time suitable for real-time applications. In general, the results demonstrate that deep learning–based image processing is applicable to practical wildfire monitoring. The highest performance model among the tested was ResNet50, with an accuracy of 98.99%, a precision of 97.92%, a recall of 97.92% and an F1-score of 97.92%.</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

Intelligent real-time wildfire detection using image processing techniques with edge-AI integration

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月13日

查看全文
  • Natesh Mahadev,
  • Archana Bhat,
  • V. L. Sowmya,
  • Anitha Premkumar,
  • R. Shankar,
  • Rajesh Natarajan

摘要

Wildfires are a major threat to ecosystems, infrastructure, and lives; therefore, fast and accurate detection systems are crucial. In this paper, we present an intelligent real-time wildfire detection framework, which combines image  processing techniques and Edge Artificial Intelligence (Edge-AI) to achieve improved on-site analysis. The framework uses RGB imagery captured by surveillance cameras and aerial platforms to distinguish fire from non-fire scenarios in different environmental conditions. We show more than one deep learning-based classifier to test the reliability of detection and robustness. The proposed model architecture is tailored to run on the edge with lightweight computing devices and guarantees low-latency inference in a more connectivity-independent manner, dependent on the cloud. Experiments show the promising detection performance achieved with a running time suitable for real-time applications. In general, the results demonstrate that deep learning–based image processing is applicable to practical wildfire monitoring. The highest performance model among the tested was ResNet50, with an accuracy of 98.99%, a precision of 97.92%, a recall of 97.92% and an F1-score of 97.92%.