Scholarlink

<p>Forest fires pose a significant threat to ecosystems, human settlements, and natural resources globally. Timely and accurate detection of forest fires is crucial to mitigate their impact and enhance response efforts. Despite the numerous techniques proposed for forest fire detection, achieving accuracy remains a significant challenge due to the inherent complexity and difficulty of the task. Recent advancements in deep learning have shown promise in improving fire detection capabilities, offering a valuable tool for early warning systems. In this study, we present an approach for forest fire detection using deep learning algorithms. Our method leverages six pre-trained models: AlexNet, GoogLeNet, SqueezeNet, ResNet-18, MobileNetV2, and EfficientNetB0. We conducted an extensive series of experiments to determine the optimal range of their hyperparameters, leading to highly accurate fire detection. These models were fine-tuned using two publicly available and widely used datasets from Kaggle, namely the DeepFire dataset and the Forest Fire Images dataset. The best-performing network on the DeepFire dataset was AlexNet with an accuracy of 99.74%, while ResNet-18 achieved the highest accuracy of 98% on the Forest Fire Images dataset. The proposed approach not only outperforms existing state-of-the-art methods but also provides a robust and reliable solution for wildfire detection.</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

Advanced evaluation of pre-trained CNN models for accurate forest fire detection

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Siham Khirani,
  • Abdelkerim Souahlia,
  • Abdelhalim Rabehi,
  • Mohammed Bourennane,
  • Mawloud Guermoui,
  • Imad Eddine Tibermacine,
  • Abdelaziz Rabehi

摘要

Forest fires pose a significant threat to ecosystems, human settlements, and natural resources globally. Timely and accurate detection of forest fires is crucial to mitigate their impact and enhance response efforts. Despite the numerous techniques proposed for forest fire detection, achieving accuracy remains a significant challenge due to the inherent complexity and difficulty of the task. Recent advancements in deep learning have shown promise in improving fire detection capabilities, offering a valuable tool for early warning systems. In this study, we present an approach for forest fire detection using deep learning algorithms. Our method leverages six pre-trained models: AlexNet, GoogLeNet, SqueezeNet, ResNet-18, MobileNetV2, and EfficientNetB0. We conducted an extensive series of experiments to determine the optimal range of their hyperparameters, leading to highly accurate fire detection. These models were fine-tuned using two publicly available and widely used datasets from Kaggle, namely the DeepFire dataset and the Forest Fire Images dataset. The best-performing network on the DeepFire dataset was AlexNet with an accuracy of 99.74%, while ResNet-18 achieved the highest accuracy of 98% on the Forest Fire Images dataset. The proposed approach not only outperforms existing state-of-the-art methods but also provides a robust and reliable solution for wildfire detection.