Scholarlink

<p>Prompt detection of forest fires is vital to curb the ruinous effects of wildfires, yet expansive terrains, erratic flames, and shifting conditions pose daunting challenges. Conventional sensors, like those for smoke or heat, lack broad coverage and often yield false positives in dynamic settings. Many modern deep learning models, being computationally demanding, are impractical for real-time use in remote forests. To address these challenges, this paper proposes a lightweight forest fire smoke detection model based on the YOLOv11n framework, which integrates novel convolutional modules to enhance detection accuracy and efficiency. Specifically, the GEConv module improves local feature extraction with minimal computational cost, ensuring accurate flame localization. The FastStarConv module optimizes the neck structure for robust feature fusion, which enhances performance in complex environments. The StrixDetect detection head reduced model parameters and inference time while preserving detection accuracy. Experimental evaluations on public forest fire detection datasets demonstrated that the proposed model achieves a 3.1% increase in detection accuracy and a 12% improvement in inference speed compared to the baseline YOLOv11n. Furthermore, when benchmarked against other state-of-the-art lightweight detection models, it exhibits superior performance in both precision and efficiency. These results highlight the model’s effectiveness and suitability for practical deployment in forest fire monitoring systems, offering a balanced solution between performance and resource constraints.</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

Lightweight multi-module convolutional network for efficient forest fire smoke detection

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Jianjia Guan,
  • Yuping Sun

摘要

Prompt detection of forest fires is vital to curb the ruinous effects of wildfires, yet expansive terrains, erratic flames, and shifting conditions pose daunting challenges. Conventional sensors, like those for smoke or heat, lack broad coverage and often yield false positives in dynamic settings. Many modern deep learning models, being computationally demanding, are impractical for real-time use in remote forests. To address these challenges, this paper proposes a lightweight forest fire smoke detection model based on the YOLOv11n framework, which integrates novel convolutional modules to enhance detection accuracy and efficiency. Specifically, the GEConv module improves local feature extraction with minimal computational cost, ensuring accurate flame localization. The FastStarConv module optimizes the neck structure for robust feature fusion, which enhances performance in complex environments. The StrixDetect detection head reduced model parameters and inference time while preserving detection accuracy. Experimental evaluations on public forest fire detection datasets demonstrated that the proposed model achieves a 3.1% increase in detection accuracy and a 12% improvement in inference speed compared to the baseline YOLOv11n. Furthermore, when benchmarked against other state-of-the-art lightweight detection models, it exhibits superior performance in both precision and efficiency. These results highlight the model’s effectiveness and suitability for practical deployment in forest fire monitoring systems, offering a balanced solution between performance and resource constraints.