Scholarlink

This paper discusses the ongoing development of an energy-efficient YOLO-based fire detection system optimized for edge devices. Using Knowledge Distillation, we compress the YOLOv8m model into YOLOv8n, making it more suitable for deployment on energy-constrained edge devices while maintaining its accuracy. Additionally, we are designing a real-time dynamic energy control mechanism to manage energy usage during the inference process based on real-time power monitoring. Initial results demonstrate that the proposed method reduces model size and power consumption without compromising performance.

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

Energy-Efficient YOLO with Knowledge Distillation and Dynamic Energy Control for Edge Devices

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Anggi Andriyadi,
  • Chandra Wijaya,
  • Shih-Yen Chen,
  • Ding-Hsiang Huang,
  • Chao-Tung Yang

摘要

This paper discusses the ongoing development of an energy-efficient YOLO-based fire detection system optimized for edge devices. Using Knowledge Distillation, we compress the YOLOv8m model into YOLOv8n, making it more suitable for deployment on energy-constrained edge devices while maintaining its accuracy. Additionally, we are designing a real-time dynamic energy control mechanism to manage energy usage during the inference process based on real-time power monitoring. Initial results demonstrate that the proposed method reduces model size and power consumption without compromising performance.