基于YOLOv8改进的轻量化森林火灾检测算法  

An Improved Lightweight Forest Fire Detection Algorithm Based on YOLOv8

作　　者：蒋端 彭龑 JIANG Duan;PENG Yan(School of Computer Science and Engineering,Sichuan University of Science and Engineering,Yibin Sichuan 644000,China)

机构地区：[1]四川轻化工大学计算机科学与工程学院,四川宜宾644000

出　　处：《兰州工业学院学报》2025年第2期32-37,共6页Journal of Lanzhou Institute of Technology

基　　金：自贡市科技局科技计划资助项目(2018GYCX33)。

摘　　要：针对目前基于深度学习的森林火灾检测算法参数量和计算量较大,难以在终端部署等问题,提出一种基于YOLOv8n改进的轻量化森林火灾检测模型。在模型的Backbone端和Neck端使用轻量级的C2f_faster模块代替原模型中的C2f模块,降低模型的参数量和内存占用大小;在模型的Neck端引入DySample进行上采样,在不增加计算负担的同时保证检测精度。实验结果表明,改进后的算法mAP@0.5值达到92%,相较于YOLOv8,新算法的参数量减少约22.7%,GFLOPs减少约21%,Size减少约22.2%,为森林火灾检测的轻量化研究提供理论参考。Aiming at the problem that the current forest fire detection algorithm based on deep learning is difficult to deploy in terminal due to the large number of parameters and computation,an improved lightweight forest fire detection model based on YOLOv8n is proposed.The lightweight C2f_faster module is used to replace the C2f module in the Backbone and Neck ends of the model to reduce the number of parameters and memory footprint of the model.DySample is introduced into the Neck end of the model for up-sampling,which ensures detection accuracy without increasing computational burden.The experimental results show that the mAP@0.5 value of the improved algorithm reaches 92%,compared with YOLOv8,the new algorithm reduces the number of parameters by 22.7%,GFLOPs by 21%,and Size by 22.2%,which provides a theoretical reference for the lightweight research of forest fire detection.

关 键 词：森林火灾检测 轻量化 DySample YOLOv8 

分 类 号：X932[环境科学与工程—安全科学]