基于改进型YOLOv5模型和激光SLAM建图提高森林火灾应对能力的研究  

Research on improving forest fire response ability based on improved YOLOv5 model and laser SLAM mapping

作　　者：马晨浩 马江涛 宋金龙 梅良才[1] MA Chenhao;MA Jiangtao;SONG Jinlong;MEI Liangcai(Beijing Institute of Technology,Zhuhai,Zhuhai,Guangdong 519088,China)

机构地区：[1]北京理工大学珠海学院,广东珠海519088

出　　处：《计算机应用文摘》2023年第18期101-104,共4页Chinese Journal of Computer Application

基　　金：广东省大学生创新创业训练计划项目:基于神经网络算法的无人机烟火检测系统研究(S202213675011);北京理工大学珠海学院大数据研究中心及广东省本科高校教学质量与教学改革工程项目:基于SRTP的应用型高校本科生科研能力培养的研究与实践——以数据科学与大数据技术专业为例(XJ⁃2018⁃05)。

摘　　要：文章介绍了一种改进型的YOLOv5模型,它结合了数据增强和伪标签技术,以提高森林火灾检测的准确性。该模型在各种数据集上进行评估,并在检测森林环境中的火源方面实现了最先进的性能。此外,提出了一种使用SLAM(同时定位和映射)技术对森林火灾场景进行建模的新方法。该方法利用LiDAR和RGB-D相机构建环境的3D地图,并实时检测火源。实验结果表明,该方法在准确检测森林火源并及时提供警报以防止潜在灾害方面具有有效性,进而为森林火灾安全应用的先进计算机视觉和机器人技术的发展做出了贡献。The article introduces an improved YOLOv5 model that combines data augmentation and pseudo-labeling techniques to improve the accuracy of forest fire detection.The model is evaluated on various datasets and achieves state-of-the-art.performance in detecting fire sources in forest,environments.In addition,a new method for modeling forest fire scenarios using SLAM(Simultaneous Localization and Mapping)technology is proposed.The method utilizes LiDAR and RGB-D cameras to build a 3D map of the environment and detect fire sources in real time.Experimental results show that the method is effective in accurately detecting forest fire sources and providing alerts in a timely manner to prevent potential disasters,which in turn contributes to the development of advanced computer vision and robotics for forest fire safety applications.

关 键 词：森林防火 数据增强 YOLOv5 目标检测 激光SLAM 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]