摘要
相较于传统烟火、烟雾检测,基于卷积神经网络算法的烟火检测具有更高检测精度和效率,提出基于改进YOLOv3算法的烟火识别方法,应用高斯参数设计损失函数并建立YOLOv3边界框模型,实现边界框置信度计算以减少负样本.为充分利用图像局部特征信息,对网络结构进行改进,以实际烟火现场图片为待检对象,完成烟火识别过程计算.结果表明,与基础YOLOv3对比,本研究提出的改进YOLOv3算法平均精度提高5.5%,该方法有助于提升智能烟火预警、人员救助和险情跟踪作业水平,最终提升事故灾害应急和管理能力.
Compared with conventional fire and smoke recognition methods,convolutional neural network(CNN)based algorithms are able to provide multiple merits including high accuracy and efficiency.In this work,a novel fire recognition method based on improved You Only Look Once Version 3(YOLOv3)algorithm has been proposed.Using Gaussian parameters,we designed a loss function for establishing the bounding box model of YOLOv3 network architecture which can also predict positioning uncertainty to reduce the false positive.Then,in order to make full utilization of the local feature information of targeted images and improve the network framework,we employed actual fire scene pictures to complete the fire identification.Thereafter,we compared the improved YOLOv3 with the original one using predefined images.Testing results indicate that the average accuracy has been improved by 5.5%using the presented algorithm.The study holds potential to increase the ability of fire alerting,rescue assistance and emergency tracking which may eventually enhance the response capability for containing accidents and disasters.
作者
陈金鹏
孙浩
东辉
范龙翔
李晨
姚立纲
CHEN Jinpeng;SUN Hao;DONG Hui;FAN Longxiang;LI Chen;YAO Ligang(College of Mechanical Engineering and Automation,Fuzhou University,Fuzhou,Fujian 350108,China;Fujian Provincial Collaborative Innovation Center,Fuzhou,Fujian 350001,China;Institute of Intelligent Manufacturing and Simulation,Fuzhou University,Fuzhou,Fujian 350108,China;Research Institute of Engineering and Technology,Harbin Institute of Technology,Quanzhou,Fujian 362011,China)
出处
《福州大学学报（自然科学版）》
CAS
北大核心
2021年第3期309-315,共7页
Journal of Fuzhou University(Natural Science Edition)
基金
国家自然科学基金资助项目(51605092)
江苏省先进机器人技术重点实验室基金资助项目(JAR202003)
贵州航天智慧农业有限公司项目(00201922)。
