基于BIMFire-YOLO的消防监控图像识别方法  

Image recognition method of fire monitoring based on BIMFire⁃YOLO

作　　者：侯戬炜 吴建新 刘超 HOU Jianwei;WU Jianxin;LIU Chao(Beijing Tongren Hospital Affiliated to Capital Medical University,Beijing 100730,China)

机构地区：[1]首都医科大学附属北京同仁医院,北京100730

出　　处：《电子设计工程》2024年第11期105-109,共5页Electronic Design Engineering

摘　　要：在传统条件下,通常存在着火灾现场情况多变,建筑物内构造复杂等问题,使得消防监控系统难以通过现有方法精准检测识别火灾第一现场和所处室内环境。为了优化检测算法在复杂场景下的应用效果,文中提出BIMFire-YOLO检测方法,在以往YOLOv3主干网络中增加了新的卷积方式,在扩大了卷积感受野范围的同时,也保证了输出特征的不变性,对小火焰和烟雾等目标的检测能力有所提升。通过引入新的损失函数,降低了火灾发生的目标漏检率,融合BIM技术不仅提升了定位精度也实现了自动识别视频和图像中失火的场景。该算法不仅提高了检测效率,也实现了火灾的信息共享与管理的统一,在实际医院场景下的综合识别精度达到了45.6%,检测速度和精度均优于其他同类方法。Under the traditional conditions,there are many problems such as the changeable fire scene and the complex structure in the building,which makes it difficult for the fire monitoring system to accurately detect and identify the first fire scene and the indoor environment through the existing methods.In order to optimize the application technology of the detection algorithm in complex scenes,the BIMFire⁃YOLO detection method is proposed in this paper.A new convolution method is added to the previous YOLOv3 backbone network,which expands the range of convolution receptive field,ensures the invariance of output features,and improves the detection ability of small flames,smoke and other targets.By introducing a new loss function,the missed detection rate of fire targets is reduced.The fusion of BIM technology not only improves the positioning accuracy,but also realizes automatic recognition of fire scenes in videos and images.This algorithm not only improves the detection efficiency,but also realizes the unification of fire information sharing and management.The comprehensive recognition accuracy in the actual scene reaches 45.6%,and the detection speed and accuracy are superior to other similar methods.

关 键 词：BIM YOLOv3 EIOU 火灾图像识别 火情定位 

分 类 号：TN929.5[电子电信—通信与信息系统]