摘要
古建筑以木结构为主体,导致人为造成的古建筑火灾频发。古建筑中为祭拜、照明或营造氛围一般都存在大量和长时间的点火、焚烧等情况,对建筑内设置的常规点型感烟、吸气式、线型光束和图像型等火灾探测器产生严重的干扰和影响,存在误报率高的问题。因此,本文针对大型古建筑内部大空间火灾预警探测技术难题,提出一种新的基于人工智能图像处理技术的古建筑受控火干扰辨识方法,通过设计的火灾智能检测算法、受控火智能辨识与确认算法,实现了典型场景下真实火灾与受控火的“行为”图像特征智能辨识。试验结果表明,本方法可在4 s内分别对真实火灾和受控火做出预警提示,有效解决大型古建筑火灾防控技术难题。
The use of wooden structures as the main structure in ancient buildings has led to frequent human caused fires.Due to the large and long-term ignition,burning,and other situations in ancient buildings for worship,lighting,or creating an atmosphere,it seriously interferes and affects the conventional point type smoke detectors,suction type,linear light beams,and image type fire detectors installed inside the building,resulting in a high false alarm rate.Therefore,this article proposes a new method for identifying controlled fire interference in ancient buildings based on artificial intelligence image processing technology,aiming at the technical difficulties of fire early warning and detection in large spaces inside large ancient buildings.Through the design of fire intelligent detection algorithms and controlled fire intelligent identification and confirmation algorithms,the intelligent identification of"behavior"image features of real fires and controlled fires in typical scenarios is achieved.The experimental results show that this method can provide warning prompts for both real and controlled fires within 4 s,effectively solving the technical difficulties of fire prevention and control in large ancient buildings.
作者
张曦
李晓旭
李泊宁
于春雨
Zhang Xi;Li Xiaoxu;Li Boning;Yu Chun yu(Shenyang Fire Science and Technology Research Institute of MEM,Liaoning Shenyang 110034,China;Liaoning Key Laboratory of Fire Prevention Technology,Liaoning Shenyang 110034,China;National Engineering Research Center of Fire and Emergency Rescue,Liaoning Shenyang 110034,China)
出处
《消防科学与技术》
CAS
北大核心
2024年第11期1528-1532,共5页
Fire Science and Technology
基金
应急管理部消防救援局科研计划项目(2019XFCX44)。
关键词
火灾检测
古建筑
受控火
深度学习
fire detection
ancient architecture
controlled fire
deep learning
