基于多特征无量纲向量的图像型管廊火灾探测技术研究  

Study on image-type fire detection technology of pipe gallery based on multi-feature dimensionless vector

作　　者：王成昌[1] 刘华[1] 陈月 李享 WANG Chengchang;LIU Hua;CHEN Yue;LI Xiang(School of Safety Science and Emergency Management,Wuhan University of Technology,Wuhan Hubei 430070,China;School of Environmental Engineering,Wuhan Textile University,Wuhan Hubei 430299,China)

机构地区：[1]武汉理工大学安全科学与应急管理学院,湖北武汉430070 [2]武汉纺织大学环境工程学院,湖北武汉430299

出　　处：《中国安全生产科学技术》2023年第4期14-20,共7页Journal of Safety Science and Technology

基　　金：国家自然科学基金项目(52074202)。

摘　　要：针对管廊火灾检测中背景干扰与小目标检测困难的问题,提出采用无量纲特征参数描述图像型火焰动静态特征的火灾检测方法。利用背景差分法分割火焰目标区域,分析火焰发展特性,提取、计算角二阶矩变化率(A^(*)_(ASM))、对比度变化率(C^(*)_(CON))、自相关度变化率(C^(*)_(COR))、面积变化率(S^(*))等无量纲特征参数;拟合无量纲特征融合向量T^(*)并在SVM模型中检验其可靠性。研究结果表明:通过对不同可燃物在不同空间环境下的燃烧视频随机200帧序列进行检测,发现提出的无量纲特征参数具有“稳定”的变化范围,对火灾火焰与一般火焰的识别率最高可达97.5%,98.0%,受监测空间环境的影响较小。研究结果可为图像型火焰特征提取提供借鉴。Aiming at the problems of background interference and small target detection in the pipe gallery fire detection,a fire detection method using dimensionless feature parameters to describe the dynamic and static characteristics of image-type flame was proposed.The background difference method was used to segment the flame target area,and the flame development characteristics were analyzed.The dimensionless feature parameters were extracted and calculated,such as the angular second moment change rate(A^(*)_(ASM)),contrast change rate(C^(*)_(CON)),autocorrelation change rate(C^(*)_(COR))and area change rate(S^(*)),then the dimensionless feature fusion vector T^(*)was fitted,and its reliability was tested in the SVM model.The results showed that through detecting the random 200 frame sequences of combustion video of different combustibles in different spatial environments,it was found that the proposed dimensionless feature parameters had a stable range of variation,and the accuracy of fire detection was as high as 98%,which was less affected by the monitoring spatial environment.The research results can provide reference for the image-type flame feature extraction.

关 键 词：多特征融合 无量纲 背景差分法 管廊火灾 图像型火灾探测 

分 类 号：X932[环境科学与工程—安全科学]