The gas explosion in residential building has always been a highly concerned problem.Explosions in homogeneous mixtures have been extensively studied.However,mixtures are often inhomogeneous in the practical scenarios...The gas explosion in residential building has always been a highly concerned problem.Explosions in homogeneous mixtures have been extensively studied.However,mixtures are often inhomogeneous in the practical scenarios due to the differences in the densities of methane and air.In order to investigate the effects of gas explosions in inhomogeneous mixtures,experimental studies involving gas leakage and explosion are conducted in a full-scale residential building to reproduce the process of gas explosion.By fitting the dimensionless buoyancy as a function of dimensionless height and dimensionless time,a distribution model of gas in large-scale spaces is established,and the mechanism of inhomogeneous distribution of methane is also be revealed.Furthermore,the stratified reconstruction method(SRM)is introduced for efficiently setting up inhomogeneous concentration fields in FLACS.The simulation results highlight that for the internal overpressure,the distribution of methane has no effect on the first overpressure peak(ΔP1),while it significantly influences the subsequent overpressure peak(ΔP2),and the maximum difference between the overpressure of homogeneous and inhomogeneous distribution is174.3%.Moreover,the initial concentration distribution also has a certain impact on the external overpressure.展开更多
针对传统基于梯度方向直方图特征检测算法对解决目标模型单一、发生形变、存在遮挡及目标受干扰下定位困难的问题,提出一种基于HOG特征混合模型结合隐SVM的感兴趣目标检测算法。首先利用用训练图像的HOG特征金字塔表示得到包含感兴趣目...针对传统基于梯度方向直方图特征检测算法对解决目标模型单一、发生形变、存在遮挡及目标受干扰下定位困难的问题,提出一种基于HOG特征混合模型结合隐SVM的感兴趣目标检测算法。首先利用用训练图像的HOG特征金字塔表示得到包含感兴趣目标根模型、部件模型和对应可变形部件特征表示,该模型不仅描述目标的整体轮廓,而且能够捕捉到更为精细的目标部件轮廓,在一定程度上提高了检测算法在目标姿态复杂情况下的鲁棒性。然后利用HOG特征混合特征训练部件检测分类器LSVM(Latent Support Vector Machine)。最后通过动态规划和距离转换算法在测试图上扫描出与可变形部件模型相匹配的区域,实现感兴趣目标的检测定位。经过多组实验结果表明,所提出的算法能较好地解决目标在发生较大形变和存在遮挡等复杂姿态下的定位问题。展开更多
基金the financial support from National Natural Science Foundation of China(Grant No.52378488)Postgraduate Research&Practice Innovation Program of Jiangsu Province(Grant No.KYCX22_0222)。
文摘The gas explosion in residential building has always been a highly concerned problem.Explosions in homogeneous mixtures have been extensively studied.However,mixtures are often inhomogeneous in the practical scenarios due to the differences in the densities of methane and air.In order to investigate the effects of gas explosions in inhomogeneous mixtures,experimental studies involving gas leakage and explosion are conducted in a full-scale residential building to reproduce the process of gas explosion.By fitting the dimensionless buoyancy as a function of dimensionless height and dimensionless time,a distribution model of gas in large-scale spaces is established,and the mechanism of inhomogeneous distribution of methane is also be revealed.Furthermore,the stratified reconstruction method(SRM)is introduced for efficiently setting up inhomogeneous concentration fields in FLACS.The simulation results highlight that for the internal overpressure,the distribution of methane has no effect on the first overpressure peak(ΔP1),while it significantly influences the subsequent overpressure peak(ΔP2),and the maximum difference between the overpressure of homogeneous and inhomogeneous distribution is174.3%.Moreover,the initial concentration distribution also has a certain impact on the external overpressure.
文摘针对传统基于梯度方向直方图特征检测算法对解决目标模型单一、发生形变、存在遮挡及目标受干扰下定位困难的问题,提出一种基于HOG特征混合模型结合隐SVM的感兴趣目标检测算法。首先利用用训练图像的HOG特征金字塔表示得到包含感兴趣目标根模型、部件模型和对应可变形部件特征表示,该模型不仅描述目标的整体轮廓,而且能够捕捉到更为精细的目标部件轮廓,在一定程度上提高了检测算法在目标姿态复杂情况下的鲁棒性。然后利用HOG特征混合特征训练部件检测分类器LSVM(Latent Support Vector Machine)。最后通过动态规划和距离转换算法在测试图上扫描出与可变形部件模型相匹配的区域,实现感兴趣目标的检测定位。经过多组实验结果表明,所提出的算法能较好地解决目标在发生较大形变和存在遮挡等复杂姿态下的定位问题。
