基于BP神经网络的公路隧道视频火焰识别

Flame Identification in Video for Highway Tunnel Based on BP Neural Network

为提高基于视频图像的公路隧道火灾火焰识别率，在对火焰动态特征研究成果之上，利用BP神经网络融合火焰静态特征，对公路隧道视频火焰进行综合识别．火焰动态特征选取作者研究的火焰边缘运动量（AM FE）和火焰区域跳动特征，火焰静态特征选取前人研究的尖角数目、火焰颜色特征和圆形度．将此5种火焰特征作为BP神经网络的输入，达到融合火焰多特征信息并实现火焰综合识别的目的．实验结果表明，火焰识别率稳定在86．2％～96．5％之间，验证了该方法的可靠性．

In order to improve the flame identification rate for highway tunnel based on video ,BP neu‐ral network was employed for integration of static features to identify flame on the basis of dynamic fea‐tures ,such as the amount of movement of flame edge（AMFE） and flame area beating feature ,from our ow n previous studies .Flame static features such as flame circular degree ,flame color feature ,and flame circular degree were adopted from the previous studies of others .And the five features were used as the in‐put of BP neural network to identify flame .The experiment results showed that the rage of flame identifica‐tion rate was 86 .2% —96 .5% ,which verified the reliability of the method .