提出了一种基于神经网络的火灾烟雾识别方法 ,以波长为 6 70nm、10 6 0nm、15 5 0nm的三束激光的三对消光系数比作为网络的输入 ,网络的输出为“火灾烟雾”和“非火灾因素” ,从典型火灾烟雾和非火灾因素对多波长激光的衰减实验中选取数据 ,组成 2 6种网络样本模式定义表 ,经 391次仿真训练后 ,输出误差小于 0 .0 0 0 1,并经验证实验表明 ,本方法对火灾烟雾和非火灾因素能进行有效的识别 ,是处理烟雾识别等非结构问题的一种行之有效的方法。 A method of fire smoke recognition based on neural network is proposed. The three extinction coefficient ratios of three laser beams with wavelengths of 670nm, 106nm and 150nm are used as input of the network. The output of the network is “fire smoke” and “Non-fire factors”, from the typical fire smoke and non-fire factors on the multi-wavelength laser attenuation experiment selected data to form 26 kinds of network sample pattern definition table, after 391 simulation training, the output error is less than 0.000 1, and verified the experiment shows that this method can effectively identify the fire smoke and non-fire factors and is an effective method to deal with the non-structural problems such as smoke identification.