在设计红外火焰探测器的过程中，人工光源常常会引起探测器的误报。为有效区分人工光源与火焰信号，本文首先对采集的 1种人工光源以及 3种火焰信号进行分析，将采集的信号进行小波包 4层分解，得到信号的能量谱。通过分析发现选取第 2,3,4,7频段的能量值能够将这 4种信号有效区分。为进一步验证实验结果，本文将双通道两路信号的第 2,3,4,7频段的 8个能量值作为一组特征向量，与 BP神经网络结合进行模式识别。结果表明，通过这样的方法不仅可以区分火焰和人工光源，同时可以对 3种火焰进行识别，其识别的正确率为 84.1%。因此，基于小波包能量分析的方法提取这 8个能量值作为特征值具有一定的可行性，能有效减少人工光源引起的误报，同时为火焰种类的识别以及以后的灭火自动化提供了新的可能性。

In process of design of infrared flame detector, the recognition rate of which was reduced by some artificial lights. In order to distinguish artificial light and flame, an artificial light and three kinds of flame were analyzed firstly. The collected signals were decomposed by wavelet packet, then the wavelet packet energy spectrum was acquired. The results showed that the four kinds of signal can be distinguished by the energy of 2,3,4 and7 frequency band. In order to verify the experimental results, the eight energy values of the 2,3,4,7 spectrum from two sensors were regarded as a group of feature vector. The mode of recognition was carried out by applying BP neural network and the feature vector. The results showed that the method can distinguish the artificial light and the flame. In addition, the three kinds of flame were also identified. The recognition rate can reach 84.1%. Therefore, the method based on wavelet packet energy analysis has certain feasibility, which can increase the accuracy of flame identification. It also provides a new possibility to put out the fire automatically in the feature.