基于布里渊光时域分析仪的全分布式光纤传感系统中，光纤沿途的探测信号含有噪声导致被测量的温度或应变信息难以识别，光谱拟合的精确度对传感信息的识别非常重要。在传感系统低信噪比的情况下，提出了一种提取高精度布里渊散射谱特征的拟合方法，利用小波去噪结合莱文伯-马奈特（LM）算法调节权值后向传输(BP)网络对布里渊散射谱进行特征提取。克服了传统BP神经网络易陷入局部极值的缺点，保证求解的精度。数值仿真表明，该方法适合不同权重比、不同线宽和低信噪比以及大测量范围的情况进行光谱拟合，并且在信噪比为10 dB的情况下得到拟合度均超过0.96。实验结果表明，该方法适用于多种泵浦功率情况下的布里渊散射谱的特征提取，优于传统BP神经网络算法且具有较高的拟合精度。

In distributed fiber optic sensing system based on stimulated Brillouin scattering optical time domain analysis,the information in temperature or strain measurements is difficult to identify because of the noise mixed into the probe signal.The accuracy of spectral fitting is very important for the identification of sensor information.For the case of the low SNR sensing system, this paper proposes a fitting method to extract high accuracy Brillouin scattering spectral features, which uses wavelet analysis combining the BP (back propagation) network of Levenberg-Marquardt(LM) algorithm to adjust the connection weights.This method overcomes the shortcoming of BP neural network’s easy falling into local minima.Meanwhile, the method ensures the precision of values. Numerical simulations show that the method is suitable for different weight ratios, different spectral line widths, low signal to noise ratio and spectral fitting in large scope. With the SNR of 10 dB,the degrees of fitting obtained are more than 0.96. In addition, experimental results demonstrate that this method is suitable for the extraction of the feature of Brillouin scattering spectrum under the circumstances of multiple pump power extraction, and it has higher precision than traditional BP neural network algorithm.