Abstract: Because the wavelet transform has the characteristic of multi-scale analysis and it can characterize signals local feature <参考文献原文>, this article uses the wavelet decomposition method to study the sensitivity of different time-frequency components of the Magnetic Barkhausen Noise signal with changes in temperature and stress. After using the db5 wavelet with six layers to decompose the MBN signal, we extract the mean and RMS value of each layer decomposition coefficients and discuss the relationship between the relative change of the features and applied temperature and stress. It is found that within the elastic range of the sample, the mean and RMS value of high-frequency coefficients of each layer and low-frequency coefficients both decrease with increasing compressive stress. The mean and RMS value of high-frequency coefficients of each layer decrease, whereas the values of the low-frequency coefficients increase with increasing temperature respectively. This article takes temperature, the mean and RMS value of the original MBN signal and the decomposition coefficients as the input of the neural network and takes the stress as the output of the neural network to build the neural network model. It is shown that using this neural network model to detect stress has higher accuracy than the former BP neural network model in which the wavelet decomposition is not used.