Abstract: Transient electromagnetic signals are easily affected by electromagnetic interference, which reduces the signal-to-noise ratio of data and distorts the attenuation curve. However, a single filtering method has some disadvantages, such as easy loss of geological information and excessive smoothness, so it is difficult to obtain high-precision imaging results. Therefore, a hybrid algorithm based on wavelet packet transform -BP neural network is proposed, which makes use of wavelet packet transform's ability to extract energy features, decompose and reconstruct signals and BP neural network's learning and feedback ability to filter transient electromagnetic signals. Fourier transform is used to obtain the frequency domain characteristics of the transient electromagnetic signal, and the difference between the interfered signal and the undisturbed signal is obtained by comparing them. Using three-layer wavelet packet decomposition to obtain the energy ratio of the third-layer nodes, extracting the characteristics of the reconstructed signal, and preliminarily decomposing and reconstructing the transient electromagnetic signal; The trained neural network model is called to train the features of the reconstructed signal, and the final filtered transient electromagnetic signal is obtained. The theoretical and measured data research shows that the hybrid algorithm is more practical and accurate than the commonly used S-G filtering and mean filtering. It keeps the real geological information while filtering, enhances the accuracy of data interpretation, and has a good application effect, which provides a strong technical support for data processing.