Abstract: Ultrasonic testing is a new way to detect power equipment. Its echo is often interfered by noise. In order to improve the accuracy of detection, it is necessary to denoise the echo signal and improve the signal quality. In this paper, a denoising algorithm based on the combination of complete ensemble empirical mode decomposition with adaptive noise (CEEM) and fast independent component analysis (FastICA) is proposed. The noisy signal is decomposed into several modal components (IMF) by CEEMDAN algorithm to meet the requirements of blind source separation for signal positive or overdetermined, and then the multi-source signal is constructed by FastICA for IMF, and finally the Hurst index is used. The threshold distinguishes the noise in the multi-dimensional signal, completes the filtering and reconstructs the ultrasonic signal. Through simulation and experiment, this method can remove the noise signal better than other algorithms, retain the original information such as the starting position of the echo, and has higher signal-to-noise ratio, lower mean square error and the shortest running time. It can improve the accuracy of the ultrasonic detection sleeve lead state and has certain application value.