The intensity of the lidar echo signal decays with the square of the distance. When the detection distance is large, the signal will be submerged in the strong noise. Therefore, it is crucial to extract valid signal efficiently from strong background noise. In this paper, the lidar echo signal is decomposed through the empirical mode decomposition. According to the correlation between the intrinsic mode function and the echo signal, the effect of lidar echo signal denoising is improved effectively by combining with soft threshold and roughness penalty techniques. The experimental results of this method show that the output signaltonoise ratio is 16.67 dB and the root mean square error is 1.49×10-11 when 5 dB Gaussian white noise is added. Compared with other denoising techniques, this method achieves the higher signaltonoise ratio and the lower root mean square error, thereby the effectiveness of this method is proved.