Lidar, widely used in practical applications, is often affected by background light noise and system shot noise, leading to a decline in signal quality and difficulty in extracting valid information. To address this issue, this paper proposes a dual noise filtering method combining local mean decomposition (LMD) and successive variational mode decomposition (SVMD), which is applied to the field of lidar cloud layer detection. The method employs LMD to decompose the original signal and uses sample entropy to categorize the signal into low-frequency and mixed signals. Subsequently, SVMD is applied to the mixed signals, and mutual information is utilized to further segregate the signals into high-frequency and low-frequency effective signals. Finally, the low-frequency and low-frequency effective signals are reconstructed into a denoised signal. The denoising performance of the proposed method is verified using both simulated and measured data. The experimental results show that for the Bumps signal with an input SNR of -6 dB, the output SNR of the proposed method reaches 8.49 dB, and the root mean square error (RMSE) is as low as 0.67. For the lidar simulated echo signal with an input SNR of 9 dB, the output SNR is 19.344 8 dB and the RMSE is 5.181 7×10-12, which exhibits significant advantages over the methods such as WT, LMD, EMD-WT, SVMD, and EEMD-GWO-SVD. Moreover, in the application of lidar cloud parameter detection, the denoised signal from this method is closest to the original signal in terms of cloud base and top heights, effectively preserving signal details and features, which is beneficial for enhancing the inversion accuracy of lidar data.