The echo of the all waveform lidar contains the vertical structure information of the ground object. The traditional all waveform data processing methods rely too much on the initial parameters when extracting these information, resulting in the low availability and accuracy of the data in the terrain complex area. To solve this problem, a waveform decomposition method based on genetic algorithm was proposed. The improved algorithm did not need to provide accurate initial parameters, and used probabilistic transfer rules instead of deterministic rules, which had the characteristics of global optimization. The experiment was carried out with the full waveform data of GF-7 satellite laser. The results show that the correlation coefficient between the echo waveform fitted by the improved waveform processing method and the preprocessed waveform is more than 99%. The inversion of the maximum tree height in the forest area was compared with the forest canopy height parameter in the ATL08 data of ICESat-2. The correlation coefficient is 0.85 and the mean square error is 1.1 m, which shows that the method can extract the feature information of complex waveform more accurately.