Abstract: ICESat-2 photon-counting lidar data are widely utilized in various fields such as terrain mapping, forest monitoring, and water quality assessment. When the ICESat-2 overfly the sea surface, it can detect faint photon signals reflected back by the sea surface, which can reflect the fluctuation state of the sea surface. However, it also contains a large number of noise photons from equipment noise, atmospheric particle scattering noise, background interference noise, etc. This paper proposes a new method for sea surface signal photon extraction and wave parameters retrieved. First, the original ICESat-2 data is evenly divided into segments along the vertical direction, and the number of photons in each segment is counted. Gaussian fitting and iterative median filtering are performed on it to determine the precise upper and lower bounds of sea surface photons, thereby extracting sea surface signal photons. Support vector regression fitting is further used to extract the instantaneous sea surface profile. On this basis, the spectrum analysis method is used to calculate the significant wave height and preliminary wave peak wavelength. Through phase coherence analysis of synchronized strong and weak beams, the wave direction is estimated, the wave peak wavelength is corrected, and the wave peak period is calculated. Sea surface photon extraction and wave parameters retrieved were performed on 24 orbital ICESat-2 data from the Cape of Good Hope and Samoa waters, and the results were compared and verified with ERA5 reanalysis data. The verification results show that the root mean square error of the significant wave height, wave direction and peak period are 0.12 m, 28.58 degrees and 0.48 seconds, respectively, proving that this method has high accuracy in retrieving wave parameters and is capable of further exploring a wider range of laser wave feature extraction and analysis potential of lidar data.