A large amount of noise will be generated while spaceborne photon counting LIDAR receive signals， and the signal-to-noise ratio is lower in complex mountainous land， which greatly affects the accurate extraction of vegetation point cloud signals. This paper proposes a density clustering algorithm based on the mountain slope to solve this problem. By analyzing the density of point cloud data and the terrain characteristics of forest targets， coarse noise removal is performed by using the maximum density center search method， and then the slope angle is calculated based on the point cloud data to optimize density clustering and complete the data fine noise removal. By classifying the extracted forest region signal， fitting the vegetation canopy profile and the surface profile， the results show that the proposed algorithm has high accuracy in the extraction of vegetation photon signal， and the RMSE of the ground and canopy profiles are 0.3588 m and 3.7449 m， respectively， which is more suitable for vegetation remote sensing point cloud data processing.