Photon-counting LiDAR is very sensitive to ambient interference, target features, and instrument performance, especially for long-distance detection of spaceborne laser altimeter and measurement of complex land-cover types with steep terrain. It is crucial to extract the signal photons on the ground surface from the collected photon point cloud (PPC). An adaptive signal photon detection method is presented in this paper, which combines histogram statistics and boxplot analysis with density-based spatial clustering of applications with noise (DBSCAN), to denoises the PPC data with strong and weak noise obtained by ICESat-2 laser altimeter. First, a coarse denoising with histogram of elevation is conducted on the raw PPC to reduce the calculation amount. Second, a fine denoising based on adaptive DBSCAN is used to extract the signal photons, where the key parameters of elliptic filter kernel are automatically determined according to the topographic data situation. We compared it with other methods, including local distance statistics (LDS), traditional and modified DBSCAN, traditional and modified ordering points to identify cluster structure (OPTICS), and ATL08 data. Some quantitative indicators, such as recall (R), precision (P), and F-score (F), are used to evaluate its performance. The results show that (1) the adaptive DBSCAN has the best performance on preserving the vertical structural characteristics of ground objects. (2) The adaptive DBSCAN in the mean R, P and F of three land covers (i.e., mountain forest, urban and water areas) can get up to the maximum are 0.9852, 0.9675 and 0.9761, respectively; followed by ATL08 data with 0.9773, 0.9412 and 0.9536, modified OPTICS with 0.9684, 0.9460 and 0.9586, and modified DBSCAN with 0.9613, 0.9474 and 0.9544; and then OPTICS with 0.9444, 0.9397 and 0.9378, and the DBSCAN with 0.9444, 0.9355 and 0.9554; the last one is LDS with 0.9382, 0.9261 and 0.9422. The proposed method provides an alternative approach for rapid and accurate processing of PPC on complex terrain.