Aiming at the defects of the traditional differential zero-crossing method, this paper proposes a two-threshold cloud detection algorithm based on cluster analysis. In this algorithm, the lidar signal is processed with the improved thresholds of peak-to-bottom ratio and background noise, and the cloud information is extracted accurately. In addition, for the layered cloud processing, this paper adopts the ISODATA algorithm for the cluster analysis of differential zeros to achieve accurate layered cloud processing. Experimental results show that compared with the traditional differential zero-crossing method, the proposed algorithm significantly improves the inversion accuracy, effectively eliminates the interference of aerosol signals, extracts cloud information, and has a good detection effect on the clouds with different heights. It is found that on the basis of hierarchical cloud processing, the echo signal characteristics with different types of clouds and different cloud structures can be used to distinguish cloud types such as cumulus, stratus, and cirrus, as well as cloud phase states such as ice cloud, water cloud, and ice-water mixed cloud.