Mie lidar is widely used in the spatial distribution detection of atmospheric aerosol and cloud, but there are some difficulties in cloud detection from return signal automatically and accurately. The differential zero-crossing method is one of the dominant methods, but it has a disadvantage of big uncertainty when the signal-to-noise ratio is low. The differential zero-crossing method is improved based on the feature of the return signal of LiDAR. Namely the former and latter cloud information and signal intensity are referred to in the improved method, so obvious errors are efficiently avoided. Finally, the cloud extinction coefficient and optical thickness are retrieved based on the cloud detection result. The variations feature of the optical properties is represented objective. Further more, precise and automatic lidar cloud detection and coefficient retrieval are achieved preliminarily.