Aiming at the detection of celestial targets under strong skylight background using sparse domain, the over-complete dictionary of skylight background and target signal is established respectively to distinguish the skylight background area and target area, which based on the difference between skylight background and target signal in their morphological components. The signal energy is accumulated while in suppressing the background, and the signal to noise ratio is also improved. Simulation and experimental results show that the target can be well subtracted, while the centroid deviation, the root mean square and peak-valley values of the centroid are small compared with the subtracting threshold method.