Among images observed by systems in starry sky, because stars are far away from the systems, only several pixels in the images and much noise, lots of dim targets with lower signal-to-noise ratio are submerged. So it should be detected effectively during pre-processing so as to reduce false alarm rate of identification and tracking of subsequent targets. Firstly, noise models of the images are analyzed and image background parameters are got by the least squares fitting method. And then through secondary test method, targets are tested coarsely for the first time and pipeline filtering method is used for identification for the second time. So noise is filtered and the target is got. Finally, dim targets in original images are enhanced by energy accumulated method. Simulation results show that the algorithm is effective.