Abstract: The 3D pose estimation of space object is a significant aspect of space situational awareness.We proposed a method to estimate the 3D pose of satellite as well as its sub-components based on as tronomical observed images.Firstly，a multi-viewpoints pose model database was established by use of equal-area subdivision grid. Secondly，the image restoration and filtering were operated on the observed image with the use of maximum likelihood algorithm based on mixed noises，and then the segment and feature abstraction were performed. Thirdly，a scale-invariant feature indexing method was proposed to encode the geometric features of satellite. Finally，the minimum distance match was used to estimate the 3D pose of space object among the pose database. The experiment result verifies the validity and correction of proposed method.