Abstract: Aiming at the problem of combination detection of targets with various shapes in space based infrared surveillance system, an adaptive target detection method based on extended Markov random field (E-MRF) was proposed. First, the target characteristics in space based infrared surveillance system were analyzed and the extended MRF neighborhood system was established by using the typical target shapes as templates. Then the new MRF potential function was constructed. By using the potential differences between the infrared background and target, the detection problem of targets with various shapes under complex background was formulated as a discrimination problem of MRF potential function and finally the problem of mixed shape targets detection under MRF theory framework was efficiently solved. Simulation and experimental results show that the proposed method can adaptively adjust the operator according to the target shapes and perfectly detect targets with different shapes under various image SCRs. So the proposed method has a stabile performance.