With the increasing number of Geosynchronous Orbit (GEO) satellites and the complexity of their operational environments, evasion maneuver strategies have become critically important for the space safety, particularly against satellites with active observation capabilities . This paper addresses the satellite evasion problem to avoid close observations by non-cooperative targets. An evasion maneuver optimization model is developed incorporating space-based optical visibility constraints, using a differential evolution algorithm. The model includes dynamics equations for two-impulse maneuvers, and considers multiple factors such as observation distance, earth′s shadow, and solar illumination angle constraints. Finally, numerical simulations indicate that the proposed method can achieve evasion maneuvers which can significantly reduce the visible arc segments of the observed satellite, enhancing its stealth capabilities and survivability in GEO environments. The research is of great importance in improving the safety and operational effectiveness of GEO satellites.