Observation scheduling problem for agile earth observation satellites (OSPFAS) plays a critical role in management of agile earth observation satellites (AEOSs). Active imaging enriches the extension of OSPFAS, we call the novel problem as observation scheduling problem for AEOS with variable image duration (OSWVID). A cumulative image quality and a detailed energy consumption is proposed to build OSWVID as a bi-objective optimization model. Three multi-objective memetic algorithms, PD+NSGA-II, LA+NSGA-II and ALNS+NSGA-II, are then designed to solve OSWVID. Considering the heuristic knowledge summarized in our previous research, several operators are designed for improving these three algorithms respectively. Based on existing instances, we analyze the critical parameters optimization, operators evolution, and efficiency of these three algorithms according to extensive simulation experiments.