Satellite observation scheduling plays a significant role in improving the efficiency of satellite observation systems. This paper proposes an ensemble of heuristic and exact algorithm based on a divide-and-conquer framework (EHE-DCF) consisting of two iterative phases: task allocation among multiple orbits and task scheduling on a single orbit. In the task allocation phase, we propose a heuristic task allocation algorithm. In detail, we calculate the allocation probability between the task and orbit, and the task is allocated to the orbit with a probabilistic selection rule. During the scheduling phase, we construct a task scheduling model for each single orbit, and use exact method (e.g., CPLEX) to solve this model. The task allocation and task scheduling phases are performed iteratively until the algorithm termination conditions are met. To validate the performance of EHE-DCF, we compare it with the method that directly utilizes CPLEX to solve the scheduling problem without task allocation, as well as three heuristic algorithms, including greedy algorithm and two simulated annealing algorithms considering the task allocation phase. Experimental results show that EHE-DCF is more efficient for the satellite observation scheduling problem with large-scale tasks. Moreover, the scheduling profits and the number of scheduled tasks are higher and more stable than those of the comparison algorithms.