Multi-UAV reconnaissance mission planning problems (MURMPP) refer to using a limited number of UAVs to obtain critical enemy intelligence and thus control the initiative in warfare while satisfying constraints such as voyage and sensor performance. Mission assignment and path planning are the two core components of the MURMPP. Due to the difficulty and high coupling of the problem, existing stochastic intelligent optimization algorithms or heuristics generally perform poorly on this problem, especially when the problem size gradually increases. Therefore, this paper proposes a hierarchical mission planning framework based on problem decomposition to improve problem-solving efficiency: enhanced consensus based bundle algorithm-deep reinforcement learning (ECBBA-DRL). Specifically, this study first decomposes the MURMPP into task assignment and path planning subtasks, which are solved by ECBBA and DRL, respectively. For task allocation, different types of ECBBA are used to deal with task allocation problems in different environments; for the no-fly zone that exists in the flight environment, multi-agent deep deterministic policy gradient (MADDPG) is used to allow the agent body to fully explore the environment and complete the reconnaissance task while achieving the obstacle avoidance function. The article concludes with multiple simulations for different scenarios to verify the effectiveness and generalization capability of the proposed ECBBA-DRL framework.