Abstract: In recent years cellular access UAVs have been widely used in addressing the limitations of traditional UAV models. However, the related research has not yet resulted in an effective path planning method for urban multi-building complexes and sudden obstacle environments. In this paper, we propose a new UAV multi-level path planning framework, which consists of a connectivity-aware global planning layer and a collision-free local planning layer. The global planning takes the ground base station signal coverage as the minimum planning unit and solves the optimal macro path based on the graph theory algorithm to improve the solution efficiency, while the local planning focuses on flight obstacle avoidance. An improved DQN algorithm is used to solve the optimal collision-free flight path. Simulation experiments were carried out. The results show that the proposed method can effectively reduce the flight distance by 16.02% and improve the solution efficiency by 76.6% compared with the traditional method. The method can meet the UAV pathfinding requirements in complex urban environments. Through the above hierarchical planning, the framework provides an effective solution for UAV path planning in complex urban environments.