As air-to-ground links tend to exhibit a high probability of being line-of-sight (LoS), unmanned aerial vehicles (UAVs) are widely used to improve the performance of wireless communications. The design of the UAV flight path plays a pivotal role in determining the effectiveness of UAV communication systems. However, the air-to-ground links with a high probability of LoS introduce a heightened risk of eavesdropping, posing a significant security challenge. In this work, we investigate the problem of ensuring secure data acquisition for quadrotor UAV-based communication systems in the presence of multiple location-uncertain terrestrial eavesdroppers. The bandwidth allocation and the three-dimensional trajectory of the UAV are jointly designed to maximize the system’s overall fair secrecy rate. This design also considers the UAV’s energy consumption during flight and aims to ensure fairness among users. Solving this problem poses a challenge since it is a non-convex and involves multiple variables, making it difficult to address using conventional optimization methods. Therefore, a deep reinforcement learning algorithm is developed based on the deep deterministic policy gradient algorithm. Simulation results are given to verify the effectiveness of the proposed algorithm in improving the security of aerial Internet of Things systems.