With the advancement of intelligent technologies, particularly deep reinforcement learning, Unmanned Aerial Vehicles (UAVs) are being widely deployed across various scenarios. However, current remote-controlled UAV methods and existing intelligent decision-making algorithms struggle to effectively respond to dynamic and partially observable situations. To tackle this issue, we propose the Memory-augmented Deep Deterministic Policy Gradient (MemDDPG) algorithm. The MemDDPG algorithm leverages a Long Short-Term Memory (LSTM) network to use past state information for current decision-making, enabling UAVs to perform accurately in dynamic scenarios. By incorporating memory-based mechanisms, MemDDPG addresses the limitations of conventional methods, enhancing UAVs’ ability to manage complex environments. This algorithm improves decision-making and ensures UAVs operate more autonomously and efficiently in unpredictable situations. Experiments demonstrate the effectiveness and feasibility of the MemDDPG algorithm in enhancing the intelligent decision-making capabilities of UAVs, showing significant improvements in adaptability and decision-making accuracy in dynamic situations.