Existing key extraction methods fail to work in dynamic unmanned aerial vehicles (UAVs) scenarios. This paper proposes a new key extraction framework, where a clock adjustment algorithm and a Canonical Correlation Analysis (CCA)-based sliding window smoothing method are employed to mitigate noise and enhance the reliability of the RSSI data. The Level Crossing Algorithm (LCA) is optimized for the dynamic and high-mobility UAV environment. A grid-search method is adopted to find the optimal parameters. Extensive experiments in indoor and outdoor scenarios, including real-world UAV flights, demonstrate the effectiveness of our method.