This paper addresses the limitations of traditional brain-controlled UAV systems in real-time performance, control stability, and adaptability to complex environments by designing and implementing a closed-loop brain-controlled UAV system based on a Transformer-based SSVEP decoding method. The system adopts a four-layer architecture consisting of electroencephalogram(EEG） acquisition, signal processing, command mapping, and UAV control. Steady-state visual evoked potentials (SSVEPs) are used as the control paradigm to achieve online control of eight types of flight actions: takeoff, landing, forward, backward, leftward, rightward, ascent, and descent. In the decoding layer, a Transformer-based SSVEP decoder is employed to model the spatiotemporal features of multi-channel EEG signals. In addition, a dual-verification mechanism combining a confidence threshold and a consecutive consistency constraint is introduced to reduce false triggers and improve the stability of online control. To validate the applicability of the proposed approach, offline comparative experiments were first conducted on the two public datasets, BETA and Benchmark, and comparisons were made with methods such as CCA, FBCCA, and EEGNet. Subsequently, online brain-controlled flight experiments were carried out with 12 real participants, and the system performance was evaluated in terms of recognition accuracy, response time, information transfer rate, and false trigger rate. The experimental results demonstrate that the proposed system can stably accomplish online closed-loop control of eight flight commands in a real-time scenario. The Transformer-based decoding method shows good applicability in both offline and online experiments, and the designed dual-verification mechanism effectively improves control stability and reduces the risk of false triggers. This study provides experimental evidence for the system integration and engineering application of Transformer-based SSVEP decoding methods in real-time brain-controlled UAV scenarios.