For the exploration of unknown environments, such as search and rescue, chase and escape scenarios, the UAV needs to explore (perceive) the environment while completing current trajectory planning (action selection).Aiming at the above scenarios, in order to achieve efficient environment exploration, an improved Deep Double Q Network (DDQN) exploration trajectory planning method based on Long Short-Term Memory (LSTM) network is proposed.A simulation map is built, the environmental information in the UAVs field of view is taken as input, the LSTM network is introduced, and the choice of action direction is outputted.The priority of exploration experience samples is set to improve training efficiency.Flight dynamics constraints are added, and reasonable state, action space and one-step reward function are designed.Using the proposed algorithm, the UAV can autonomously plan a collision-free track with a wide range of environmental exploration.The simulation results show that the proposed algorithm is better than the traditional DDQN algorithm in the exploration area ratio and the average reward of one-step exploration in unknown environments.