UAV Track Planning Algorithm Based on Graph Attention Network and Deep Q Network.

To solve the problem that the deviation of the Unmanned Aerial Vehicle (UAV) flight status data collected by the equipment during flight leads to the failure of the mission, this paper proposes a UAV track planning algorithm based on Graph Attention Network and Deep Q Network (DQN). Firstly, we use the camera to collect images and apply pre-trained ResNet to extract image features. Secondly, we adopt the Graph Attention Network to establish the connection between the sensor-measured flight state information and the actual flight state information. Thirdly, we build the optimization model of flight state. Moreover, based on the Deep Reinforcement Learning (DRL) theory, the DQN-based UAV track planning system is trained. Finally, the system combined the optimized flight state to complete the optimal flight action output to realize the track planning. Simulation results show that, compared with the original algorithm which is under the same flight conditions as the proposed algorithm, the velocity deviation rate of the proposed algorithm is improved by 46.79%, which can plan a high-quality track and has good engineering application value.