Route Planning Based on Deep Reinforcement Learning to Minimize Energy Consumption in UAV Photogrammetry

In photogrammetry, limited battery energy and storage capacity require unmanned aerial vehicles (UAVs) to travel frequently to and from depots. Therefore, route planning is necessary to ensure safety and efficiency. In this paper, the traditional iterative search algorithm is improved through neural network-assisted deep reinforcement learning with an attention layer and the strategy of separating improvement and reconstruction operators, and applied to the vehicle route problem with energy and capacity constraints. Simulation and practice show that energy consumption of the route generated by the improved algorithm is reduced by 10.63%, and the calculation time is only 8.92% that of the optimal traditional algorithm. Therefore, deep reinforcement learning can be applied in the simulation and practice of UAV photogrammetry, and it has been verified to be reliable and efficient, which lays the foundation for minimizing the energy consumption of photogrammetry with multi-aircraft cooperation.