Abstract
In recent years, UAVs have been extensively employed for reconnaissance missions. Our focus is prioritizing reconnaissance of high-priority targets while minimizing the flight duration when UAV power is constrained. We introduce a framework called DQN-ALNS, which integrates Deep Q Network (DQN) into the Adaptive Large Neighborhood Search (ALNS) metaheuristic algorithm to optimize the process through the current solution’s search state. Specifically, the agent is utilized to select the destroy-repair operators to update a new solution, thereby iteratively optimizing the UAV reconnaissance routes. Experimental results reveal that DQN-ALNS achieves superior solutions and faster convergence than other comparison algorithms. The algorithm leverages the exploratory potential of the current solution and demonstrates robust stability. The final sensitivity analysis showcases that reconnaissance missions with high priority are better accomplished when the UAV power is moderate and the target priority is concentrated at smaller values.