The detection of contraband in complex scenarios such as customs cargo yards and criminal investigation identification is critically important, as it is closely related to national security, social stability, and the safety of people's lives and property. The utilization of unmanned systems for non-contact detection of contraband has emerged as a prominent research frontier and a cutting-edge focus area. In reconnaissance missions involving unmanned systems for contraband detection, the optimization of shortest path planning is a pivotal factor for enhancing task efficiency and success rates. This paper proposes a novel path planning method based on a 3D Digital Twin and Reinforcement Learning DQN algorithm (3D-DQN), designed for random operational scenarios to achieve optimal reconnaissance path planning. By utilizing 3D Digital Twin technology, reconnaissance scenarios with randomly distributed obstacles and targets are simulated, enabling precise multi-functional scenario modeling. A comparative performance analysis of the Reinforcement Learning DQN algorithm and the A* automatic routing algorithm in random scenarios demonstrates that the 3D-DQN method offers better environmental adaptability. It exhibits higher efficiency and accuracy in various randomly generated scenarios, enabling the rapid identification of the shortest path and enhancing the reliability of task completion. This approach provides a new method for shortest path planning in non-contact reconnaissance missions using unmanned systems.