Due to constrained resources and complex terrain, the autonomous navigation of rovers is susceptible to wheel slip and path deviation, which jeopardize mission safety. Current path planning algorithms that account for slip safety often suffer from low prediction accuracy and extended computation times. Consequently, this paper introduces a deep learning-based path planning algorithm designed to enhance terrain classification and slip prediction. Specifically, a deep learning model is employed for precise extraterrestrial terrain classification, followed by the construction of a U-Net slip prediction network to generate a slip probability map. This slip probability map is subsequently integrated into the heuristic function of the A* algorithm, enabling optimal path planning that balances safety and path length. Comparative experiments conducted on a 3D simulation map of Mars demonstrate that the proposed method can effectively plan safe paths for rovers using fewer computational resources, avoid high-slip areas, and improve overall path optimality.