Trajectory planning for reentry vehicles faces challenges due to nonlinear dynamics, terminal constraints, and sensitivity to initial guesses in traditional sequential convex programming (SCP) methods. This study proposes a hybrid framework integrating a deep neural network (DNN) with SCP to enhance the performance. By leveraging DNN-generated data-driven initializations, the method reduces dependency on manual initial guesses and increases convergence stability. Meanwhile, numerical simulations demonstrate significant improvements in computational efficiency, convergence speed, and mission success rate compared to conventional SCP. Additionally, the optimized trajectories exhibit superior solution quality, with an increase in maximum range and reduced control oscillations. These advancements highlight the framework’s potential for real-time, reliable reentry trajectory planning in complex aerospace applications.