With the rapid advancement in autonomous driving and vehicle networking technology, the demand for efficient computation offloading in vehicle edge computing (VEC) systems has become increasingly critical. This study introduces a novel VEC task offloading model leveraging Deep Reinforcement Learning (DRL) to optimize computational resource utilization in heterogeneous vehicular networks. The proposed model utilizes idle resources in vehicles to assist in computation offloading, focusing on minimizing system costs, which include total latency and energy consumption. By integrating reinforcement learning (RL) with deep learning (DL), the model enhances convergence efficiency and decision-making accuracy. An improved Q-learning algorithm is developed to address the task offloading and processing challenge, aiming to achieve optimal offloading decisions. Simulation results demonstrate that the improved Q-learning algorithm effectively reduces the total system cost and enhances the quality of service in VEC systems. The model not only offers a practical approach to computation offloading but also opens avenues for further research in AI-driven vehicular technologies.