The Vehicle-to-Everything (V2X) communication system allows vehicles to share data instantly with other vehicles, infrastructure and pedestrians, which is necessary for autonomous driving and intelligent transportation systems. The traditional resource allocation methods, TDMA and FDMA, together with heuristic algorithms, experience difficulties in dynamic vehicular networks because of interference problems and mobility and latency restrictions. The proposed framework combines Reinforcement Learning (RL) with Deep Neural Networks (DNNs) to create an adaptive system which optimizes resource allocation and channel selection in V2X networks. The RL agent performs realtime optimization of bandwidth and power distribution through the DNN system, which uses SINR, traffic load and interference level data to predict channel quality. The framework achieves a 40.3% improvement in throughput, 52.5% latency reduction and 33% fairness gain when evaluated through simulations of urban V2X environments using synthetic data. The proposed system provides a scalable real-time solution for next-generation V2X systems, which can be integrated into 5G/6G-enabled edge networks.