Intelligent transportation systems (ITS) rely on federated learning (FL) to enable distributed collaborative sensing while preserving privacy. However, the deployment of FL in heterogeneous vehicle-to-everything (V2X) networks is severely bottlenecked by limited communication resources and diverse channel conditions. Traditional digital orthogonal multiple access (OMA) suffers from high latency that scales linearly with the number of users. Emerging digital over-the-air computation (AirComp), though solving the analog compatibility issue, still faces critical limitations in heterogeneous channels, specifically power mismatch and coupled aggregation error. To address these challenges, we propose a hybrid digital transmission FL (HDT-FL) framework to operate within a unified digital architecture and feature two core mechanisms: (1) A three-region dynamic user partitioning algorithm based on “power feasibility and power margin” to adaptively classify users into digital OMA and digital AirComp groups for high-precision transmission and high-efficiency aggregation respectively; (2) An MMSE-based layered weighted aggregation scheme to optimally fuse heterogeneous digital data. Numerical results in a homogeneous channel environment demonstrate that this hybrid digital scheme significantly reduces global aggregation error while maintaining low latency, validating the effectiveness of the proposed hybrid digital transmission paradigm.