<p>Federated learning (FL) technology is used to process the data sharing in Internet of Vehicles (IoV). During the development and deployment of federated models, the data resource heterogeneity, limited communication and computation resources in IoV pose challenges to the performance of FL. To address these challenges, one of the recent approaches is to select suitable participants from available vehicles during training. However, there are still many issues in this field that urgently need to be addressed, many works are still based on the assumption that participants voluntarily contribute their data resources. Therefore, this paper proposes a novel participant selection and incentive scheme called FedSI, which includes Blockchain layer, Road Side Unit (RSU) layer and Vehicle layer. In the scheme, FL model can realize an online client selection with high learning accuracy; and adequately motivate the selected vehicles to participate in model training under a stable and long-term cooperative strategy. Performance evaluations in the experiments show the feasibility and effectiveness of FedSI.</p>