Federated learning allows multiple participants to collaboratively train models while keeping their data private. However, many studies have shown that an attacker may recover some private information from local training results, and the server responsible for aggregating these local gradients may also return false aggregated results. In addition, in the FL training process, the computing resources of clients vary, and the relatively insufficient computing resources of some clients can affect the efficiency of data aggregation. In this paper, we propose an efficient federated learning scheme with privacy protection and verifiability. Specifically, by adding a blind factor to the local training results, the privacy of uploaded parameters is protected. Secondly, an effective verification mechanism is designed to verify the correctness of the data aggregation results. Thirdly, a drop-out mechanism is introduced to prevent clients with insufficient computing resources from affecting the aggregation efficiency. Finally, a security analysis and performance analysis of the protocol are conducted, proving that the protocol is secure and exhibits high computational and communication efficiency.