<p>Federated learning (FL) enables multiple users to collaboratively train a shared model while keeping their raw data locally stored, and it has demonstrated significant potential across various domains. However, existing FL schemes still face critical challenges in terms of privacy preservation, utilization of low-quality data, and verification of cloud-based aggregation results. To address these issues, this paper proposes an efficient and verifiable secure aggregation scheme for FL specifically designed to handle low-quality data. First, secret sharing and double masking are employed to encrypt client-uploaded gradients, preventing cloud servers from accessing plaintext gradient information and thereby protecting user privacy. Second, a truth discovery mechanism is introduced to assess data quality and assign differentiated weights, mitigating the negative impact of low-quality data on model performance. Third, a lightweight verification method using a single mask and the Hadamard product is designed to generate verification tags, enabling detection of potential server misbehavior. Finally, simulation experiments are conducted to demonstrate the feasibility and effectiveness of the proposed scheme.</p>