<p>In the current NISQ era, the client lacks sufficient quantum capabilities to construct complex quantum neural networks locally. Although the proposed quantum federated learning methods deploy training models on servers with powerful quantum capabilities, challenges remain, such as preparing highly entangled brickwork states and preventing unavoidable leakage of model parameters to the server. Therefore, this paper proposes a quantum federated learning method utilizing measurement-based quantum computation. Firstly, a quantum measurement model based on five-qubit entangled states is constructed to enable the deployment of an encrypted quantum neural network on the server side, with only measurement operations performed on the client side. This ensures that the client achieves the desired rotation gates and angles through measurement operations while performing a quantum one-time pad to encrypt quantum states. This prevents an untrusted server from extracting information about private data, model parameters, and model outputs. Secondly, to ensure the security of the client’s gradient information, a quantum measurement model utilizing three-qubit entangled states is developed to implement a secure aggregation method. Finally, a security analysis demonstrates that the proposed scheme protects the client's private data, model parameters, and model output. Furthermore, we conduct a binary classification experiment on the MNIST dataset using the measurement-based quantum computation framework provided by Paddle quantum, validating the feasibility of the proposed method.</p>