The accuracy of the integrated navigation is significantly affected by calibration, with the system integrating a strapdown inertial navigation system (SINS) and a Doppler velocity log (DVL). Many current calibration methods depend on a DVL error model which is established in advance or require the underwater vehicles to drive a specific trajectory, both of which have limitations. A novel underwater PSO-SVR calibration method for SINS/DVL is proposed. This approach utilizes support vector regression (SVR) for constructing the model, while an improved particle swarm optimization (PSO) technique is used to optimize the parameters. Subsequently, these components are used to develop a predictor that calibrates the velocity of the DVL. The training data is collected by the integrated navigation system to construct the training target output. Specifically, when the underwater vehicles dive into underwater without GPS signals, the output of the PSO-SVR model is utilized as the calibration for constructing the SINS/DVL system. Following this, the trained model is used to predict the velocity of DVL. The results of simulation demonstrate that the proposed algorithm surpasses other conventional calibration algorithms in accuracy.