Abstract:Magnetic fields generated by underwater magnetic targets becomes an important source of non-acoustic detection signals for underwater targets. In this paper, the principles of static magnetic field modeling of underwater magnetic targets are investigated. The rotating ellipsoid model with uniform magnetization and the magnetic dipole model are implemented, and the adaptability of the two models to the simulation of the target magnetic anomaly signal is analyzed, which shows that the rotating ellipsoid model can also perform well when the sensor is close to the target compared to the magnetic dipole model. Therefore, using the rotating ellipsoid model, we firstly construct the close-range magnetic anomaly signal dataset of the target in different motion speed states by simulation, and then carry out the research on the underwater magnetic target motion speed recognition method based on the deep learning method. Finally, the recognition effect achieved good error accuracy, which was 0.17m/s on the validation set and 0.74m/s on the test set, and the work of this paper laid an important foundation for the subsequent target heading and other state recognition.