Abstract: The traditional positioning methods with magnetic gradient tensor are basically based on single magnetic dipole model，which results in insufficient positioning accuracy. Moreover，these positioning methods are easily affected by environmental noise and measurement accuracy. Especially in the dipole characteristic plane，little disturbance or insufficient observation accuracy will lead to a big deviation of positioning results. A new magnetic gradient tracking method for scale magnetic targets is proposed based on the combination model of multiple dipoles and the robust nonlinear Huber unscented Kalman filter. The proposed method can effectively improve the positioning accuracy of scale magnetic targets，especially when abnormal observation occurs，it can still accurately locate the target. In order to identify targets such as mines or underwater unexploded projectiles，and considering the mathematical meaning of principal component analysis (PCA)，a inversion algorithm based on the dipole distribution obtained by NHUKF algorithm is proposed. The inversion algorithm can identify the main scale as line structure or surface structure，and the inversion error of main scale is less than 0.3 m，the inversion error of the direction is less than 2 degrees，which provides a reference for the target classification.

Key words: magnetictarget, magneticgradienttensor, underwatermovingplatform, robustness, characteristicsinversion