In order to solve the problem of magnetic moment estimation of magnetic targets, the inverse problem of magnetic moment estimation was constructed based on the hybrid model of ellipsoid and magnetic dipole array. In order to solve the ill-posed problem of the magnetic moment estimation equations, the method was designed to estimate the magnetic moment parameters of the ellipsoid and magnetic dipole array, and the improved discrepancy principle and the maximum chi-square distribution stop criterion were introduced to improve the semi-convergence behavior of the conjugate gradient least squares method. Through simulation examples of magnetic target magnetic moment estimation and ship model measurement data, the performance of two conjugate gradient least squares methods, Tikhonov algorithm and stepwise regression method, was compared and analyzed from four aspects: relative error of magnetic moment estimation, relative error of magnetic field fitting, relative error of magnetic field extrapolation, and computational time complexity. The conjugate gradient least squares method has the advantages of high accuracy of magnetic moment estimation, high immunity of magnetic moment estimation to interference, high accuracy of magnetic field fitting and extrapolation, and low computational time complexity.