Abstract: In quaternion model, both azimuth and elevation angles of signals should be known before using the conventional quaternion estimation of signal parameters via rotational invariance techniques (Q-ESPRIT) to estimate the polarized signal angles with uniform linear array (ULA), which consists of co-centered orthogonal loop and dipole (COLD) antennas. The existence of "quaternion model coherence problem" also results in the loss of orthogonal information and the necessity of long vector model for signals' the polarization information estimation. This paper solves the above problem effectively by proposing a novel dimension reduction quaternion ESPRIT(DRQ-ESPRIT) algorithm based on cross-correlation matrix. Making full use of the orthogonality and data structure of the quaternion model, the proposed algorithm can estimate the direction of arrivals (DOA) and polarization information of signals with one time eigenvalue decomposition (EVD) operation, and has a less computation. Finally, the numerical simulations show the effectiveness of the proposed method.