In this paper, a matrix pencil characteristic equation-based source number estimator is proposed. An enhanced matrix is defined through partition-and-stacking process by the original data on the uniform linear array, and its covariance matrix is computed, which is proved to be a toeplitz conjugate symmetry matrix. The covariance matrix elements are aligned to from a vector. A measurement matrix is formed by this vector considering the characteristic equation. Gerschgorin disk estimator (GDE) is employed with the measurement matrix to estimate source number. As partition-and-stacking processing is equivalent to spatially smoothing averaging, and at the same time reduces the influence of noise, our method is efficient with correlated signals, and performs well when there are only few snapshots. Simulation results prove the efficiency of our method under severe conditions.