For solving the problems of poor initial value selection sensitivity and weak convergence performance while using fast independent component analysis (FastICA) algorithm to extract co-frequency aliased signals, a double relaxation factors modified FastICA (DLM-FastICA) algorithm is proposed. Firstly, the double relaxation factor is introduced into Newton iteration method, and the optimal weight separation matrix is obtained by adjusting the combination coefficient of separation matrix adaptively, then the sensitivity of FastICA algorithm to the initial value is improved; furthermore, the extraction signal is obtained via fast convergence characteristics of modified FastICA (M-FastICA), and the separation accuracy and convergence speed of the algorithm are improved. The simulation results show that the similarity coefficient between the extracted signal and the source signal reaches 0.9, meanwhile compared to the original algorithm, the iteration times are reduced by nearly 40%, so the proposed algorithm has faster and more stable extraction performance.