A method based on morphological filtering theory and power algorithm is put forward to identify low frequency oscillation modes, aiming at how to avoid the limitation of Prony methods that is sensitive to noise thus to cause the inaccuracy. It realizes accurate identification of oscillation modes in the condition of fixed noises. A morphology filter, which is based on mathematical morphology and hemicycle structure element is designed to effectively suppress the noise in the condition of proper size of the element. Prony algorithm is then used to detect the de-noised signal and obtain the low frequency oscillation parameters of every mode. The Matlab simulation results show that it is very necessary to preprocess the data of electric signal. Effectiveness and relatively accuracy are verified.