Abstract: The characteristics of background vibration noise in reactor primary system were studied. Based on the traditional signal extraction methods from loose parts, the de-noising autonomous adaptive threshold was chosen by using the method of adding time window to wavelet packets coefficients. At the same time, a new threshold processing function was adapted to implement the de-noise process to the loose parts impact signal, and the collected impact signal of steel ball from experimental apparatus was analyzed. The results show that the method in this article can effectively track the change of noise intensity and restrain background noise, and correctly identify impact signal of loose parts and extract relevant features.