In order to deal with the nonlinearity, non-stationarity and intermittence of intrusive vibration signal in the optical fiber perimeter system, a method combining time and frequency domains characteristics is proposed to distinguish and locate intrusive vibration signal. The minimum frame length is determined by computing embedded dimension to better reserve the dynamic characteristic of the time series signal. The two stages judging and recognizing methods of intrusive vibration signal are proposed. Firstly, short-term energy and zero-crossing measurements are used for determining whether the vibration signal is generated, and then the intrusive signal is recognized according to the characteristic of the energy distribution of each layer′s wavelet coefficients. This method effectively reduces the efficiencies of recognizing error and loss for optical fiber perimeter system. In order to improve the accuracy of locating intrusive signal, Bayesian adaptive threshold estimation in wavelet domain is applied to reduce the noise of the signal, and the reconstruction signal is finally transformed to frequency domain to find the intrusive point. The experiments result shows that the proposed algorithm is effective.