Phase sensitive optical time domain reflectometer (φ-OTDR) becomes more and more important in intrusion alarming and other dynamic sensing fields. Meanwhile, it makes much sense to classify the intrusion fast and effectively. Therefore, a fast pattern recognition method based on frequency spectrum is presented and experimentally verified. The proposed method is named EDFS, short for Euclidean distance of fast Fourier transform (FFT) frequency spectrum of the detected signals. The signal is abstracted by short-time shifted delta(SSD)and short- time energy, and the features are obtained from the abstracted signal after normalization and FFT transformation. The euclidean distance of the spectra between features and models is used to classify the intrusion. The effectivity and instantaneity are verified by three typical intrusion disturbances. It is shown experimentally that intrusions can be recognized clearly in a period less than one tenth of that by conventional dynamic time warping (DTW). The method needs fewer training models than other recognition methods, such as the neural network, and has a merit of mitigating influence of environmental noises.