Abstract: Due to the particularity of oil and gas, there are many non-destructive interference events around oil and gas pipelines, which brings up great challenges to the warning accuracy of Pipeline Security Forewarning System (PSFS). In order to improve the detection level of excavation activity (EA) in the case of background interference, an excavation activity detection (EAD) algorithm was developed in this paper. A database inclusive of EA and non excavation activity (NEA)was established based on the field data of PSFS in Wuhan-Ezhou section of West-to-East Gas Pipeline. The Pisarenko harmonic decomposition feature of vibration signal, the Itakura distance feature of adjacent sampling points, and the Mel frequency cestrum coefficient (MFCC) feature were extracted according to signal properties. And a two-layer classification structure of hidden Markov model (HMM) and support vector machine (SVM) was designed. In this structure, after the optimal state sequence is calculated in HMM, it is input into SVM classifier to detect if there is an EA in the signals. It is shown that this EAD algorithm can effectively increase the detection rate of EA signal in the case of background interference, which is 85.5% during the field test. The research results provide the theoretical basis for the application of PSFS in open sites.