Based on previous pattern recognition research which uses Mel Cepstrum coefficient method to extract frequency characteristics of disturbance signal， a fiber intrusion pattern recognition method using 1-Dimension convolutional neural network is purposed for interferometric distributed optical fiber sensing system. Hierarchical thresholds of the restored signal are used to judge and extract the intrusion behavior， which effectively reduces the calculation time compared with the framing method. A one-dimensional convolutional neural network is constructed based on the frequency domain features of the intrusion signal after Fourier transform to extract the characteristics of the disturbance signal adaptively. A line-based Sagnac interference system is set up to acquire data. By training the network with a large number of sample data， a good classification result is obtained. The average recognition rate of the verification set reaches 96.5%. The trained convolution kernels and the convoluted intrusion signals are discussed. After zscore standardization， the one-dimensional convolutional neural network can identify some features in the frequency domain of the signals， and the recognition result of the branch tapping signal with complex frequency components is greatly improved.