An event recognition method based on MFCC, superposition algorithm and deep learning for buried distributed optical fiber sensors

Abstract

An event recognition method based on Mel-frequency cepstrum coefficient (MFCC), superposition algorithm and deep learning is proposed for Φ-OTDR event classification with buried fiber. The MFCC data matrix and temporal–spatial data matrix of intrusion events are superimposed and input into the convolutional neural network (CNN) to determine the event types. Six presentation formats of data samples are selected for comparison. Experimental results based on 8185 samples of 8 event classification show that the proposed method can illustrate the event feature better and achieves the highest classification accuracy of 99.55% and 97.95%, respectively, on two networks with different depths. The accuracies are 0.69% and 1.30% higher than those of the experiment using MFCC as event feature alone.