An open dataset of φ-OTDR events with two classification models as baselines

Abstract

Event classification with machine learning methods, has now become the research spotlight for phase sensitive optical time domain reflectometer (Ф-OTDR, also referred to as DAS or DVS) system. Unlike in the computer science field, however, each team works on its own dataset and their codes in private, which makes it difficult to compare and evaluate different models on a fair basis and impairs the value of those work. Hence, we set up an intensity-based Ф-OTDR system, from which we collect data, build and make public the first open dataset, according to our searches via google. It contains six types of events with a total of 15,612 samples. We also publicize codes for two common baseline models, which are the SVM (support vector machine, 1D method) and CNN (convolutional neural network, 2D approach) models. The average recognition rates are greater than 82 % for SVM and 94 % for CNN. The dataset and codes can be used as data source and baselines for Ф-OTDR event classification, providing fair evaluation criteria and for further development of machine learning models.