Enhancing Φ-OTDR Classification Performance through Event Augmentation

As one of the research focuses in the past decades, phase sensitive optical domain reflectometer (Φ-OTDR) has come to the tipping point of its wide application. With the expansion of its application, how to classify and identify Φ-OTDR events more effectively and efficiently in practical applications has become a key and urgent issue. Over the past several years, the incorporation of machine learning methodologies has garnered considerable attention in this area. Nevertheless, the performance of those machine learning models heavily relies on the quantity and quality of the collected data. That is, the challenge of collecting rare event signals in the sensing applications strongly limits the ability of the model to classify accurately. To overcome the above weakness and further boost the capability of Ф-OTDR, we propose an event augmentation method to enhance the diversity and generalization of raw data. Experimental results show that the proposed method improves the event classification accuracy of our Ф-OTDR from 76.4% to 91.0%. The proposed method represents a promising solution to tackle with the challenges of Ф-OTDR as well as other distributed fiber optic sensors.