Pipeline Threat Event Identification Based on GAF of Distributed Fiber Optic Signals
J. Li, Ruixu Yao, Jiarui Zhang, Xinwei Zhang, Meiying Ren, Tian Ma
Abstract
This article proposes a long-distance pipeline threat event identification method based on Gramian angular field (GAF) and deep learning networks for phase sensitive Rayleigh distributed fiber optic vibration system. The 11 types of threat events of vibration signals along a 28.9-km natural pipeline are collected and converted to 2-D images using GAF. The performance of five deep learning networks, such as GoogLeNet, visual geometry group (VGG), AlexNet, ShuffleNet, and DenseNet, is compared for events classification. The results show that DenseNet achieved the highest classification accuracy within the 11 types of threat events. Moreover, the false alarm rates of manual digging (MD) and mechanical excavation (ME) among the 11 events were 0.31% and 0.18%, respectively. The proposed method is suitable for complicated pipeline safety monitoring niche and offers fast and accurate recognition solution for long-distance pipeline safety monitoring.