An Attention-Enhanced {\Phi}-OTDR Event Recognition Framework for Edge-Based Distributed Acoustic Sensing

Abstract

Phase-sensitive optical time-domain reflectometry {\Phi}-OTDR has emerged as a promising sensing technology in Internet of Things (IoT) infrastructures, enabling large-scale distributed acoustic sensing (DAS) for real-time monitoring at the edge in smart cities, industrial pipelines, and critical infrastructures. However, accurately recognizing events from massive {\Phi}-OTDR data streams remains challenging, as existing deep learning methods either disrupt the inherent spatiotemporal structure of signals or incur prohibitive computational costs, limiting their applicability in resource-constrained edge computing scenarios. To overcome these challenges, we propose a novel STFT-based Attention-Enhanced Convolutional Neural Network (STFT-AECNN), which represents multi-channel time-series data as stacked spectrograms to fully exploit their spatiotemporal characteristics while enabling efficient 2D CNN processing. A Spatial Efficient Attention Module (SEAM) is further introduced to adaptively emphasize the most informative channels, and a joint Cross-Entropy and Triplet loss is adopted to enhance the discriminability of the learned feature space. Extensive experiments on the public BJTU {\Phi}-OTDR dataset demonstrate that STFT-AECNN achieves a peak accuracy of 99.94% while maintaining high computational efficiency. These results highlight its potential for real-time, scalable, and robust event recognition in edge-based DAS systems, paving the way for reliable and intelligent IoT sensing applications.