An intelligent model integrating multi-scale features and end-to-end classifier for accurate events recognition along fiber optic fence

Pattern recognition is crucial for event detection and analysis in diverse optical fiber-based perimeter security systems. Although numerous methods and schemes have been investigated in this field, rapid and accurate pattern recognition still poses challenges for promptly identifying multiple events in practical application scenarios. For this reason, in this article, we propose and design an accurate model that integrates multi-scale features with an end-to-end classifier, enabling instantaneous and precise recognition of sensing patterns. Firstly, the original acquired time-series sensing signals are converted into two-dimensional time–frequency spectrum using a Stockwell transform, thus enabling accurate representation of the time–frequency features and phase variation characteristics. Subsequently, these acquired two-dimensional spectrum images collectively constitute the final sample dataset, which is then utilized in an end-to-end classifier that integrates a convolutional neural network with a gated recurrent unit neural network, for identifying and classifying the extent of events on the fiber optic fence. Finally, to establish the cogency and acceptability of our approach, a series of rigorous field tests have been conducted in a practical perimeter security system spanning a total sensing length of 21 km. In particular, nine types of sensing events are collected as data samples, acquired through an asymmetric dual Mach-Zehnder interferometer-based optical fiber perimeter security system. The results demonstrate that the proposed scheme outperforms previously reported schemes used for similar purposes. Verification has shown that the mean accuracy of the given nine sensing patterns achieved 98.96 %, while the mean processing time required was only 0.31 s. Thus, we believe that the proposed model holds significant promise for multiple event recognition in practical application scenarios.