Practical Aspects of Perimeter Intrusion Detection and Nuisance Suppression for Distributed Fiber-Optic Sensors

Fiber optic sensors protect resources and critical infrastructure in commercial and defense applications. Distributed fiber optic sensors can be designed using various sensing technologies, such as Mach–Zehnder interferometers (MZIs), Michelson interferometers, and phase-sensitive optical time-domain reflectometry (
$\Phi $
-OTDR). The ability to eliminate nuisance alarms without compromising the probability of detection (POD) is critical for accepting perimeter intrusion detection systems (PIDSs). In this article, we discuss the importance of sensor installation, the validity of intrusion tests, the effects of the signal-to-noise ratio (SNR) and frequency contents on time lag estimation, quantification of the POD and nuisance alarm rate (NAR), and the need to validate intrusion recognition algorithms in realistic environments. Moreover, this article demonstrates the precision of intrusion localization at various locations along the perimeter, both during torrential rain (TR) and under calm weather conditions. In a longitudinal study, this article also demonstrates the effectiveness of level crossing (LC)-based intrusion detection algorithms, integrated into Mach–Zehnder (MZ)-distributed sensors, at a practical site. During the longitudinal investigation, we found that nuisance alarms could be suppressed for rainfall rates exceeding 225 mm/day while detecting intrusions and nuisances simultaneously. The intrusion location spread in quiet and rainy conditions was within ±10 m with a 95% confidence level of 0.81. In addition, the convolutional neural network (CNN) architectures, AlexNet, ResNet-50, VGG-16, and GoogLeNet, were investigated in terms of performance and suitability to MZ-based PIDS. The CNN models can discriminate between intrusion and TR events with a 98.04% accuracy rate. The latency analysis revealed that the LC-based algorithm outperformed the CNN models in terms of processing time. This research is intended to guide the development of PIDS and its algorithms.