Probabilistic Event Discrimination Algorithm for Fiber Optic Perimeter Security Systems

This paper proposes an event discrimination algorithm with probabilistic output for fiber optic perimeter security systems. Multiscale permutation entropy and the zero-crossing rate are employed to increase the efficiency of the algorithm and extract intrusion features. A probabilistic support vector machine is used to calculate multiple event probabilities by solving a convex quadratic programming problem. The experimental results demonstrate that the proposed algorithm can distinguish six intrusion events at an average recognition rate of 92.68% and in a processing time of 0.32 s. Compared with traditional discrimination methods, the proposed algorithm obtains more detailed information (probabilities) of intrusion events. The recognition results are obtained after analyzing the probabilities, which not only reduces the decision-making costs but also reduces the losses from erroneous decisions. Therefore, the proposed high-efficiency feature extraction method and reliable discrimination algorithm can be used to improve the monitoring efficiency of fiber optic perimeter security systems.