Intrusion behavior classification method applied in a perimeter security monitoring system

Area of Science:

• Optoelectronics and Sensor Technology
• Machine Learning for Security Applications
• Signal Processing for Intrusion Detection

Background:

• Distributed optic fiber systems are vital for securing critical infrastructure but struggle to differentiate intrusion types and environmental noise.
• Existing methods are prone to false triggers, limiting their effectiveness in real-world security scenarios.
• Accurate classification of intrusion behaviors and interference events is crucial for reliable perimeter security.

Purpose of the Study:

• To develop and validate a novel vibration pattern recognition algorithm for enhanced fiber optic perimeter security.
• To improve the ability of distributed optic fiber systems to distinguish between genuine intrusions and environmental interference.
• To achieve high classification accuracy for various intrusion patterns and interference events.

Main Methods:

• A merged Sagnac interferometer structure was utilized to capture vibration signals.
• A two-part algorithm involving signal pre-processing and multi-layer perceptron neural networks (MLP-NNs) was implemented.
• High-dimensional feature vectors, derived from vibration signal power frequency, were used for pattern recognition.

Main Results:

• The proposed algorithm successfully retrieved and extracted vibration signals.
• The multi-layer perceptron neural networks effectively performed pattern recognition on diverse inputs.
• Experimental deployment on a 10 km fence demonstrated a 97.6% classification accuracy.
• The model showed robust performance in classifying intrusion patterns during integrated evaluation.

Conclusions:

• The developed vibration pattern recognition algorithm significantly enhances the performance of distributed optic fiber perimeter security systems.
• The merged Sagnac interferometer and MLP-NN approach provides a reliable method for distinguishing intrusions from environmental interference.
• The high classification accuracy validates the proposed model's effectiveness for critical infrastructure security.