Event Detection and Classification Algorithm using Wavelet and Machine Learning Technique for Vibration fence PIDS

The paper presents a design of a real-time algorithm to detect and classify an event that is occurring on a perimeter fence based on the Triboelectric sensor. The intrusion events like climbing, cutting the fence etc. produces vibrations on the fence. These vibrations are detected by Triboelectric cable which runs alongside the fence. Event detection and classification algorithm has been designed based on Wavelet and Machine learning techniques. Field data were collected for basic simulated events such as hand climb, metal tap and wood tap. These signals are pre-processed first before passing to a further part of the algorithm. The Discrete Wavelet Transform (DWT) is used here to extract features from the pre-processed data. These features are used to detect and further classify events using supervised machine learning methods like Support Vector Machine (SVM), decision trees and k-Nearest Neighbours (kNN). The accuracy and complexity of these methods were compared, and based on this analysis a machine learning method has been selected for designing the classification algorithm. The model trained by the selected classification method is further used to derive the formulation required for the classification. The derived formulation is used in designing a real-time algorithm for event detection and classification. The algorithm has been implemented on Digital Signal Processor (DSP) - Field Programmable Gate Array (FPGA) based hardware to detect and classify the events occurring on the fence in real-time.