Human activities classification based on ?-OTDR system by utilizing gammatone filter cepstrum coefficient envelope using support vector machine

• 主要な著者: Saleh N.L., Faisal B., Yusri M.S., Sulaiman A.H., Ismail M.F., Nik Zulkefli N.A.H.A., Muhamud-Kayat S., Ismail A., Abdullah F., Jamaludin M.Z., Aripin N.M.
• その他の著者: 57198797134
• フォーマット: 論文
• 出版事項: Elsevier Ltd 2024
• 主題: Distributed acoustic sensor; Energy infrastructure; Gammatone filter cepstral coefficient; Phase optical time-domain reflectometry; Support vector machine; Classification (of information); Feature extraction; Permittivity measurement; Reflection; Reflectometers; Signal to noise ratio; Time domain analysis; Acoustic Sensors; Cepstral coefficients; Energy infrastructures; Gammatone filters; Human activities; Optical time domain reflectometry; Support vectors machine; Support vector machines

Intrusion into the critical energy assets area is a serious problem that may cause essential operations to be disrupted. Besides using closed-circuit television to detect early intrusion, perimeter fencing using fiber optic distributed acoustic sensing is becoming popular. This paper proposes a phase-sensitive optical time-domain reflectometry system-based classification of human activities using a coexisting support vector machine which the Gammatone filter cepstrum coefficient envelope as input features. The detection and classification campaign consists of four phases: detection, feature extraction, classification, and evaluation. A combination of wavelet and normalized differential methods is used for detection to improve the signal-to-noise ratio. Simple dataset management methods that use envelope-wrapped, local maxima, averaging, truncation, rearrangement, and random permutation are beneficial for reducing the dimensionality of problems with lower computational power. The classification performance was considered good, which is higher than 95 %. Despite using the legacy classifier algorithm, an improvement in dataset management presented in this paper proves that it is still realistic to implement and does not require demanding computational power to achieve a good classification result. Also, the way datasets are managed in this paper is made to be flexible enough to work with other legacy classifiers. � 2023 Elsevier Ltd