An anti-noise <i>ϕ</i> -OTDR based distributed acoustic sensing system for high-speed railway intrusion detection

Abstract

Abstract We present an anti-noise ϕ -optical time-domain reflectometer-based distributed acoustic sensing system that can effectively differentiate noise and interference for high-speed railway intrusion detection. A distributed acoustic sensing interrogator unit, based on digital heterodyne detection, was deployed in a real field railway station and three types of intrusion signals were collected, including wall climbing, wall breaking, and barbed wire crossing. Sensing signals were analyzed and identified by a comprehensive deep model which consisted of a temporal relation extraction module and a spatial feature encoding module. A novel hierarchical structure of the convolutional long short-term memory network was designed for temporal pattern analysis and spatial features were extracted by a convolution neural network. To accelerate computation, signals with lower energy were filtered out and the combined spatial-temporal features were used for classification. The experiment on real field data achieved over 90% of the threat detection rate, with an approximately 10% false alarm rate, under various parameter settings. The anti-noise performance was compared with the latest high-speed railway intrusion system and it demonstrated a significant improvement of noise and threat identification.