Strategic-cum-Domestic vehicular movement detection through Deep Learning approach using designed Fiber-Optic distributed vibration sensor

This paper presents the design and proposal of vehicle identification and classification sensing mechanism using fiber-optic distributed acoustic sensing (DAS). The seismic signatures of strategic-cum-domestic vehicles were made to interact with the embedded optical fibers and the acquired backscattered optical data was processed for feature extraction. The vehicle type was identified through the signal processing using a deep learning approach resulting in a novel and intelligent vehicle identification platform. Data denoising was performed using various time-series mining algorithms with at least 500 epochs to test the model accuracy. This fully automated and end-to-end simulated sensing system used an asymmetric generative adversarial network (GAN) for denoising the time series data resulting in the training and testing accuracy of 73.63% and 69.5% respectively. The model can also be employed to measure the real-time speed and position of the vehicle.