Abstract
This paper proposes an unsupervised algorithm for vehicle event classification based Distributed Fiber Optic Sensing (DOFS) data. The objective of the proposed algorithm is to separate vehicle-induced signals from complex background noise by adaptively estimate varying noise conditions and applying a dual-threshold classification method. The vehicle events are classified through a dual-threshold mechanism that is based on physically meaningful spatiotemporal features, which are Intensity Ratio and Spatial Width. With real-world expressway dataset evaluation, the method achieved an F1- score of 89.41% and an accuracy of 85.00% without pre-training. The results also indicate stable performance across different seasons, suggesting that the feature-based framework can support efficient data reduction and event identification in long-term pavement monitoring.