\u03a6-OTDR Distributed Acoustic Sensing Vibration Events Dataset (Seven-Class Open-Set Benchmark)

This dataset provides a reproducible open-set benchmark for vibration event recognition in phase-sensitive optical time-domain reflectometry (\u03a6-OTDR) based distributed acoustic sensing (DAS) systems. Seven labeled event types are included (Digging, Environment Noise, Sawing, Shaking, Striking, Walking, Running), acquired with a self-built \u03a6-OTDR DAS platform in a perimeter-security deployment. Each sample is represented as a 224\u00d7224 Gramian Angular Difference Field (GADF) image generated from a normalized vibration trace and is assigned to predefined train\/validation\/test splits. The six classes Digging\u2013Walking (5513 samples) are used as known classes for training and evaluation, while the Running class (364 samples) is held out as unknown for open-set testing. We further provide labels supporting an alternative two-unknown-class protocol, in which Shaking and Running are jointly treated as unknown. The dataset was used to evaluate the proposed JCRNet open-set recognition framework in our manuscript \u201cOpen-Set Event Recognition for Distributed Fiber-Optic Vibration Sensing Systems Based on JCRNet,\u201d but is model-agnostic and suitable for benchmarking a wide range of methods for time-series encoding, open-set recognition, and anomaly detection in distributed fiber sensing.