To improve the vibration event recognition performance of the distributed acoustic sensing system

a distributed optical fiber vibration recognition method based on multi-time scale weight adaptation is proposed. Firstly

33-dimensional complementary features are extracted in parallel

which are divided into three categories:time-frequency features that describe dynamic energy variations in both the time and frequency domains

structural features that depict the overall waveform morphology and local stability

and information coupling features that quantify the coupling between signal amplitude and phase.Feature importance is introduced to quantify the contribution of each feature to the classification results.Secondly

conditional mutual information is employed to evaluate the independent discriminative power of features

ensuring that the fusion weights simultaneously account for both feature importance and the amount of independent information they provide.Finally

a multi-time scale weight evolution framework is constructed to adaptively adjust the feature combination strategy based on the instantaneous characteristics of vibration signals

enabling precise responses to different types of vibration events.Field experiments show that the proposed method achieves an overall recognition accuracy of 99% for vibration events such as excavator operation

hammering

trampling

and drilling

validating the effectiveness of the proposed method.