Research on Vibration Event Classification in $\Phi-$ OTDR Systems Using MFCC Feature Extraction and Improved Swin Transformer

In this paper, an event recognition method based on MFCC and improved Swin Transformer is proposed for
$\Phi-$
OTDR event classification of ground-buried sensing fibers. For the scenario of buried optical fibers, MFCC signals and time-domain signals of six types of data samples were selected for comparison. Compared to time domain signals, the average accuracy of MFCC signals on the improved Swin Transformer is 11.95% higher. The feedforward control idea is applied to the structural improvement and information processing process of Swin Transformer. The results of 6 events classification based on 5636 samples show that compared with Swin Transformer, the improved method achieves higher classification accuracy on the self-made dataset, and the average accuracy is increased from 91.499% to 92.674
^%
compared with traditional Swin Transformer.