Signal recognition method based on Mel frequency cepstral coefficients and fast dynamic time warping for optical fiber perimeter defense systems.

Abstract

To improve the accuracy of signal recognition in optical fiber perimeter defense systems, a method based on Mel frequency cepstral coefficients (MFCCs) and a fast dynamic time warping (FastDTW) algorithm is proposed. Four kinds of sensing signals are acquired by employing an in-line Sagnac interferometer system. MFCC features of the signals are extracted. The FastDTW algorithm is utilized to calculate distances of the optimal warping paths between a test sample and all reference templates. According to the nearest neighbor criterion, signal recognition results are obtained. The average accuracy is over 96%. This proposed identification method is highly efficient and easy to implement.