φ-OTDR pattern recognition based on CNN-LSTM

Abstract

Listen

Translate article

We proposed a pattern recognition strategy based on the long short-term memory network (LSTM) and convolutional neural network (CNN), with phase-sensitive optical time domain reflectometry (φ-OTDR) realizing vibration sensing and data acquisition in application scenarios. Given the time domain curve as well as its discrete wavelet transform (DWT) and short-time Fourier transform (STFT) as the input, the trained LSTM-CNN can effectively identify six kinds of target signals, which can assist users in taking appropriate measures. The significant improvement of LSTM on classification performance has been proven in comparison to the conventional artificial neural network (ANN) and CNN, providing an application example for the integration of LSTM and optical fiber sensing.