A Highly Efficient Unsupervised Recognition Method for Fiber-Optic DAS with Spiking Convolutional Neuron Network (SCNN)

In this paper an efficient unsupervised spiking convolutional neural network (SCNN) is proposed for the first time to improve the event recognition performance and timeliness simultaneously in the fiber-optic distributed acoustic sensor (DAS). Its effectiveness is validated through three groups of field tests, including small dataset A with balanced, typical and consistent data, bad datasets B and C full of inconstant and atypical data, and fully imbalanced dataset D respectively. It improves its computation efficiency by six times while maintaining its stably high recognition accuracy compared with the standard SNN; while compared with the mainstream supervised CNN, the SCNN has a stronger generalization ability with complicated inconsistent, or imbalanced samples, which often happens in practice.