Deep Learning Based Threat Classification for Fiber Optic Distributed Acoustic Sensing Using SNR Dependent Data Generation

In this study, a novel method is proposed to generate SNR dependent database and classify seismic events for fiber optic distributed acoustic sensing (DAS) systems. Optical time-domain reflectometry (OTDR) is used to acquire DAS signals. Proposed data creation method generates signals with different SNR values which is based on real channel noise characteristics. By this way, from the limited dataset, huge dataset consists of three different seismic events such as hammer hit, digging with pickaxe and digging with shovel is generated. In the classification part, two different Deep Learning algorithm (Convolutional Neural Network and fully connected neural networks) are used to identify three different seismic events. Results show that remarkable identification accuracy for the three different SNR ranges is achieved.