[13p-PA2-1~11] 3.14 Optical control devices and optical fibers

▲ [13p-PA2-10] Convolutional Neural Network for Improving Spatial Resolution of BOCDR

Jelah Nieva Caceres^1,2, Kohei Noda¹, Heeyoung Lee³, 〇Yosuke Mizuno¹, Kentaro Nakamura¹ (1.Tokyo Tech, 2.NUS, 3.SIT)

Keywords:optical fiber sensor, neural network, Brillouin scattering

We provide specific trial results of deep-learning-assisted Brillouin optical correlation-domain reflectometry (BOCDR). We employ a convolutional neural network to obtain the intrinsic Brillouin frequency shift distribution directly from the measured Brillouin gain spectrum (BGS) distribution in BOCDR. By using this algorithm, actual spatial resolution of the system can potentially be enhanced compared with nominal resolution conventionally calculated using a convolved BGS distribution. We show that the spatial resolution can be at least 4 times higher than the nominal value.