[15p-A202-1~14] 3.2 Information photonics and image engineering (formerly 3.3)

[15p-A202-11] Multimode-fiber wavemeter using the machine learning and its tolerance against Poisson intensity noise and dark counts.

〇(M2)Kosuke Okuyama¹, Daisuke Nishiwaki¹, Yutaka Matsuno¹, Naoto Namekata², Shuichiro Inoue² (1.Nihon University CST, 2.Nihon University IQS)

Keywords:Deeplearning, Multimode-fiber

We have studied on the machine learning approach for the light-wavelength estimation using a multi-mode-fiber-based wavemeter. The developed convolutional neural network (CNN) was able to well estimate (classify) a wavelength oflight from a speckle image obtained by the wavemeter with an accuracy of ~ 100% and a wavelength resolution of less than0.1 nm. Here, we report on an experimental analysis of the estimation accuracy degradation due to intensity (photon-number) fluctuations in measured light. The speckle images with photon-number fluctuations were calculated by means of the Monte Carlo method, and the wavelength estimation was carried out using them. Uncorrelated errors were observed,which indicates that the developed CNN might not learn specific (or principle) features of speckle images.