Fiber Optic Incidents Detection and Classification with Yolo Method

Abstract

in this paper we propose an automatic and real time management system of incidents in fiber optic telecommunications networks. The first step is to collect images of incidents in the fiber optic transmission networks of telecommunications operators in Senegal. Then, from these collected images, we develop a convolutional neural network architecture YOLO (You Only Look Once) to automatically detect in real time incidents in the fiber optic telecommunications networks as well as the impacts on the different services of the operator. This proposed system is an extension of two works on incident management systems in fiber optic telecommunications networks with deep learning algorithms based on convolutional neural networks. Indeed, the first work deals with the use of convolutional neural networks (CNN) and the second with the use of regional mask neural networks (Mask-RCNN). The reality is that in the current networks (Internet) in Senegal as everywhere in the world the infrastructure based on optical fiber represents the most important part of the transmission media used for the transport of information. And this infrastructure is growing more and more in parallel with the needs of users of telecommunications networks. So finding ways to detect incidents faster than Mask-RCNN and more efficiently than CNN is essential for better incident handling. Our system ensures better handling through faster, more efficient detection and analysis of downed or cut optical link incidents through the use of the YOLO algorithm. YOLO is designed to detect objects on a digital image, i.e. the status of a link impacted by an incident from the image for our system.