光纤网络哑资源智能检测与清查方法

ODN网络是FTTH网络中的重要组成部分，其质量好坏直接关系到客户使用宽带网络的体验。为了对哑资源进行高效精准管理，改善网络质量，提升宽带业务开通效率，降低维护成本，文中设计优化了哑资源智能检测及清查系统。该系统基于深度学习中的YOLOX算法，设计改进提出SGDMN优化器，该优化器能有效抑制振荡并且在加速训练的同时对梯度进行校正，以此对参数进行更新；同时对数据增强方式进行优化，在不脱离真实场景的前提下选择数据增强方式，以此实现对分光器、尾纤、标签、二维码等关键信息进行目标检测，进行分类标识，维护人员可基于这些关键信息配合资管系统信息完成哑资源清查。通过其他经典目标识别算法进行对比实验，结果表明改进后的YOLOX算法精度更高，满足哑资源智能检测及清查实际工程需求。

Recognition and Inventory Method for Optical Network Dumb Resources

ODN network is an important part of FTTH network, and its quality is directly related to the experience of customers using broadband network. In order to efficiently and accurately manage dummy resources, improve network quality, improve broadband service opening efficiency, and reduce maintenance costs, the intelligent detection and inventory system of dumb resource is designed and optimized in this paper. Based on YOLOX algorithm in deep learning, the system designs and improves SGDMN optimizer, which can effectively suppress oscillation and correct gradient while accelerating training, so as to update parameters. At the same time, optimize the data enhancement mode, select the data enhancement mode without departing from the real scene, so as to achieve target detection and classification identification of key information such as optical splitter, pigtail, label and two-dimensional code. Maintenance personnel can cooperate with the information of the asset management system to complete dumb resources inventory based on these key information. Compared with other classical target recognition algorithms, the results show that the improved YOLOX algorithm has higher accuracy and meets the practical engineering needs of intelligent detection and inventory of dumb resources.