基于深度学习的电力通信光纤配线标签识别研究  

Research on Recognition of Optical Fiber Distribution Labels for Power Communication Based on Deep Learning

作　　者：杨志敏[1] 陈一童 吴桂龙 黄强 贺云 YANG Zhimin;CHEN Yitong;WU Guilong;HUANG Qiang;HE Yun(Power Dispatching&Control Center,China Southern Power Grid,Guangzhou 510525,China)

机构地区：[1]中国南方电网电力调度控制中心,广东广州510525

出　　处：《电力信息与通信技术》2022年第4期18-23,共6页Electric Power Information and Communication Technology

基　　金：南方电网重点科技项目“通信网运行智慧管控技术研究及应用”(ZDKJXM20190013)。

摘　　要：电力光纤作为电力业务通道承载介质,其标签管理尤为重要。光纤配线标签通常粘贴在光纤端子处,软体材质容易导致标签堆叠、扭曲,加之光照不均、背景复杂、文本变形和文本方向不一致等干扰及影响因素,使得现有文本检测和识别方法难以正确识别光纤配线标签。针对这一问题,文章提出一种基于YOLO、PSENet文本检测和DenseNet、Seq2Seq、Attention文本识别与文本纠错技术相结合的光纤配线标签识别方法。通过实验对比,该标签识别方法具有较高的文本识别准确率,研究成果在通信资源数字化管理领域具有实际应用价值。As the carrier medium of the power business channel,the label management of the power optical fiber is particularly important.The optical fiber wiring label is usually pasted at the optical fiber terminal,and the soft material is easy to cause label stacking and distortion.In addition,the interference and influencing factors such as uneven illumination,complex background,text deformation and inconsistent text direction make it difficult for the existing text detection and recognition methods in identifying the optical fiber wiring label.In order to solve the problem,this paper proposes a optical fiber distribution label recognition method based on text detection of YOLO and PSEnet and the combination of text recognition of DenseNet,Seq2Seq and attention and error correction about text.By experimental comparison,the solution has achieved high text recognition accuracy,the program has good results in various indicators,and the research results have practical value in the field of digital management of communication resources.

关 键 词：文本检测 YOLO PSENet DenseNet 文本纠错 

分 类 号：TN915.853[电子电信—通信与信息系统]