Abstract: 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.