Research on End-to-End Text Spotting Algorithm for Power Equipment Nameplate Images

Text spotting for power equipment nameplate images is important for power system maintenance and stability. However, it currently shows a relatively low accuracy in text spotting algorithms for power equipment nameplate images. In this paper, an end-to-end text spotting algorithm MV-Bridge is proposed to improve the accuracy. First, the proposed MV-Bridge adopts an advanced text detection algorithm (Toward Accurate Detection of Challenging Scene Text in the Wild, MixNet) and text recognition algorithm (Vision Permutable Extractor for Fast and Efficient Scene Text Recognition, VIPTR) as the detector and recognizer to detect and recognize the texts, respectively. Then, a Bridge module is introduced to integrate the detector and recognizer into a single network, which enables the joint training between them to avoid error accumulation. An Adapter module is introduced to facilitate the joint training for better detection and recognition accuracy. Finally, the accuracy of MV-Bridge was demonstrated on a public power equipment nameplate image dataset. The results of this paper may provide some useful information for power system maintenance and stability.