Text Recognition of Power Equipment Nameplates Based on Deep Learning

Abstract

There is abundant information in power equipment nameplates. Extracting information from nameplates through image and text recognition technology enables the high effective and quick works on statics and account check of power equipment, which is also beneficial to improve the equipment management level of power system. Con-sidering the great difference of text recognition between power equipment nameplates and ordinary images, this paper proposes an algorithm of automatic recognition of power equipment nameplates based on deep learning. This algorithm consists of three parts, namely nameplate detection, text detection and text recognition, respec-tively. By improving the design of loss functions, adding the correction of text recognition, and synthesizing text images, the values of mAP and F1 reach 92.2% and 91.2% on test dataset for the nameplate and text detection models respectively, and the text recognition model yields the accuracy of 94.0% and 82.3% for the recognition of characters and text lines respectively.

Publication
Electric Power Engineering Technology
Date