基于分类模板数据库的电气铭牌识别

Electrical equipment nameplate recognition based on classification template database

电气铭牌文本行多,识别难度较大,当前技术均难以实现有效准确的识别。为解决这一问题,提出一种基于分类模板数据库的方法辅助电气铭牌识别。根据电气铭牌文本行分布情况,将识别分为不可变区域和可变区域识别。每类铭牌不可变区域相当于一张图像,所以将文字识别转化为图像分类问题。通过对不可变区域建立分类模板数据库,引进卷积神经网络对电气铭牌图像进行分类。经实验验证,该方法能准确高效地识别电气铭牌的不可变区域,从而大幅提升了电气铭牌识别的准确度。

There are many lines of items in text on the electrical equipment nameplate(EEN),so it is difficult to identify them effectively and accurately by the current technologies.To solve this problem,a method based on the classification template database is proposed to assist the EEN recognition.The EEN recognition is divided into invariable area and variable area according to the distribution of text lines on EEN.The invariable area of each type of nameplates is equivalent to an image,so that the character recognition is transformed into the image classification.By establishing the classification template database for the invariable area,the convolutional neural network is introduced to classify the EEN images.The results of experiment verification show this method can recognize the invariable areas of EEN efficiently and accurately,thus greatly improving the accuracy of EEN identification.