摘要
变电站二次回路是二次高级集成业务的基础,采用图像识别技术对二次回路的自动特征识别、信息提取,可实现二次回路的智能运维业务。而变电站采集的图片环境背景杂乱、分辨率低以及失真,使得采用图像识别技术识别不规则文本极具挑战。因此,提出一种基于注意力机制的二次回路端子文本检测与识别方法。该方法主要包含预处理、文本检测和文本识别3个部分,其中文本识别部分提出一种时空嵌入编码方法,可以更好利用图片的位置信息。在训练过程中,相较未改进方法仅需要序列级的标注信息,而无需额外细粒度的字符级别框或分割掩码。最后,通过实际工作场景数据集证明该方法不仅易用、性能好,且在识别精度上也优于其他方法。
The secondary circuit of the substation is the basis of the secondary advanced integrated business.The automatic feature recognition and information extraction of the secondary circuit by image recognition technology can realize the secondary circuit's intelligent operation and maintenance business.However,the images collected by the substation have messy backgrounds,low resolution,and distortion,making it very challenging to identify irregular text using image recognition technology.Therefore,a text detection and recognition method of a secondary loop terminal based on an attention mechanism is proposed.This method mainly includes preprocessing,text detection,and text recognition.In the text recognition part,a spatiotemporal embedding encoding method is proposed,which can better use the picture's location information.Compared with the unimproved method,only the sequence⁃level annotation information is needed in the training process,and no additional fine⁃grained character level box or segmentation mask is needed.Finally,it is proved that the proposed method is not only easy to use and has good performance but is also better than other methods in recognition accuracy.
作者
钟鸣
陶军
阿敏夫
杨逸
ZHONG Ming;TAO Jun;A Minfu;YANG Yi(Inner Mongolia Power Research Institute,Hohhot 010020,China;Inner Mongolia Autonomous Region Power System Intelligent Grid Simulation Enterprise Key Laboratory,Hohhot 010020,China)
出处
《电力科学与技术学报》
CAS
CSCD
北大核心
2023年第3期132-139,共8页
Journal of Electric Power Science And Technology
基金
内蒙古自治区科技关键技术攻关计划项目(2019GG373)。
关键词
变电站二次回路
图像识别
注意力机制
特征提取
文本识别
secondary circuit of substation
image recognition
attention mechanism
feature extraction
text recognition
