摘要
为规范电力作业中“两票”的规范管理问题,提出一种基于深度学习的轻量化图像检测与识别框架,以实现对纸质“两票”图像中的印章及日期信息的快速识别。该框架首先基于YOLOv4网络对印章及日期等关键信息进行检测,然后采用MoblileNetv3和自行设计的Ghost-OCRNet网络进行印章和日期识别;针对日期信息涉及手写与打印字体混合的现状,设计一种轻量化的Ghost-OCRNet识别网络,可对日期进行无分割的序列化识别。实验结果表明所提出的纸质工作票图像识别方法平均运行速度为5.6帧/s(FPS),识别准确率达到93.5%,在保证识别精度的前提下,能够满足实时运行要求。
In order to regularize the standard management of"two tickets"in electric power operation,a lightweight image detection and recognition framework based on deep learning is proposed to realize the rapid recognition of seals and date information in the paper"two tickets"images.This framework starts with the detection of key information such as seal and date based on the YOLOv4 network,followed by the seal and date identification based on MoblileNetv3 and indigenously designed Ghost-OCRNet network.In allusion to the current situation that the date information involves the mixture of handwriting and printed fonts,a lightweight Ghost-OCRNet identification network is designed,which is available to sequentially identify the date without division.The experimental results show that the average running speed of the proposed paper work ticket image recognition method is 5.6 frames/s(FPS)and the recognition accuracy is 93.5%,which is available for the realtime operation requirements under the premise of ensuring the recognition accuracy.
作者
来骏
仲赞
林文钊
沈炼
LAI Jun;ZHONG Zan;LIN Wenzhao;SHEN Lian(Huzhou Power Supply Company,Huzhou 313000;Changxing County Power Supply Company,Huzhou 313100)
出处
《电力安全技术》
2023年第11期71-74,共4页
Electric Safety Technology
关键词
工作票识别
深度学习
目标检测
轻量化网络
work ticket recognition
deep learning
target detection
lightweight network