作者:\n\t翟晓卉，孙凯，赵吉福，孙艳玲，邢宇，郭凯旋，王海英\n

Authors:\n\tZHAI Xiaohui,SUN Kai,ZHAO Jifu,SUN Yanling,XING Yu,GUO Kaixuan,WANG Haiying\n
摘要:\n\t针对智能电表通信模块铭牌检测和识别问题，本文提出了一种基于ResSE-SegNet的级联深度学习方法，采用高斯滤波和自适应图像对比度增强等图像预处理技术，通过深度语义分割网络来确定输入图像中厂家名称所在的区域，并采用深度编解码器网络结构进行分割，构建并训练端到端的卷积神经网络(CNN)模型用于识别不同的厂家。最后通过全维度智能电表检测系统获得通信模块图像的数据集并进行通信模块铭牌的检测和识别实验，实验结果表明，该方法的召回率为92.34%，精确率为93.97%，F_1Score为92.14%，对智能电表通信模块厂家的分类准确率为98.53%。\n

Abstract:\n\tAiming at the problem of nameplate detection and recognition of smart meter communication modules, this paper proposes a cascaded deep learning method based on ResSE-SegNet, which adopts image preprocessing techniques such as Gaussian filtering and adaptive image contrast enhancement, and determines the input through a deep semantic segmentation network.The region where the manufacturer′s name is located in the image is segmented using a deep codec network structure, and an end-to-end convolutional neural network (CNN) model is constructed and trained to identify different manufacturers.Finally, the data set of the communication module image is obtained through the full-dimensional smart meter detection system, and the detection and recognition experiment of the communication module nameplate is carried out.The experimental results show that this method achieves a recall rate of 92.34%, a precision rate of 93.97%, and an F1-Score of 92.14%. The classification accuracy rate for smart meter communication module manufacturers is 98.53%.\n


PDF全文下载地址:

可免费Download/下载PDF全文