基于深度学习的变电站巡检机器人自动抄表研究

Research on Automatic Meter Reading for Substation Inspection Robot Based on Deep Learning

为提高变电站巡检机器人自动抄表识别的精度,提出一种深度学习的自动抄表识别方法。以YOLOX网络作为基础框架,在网络通道层和空间层添加卷积注意力模块,同时采用Focal-Loss函数替代BCE-Loss函数,以提高网络的训练速度和识别精度。结果表明,相较于标准的YOLOX网络、SSD算法和DenseBox算法,改进的YOLOX网络在P_(avg)、P和R指标上表现具有明显优势,分别达91.44%,96.36%和98.89%;将改进的YOLOX网络用于变电站巡检机器人自动抄表识别中,实现了智能电表数据的准确识别,且识别的P_(avg)值达90.23%,P值达93.56%,R值达到98.12%。变电站巡检机器人的识别方法可用于自动抄表中,且具有一定的工程应用价值。

In order to improve the precision of automatic meter reading recognition of substation inspection robots,an automatic meter reading recognition method of deep learning was proposed.The YOLOX network was taken as the basic framework,the convolutional attention modules were added at the network channel level and space level,and the Focal-Loss function was used to replace BCE-Loss function,so as to improve the training speed and recognition accuracy of the network.The results showed that compared with the standard YOLOX network,SSD algorithm and DenseBox algorithm,the improved YOLOX network proposed had obvious advantages in P_(avg),P and R,reaching 91.44%,96.36%and 98.89%.The improved YOLOX network was used in the automatic meter reading recognition of substation inspection robots,and the accurate recognition of smart meter data was realized,with P_(avg) value of 90.23%,P value of 93.56%and R value of 98.12%.It can be concluded that the identification method of substation inspection robots can be used for automatic meter reading,which has certain engineering application value.