摘要
当前的电表读取算法存在精度低、读取速度慢的问题。为了能够更准确地读取电表数值,文章提出一种基于注意力机制与YOLOv3的电表自动读数算法,该算法首先对采集的电表图像进行去噪、增强等预处理;其次通过在YOLOv3网络层中添加高效金字塔分割注意力(Efficient Pyramid Split Attention,EPSA)模块对YOLOv3进行改进;利用改进后的YOLOv3对预处理后的图像进行训练,考虑到表盘刻度目标通常较小,训练分为2个阶段:先检测图像中的表盘;然后在所检测出的表盘图像部分检测指针和刻度值,实现电表数值的读取。实验结果验证了所提算法的性能较好,能够准确地读取电表数值。
The current power meter reading algorithm has the problems of low precision and slow reading speed.In order to read the value of power meter more accurately,we propose an automatic reading algorithm of power meter based on the combination of YOLOv3 and attention mechanism.Firstly,the image of power meter is preprocessed by denoising and enhancing.Secondly,the EPSA attention mechanism module is added to the network layer of YOLOv3 to improve it;The improved YOLOv3 is used to train the preprocessed image.Considering that the dial scale target is usually small,the training is divided into two stages including checking the clock dial and detecting the pointer and scale value in the detected clock dial,so as to read the value of the power meter.Experimental results has verified that our algorithm has good performance and can accurately read the value of power meter.
作者
张俊
杨光
胡东升
胡志凌
王南勤
ZHANG Jun;YANG Guang;HU Dongsheng;HU Zhiling;WANG Nanqin(Nanjing Power Supply Branch,State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 210000,China;Anhui NARI Jiyuan Electric Power Grid Tech.Co.,Ltd.,Hefei 230088,China)
出处
《电力信息与通信技术》
2021年第12期82-87,共6页
Electric Power Information and Communication Technology
基金
国家自然科学基金项目(61772032)。
