摘要
针对指针式仪表读数算法存在计算复杂和对倾斜状态鲁棒性不足的问题,提出一种基于轻量级网络的自动读数方法。采用MobileNetV3模块搭建轻量化主干逐层提取图像特征,以GSConv为基础构建网络颈部,通过自底向上和自顶向下2条路径高效融合特征,设计加权拼接结构使网络分辨有效信息。利用网络检测仪表关键点,采用最小二乘法拟合表盘椭圆,对图像进行透视变换以完成倾斜校正,通过角度法直接计算仪表读数。实验结果显示,所提方法轻量简单,相较传统方法鲁棒性更强,读数准确率高。
To address the issues of computational complexity and insufficient robustness to inclination in pointer meter reading algorithms,an automatic reading method based on a lightweight network was proposed.The lightweight backbone was constructed using the MobileNetV3 module to extract image features layer by layer.The neck was designed based on GSConv,integrating features efficiently through bottom-up and top-down paths.A weighted concatenation structure was designed to enable the network to distinguish useful information.The network detected key points of the meter,and the least squares method was used to fit the dial into an ellipse.Perspective transformation was then applied to complete inclination correction,followed by direct calculation of the meter reading using the angle method.Experimental results demonstrate that the proposed method is lightweight and simple,offering stronger robustness compared to traditional methods and achieving high reading accuracy.
作者
向诗洁
贾振堂
XIANG Shijie;JIA Zhentang(College of Electronics and Information Engineering,Shanghai University of Electric Power)
出处
《仪表技术与传感器》
北大核心
2025年第10期29-35,41,共8页
Instrument Technique and Sensor
基金
国家自然科学基金项目(62105196)。
