基于改进YOLOv4-Tiny的指针仪表自动读取方法
AUTOMATIC READING METHOD OF POINTER METER BASED ON IMPROVED YOLOV4-TINY
邵磊 1陈培栋 1孙文涛 1李超 1刘宏利1
作者信息
- 1. 天津市复杂系统控制理论及应用重点实验室天津理工大学电气电子工程学院 天津 300384
- 折叠
摘要
为了提高对指针仪表识别的准确率与泛化能力,提出一种基于改进YOLO v4-Tiny的指针仪表自动读取方法.该方法一级模型基于YOLOv4-Tiny算法,提取指针仪表表盘区域以及仪表分类,并通过加深主干特征提取网络、增加 FPN(Feature Pyramid Networks)结构与检测头、使用 PreLU(Parametric Rectified Linear Unit)激活函数,提高原YOLOv4-Tiny模型的检测精度.二级读数算法通过(PPHT)累计概率霍夫变换和最小二乘法拟合指针,通过斜率推导仪表读数.实验结果表明,改进方法能够有效、准确地完成对指针仪表的读取,改进的YOLOv4-Tiny模型相比于原模型F1分数提升了 10.65百分点,达到了 93.31％,FPS达到了 192,整体识别准确率达到了99.36％.
Abstract
In order to improve the recognition accuracy and generalization ability of pointer meters,an automatic reading method of pointer meters based on improved YOLOv4-Tiny is proposed.The first level model of this method is based on the YOLOv4-Tiny algorithm to extract the dial area of pointer instrument and instrument classification.The detection accuracy of the original YOLOv4-Tiny model is improved by deepening the main feature extraction network,adding the FPN structure and detection head,and using the PReLU activation function.The second reading algorithm uses PPHT and least square method to fit the pointer,and deduces the meter reading through the slope.The experimental results show that the improved method can read the pointer meter effectively and accurately.Compared with the original network,the improved YOLOv4-Tiny model improves the Fl score by 10.65 percentage points,reaches 93.31％,FPS reaches 192,and the overall recognition accuracy reaches 99.36％.
关键词
目标检测/指针仪表识别/YOLO/PReLU/FPNKey words
Object detection/Pointer meter reading/YOLO/PReLU/FPN引用本文复制引用
基金项目
天津市自然科学基金项目(17JCTPJC53100)
出版年
2024
国家科技期刊平台
NSTL
万方数据