摘要
随着深度学习算法的发展,指针式仪表识别方法也从原来的模板匹配算法发展到了如今通过深度学习算法进行识别,但是目前常用的深度学习读表方法多数基于YOLO,这些方法的鲁棒性相对较差,难以满足一个制造车间下不同种类仪表的识别。基于对图像分割算法的研究提出了一种改进的U-Net算法,实现对表盘的刻度及指针进行分割,并设计出了一套针对图像分割的读数方法,最后通过OCR技术中的文本区域检测和文本识别算法对表盘关键信息进行提取来实现指针式仪表的识别。使用该算法得到的指针式仪表读数的准确率高达99%。实验结果表面,该算法的稳定性及鲁棒性表现极佳,准确率也优于其他深度学习算法。
With the advancement of deep learning algorithms,the recognition methods for pointer-based instruments have evolved from traditional template matching algorithms to current deep learning-based approaches.However,most commonly used deep learning methods for reading instruments rely on YOLO,which exhibit relatively poor robustness and struggle to meet the requirements of recognizing different types of instruments in a manufacturing workshop.This paper proposes an improved U-Net algorithm for segmenting the scales and pointers of instrument dials.A reading method for image segmentation is designed,and the key information of the instrument dial is extracted through text region detection and text recognition algorithms in OCR technology to achieve pointer-based instrument recognition.The proposed algorithm achieves a reading accuracy of up to 99%for pointer-based instruments.Experimental results demonstrate excellent stability and robustness of the algorithm,with higher accuracy compared to other deep learning methods.
作者
高腾
王占举
王楠
卜繁洋
GAO Teng;WANG Zhanju;WANG Nan;BU Fanyang(School of Mechanical Engineering and Automation,Dalian Polytechnic University,Dalian 116034,China)
出处
《组合机床与自动化加工技术》
北大核心
2024年第12期12-17,共6页
Modular Machine Tool & Automatic Manufacturing Technique
基金
辽宁省教育厅高校基本科研项目(JYTZD2023023)。
