| 基于卷积神经网络的压力仪表OCR系统 |
| OCR system for pressure instruments based on convolutional neural network |
| |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 中文摘要: |
| 为了解决压力仪表传统人工抄表效率低、易出错且存在安全风险以及基于传感器和三维视觉的自动抄表技术适应性不足的问题，结合计算机视觉与人工智能技术，构建了融合数据采集、实时监控和数据分析功能的计量系统。通过改进快速区域卷积神经网络（Fast Region?Convolutional Neural Network， Fast R?CNN）算法，引入数据增强和轻量化特征提取网络，提高复杂环境下的仪表定位准确度；在DeepLabv3+模型中融合通道注意力机制与空间注意力机制，结合混合损失函数，提升字符分割效率。实验表明：改进算法在复杂工业环境中的仪表表盘定位平均准确度达84%，字符分割平均交并比为78.6%，单次计量耗时较人工抄表缩短85%，验证了系统的高效性与强适应性。本研究为工业设备智能监测提供了可扩展的技术框架，对推动计量数字化、智能化具有实践价值。 |
| 英文摘要: |
| To address the inefficiency, error?proneness, and safety risks associated with traditional manual meter reading for pressure instruments, as well as the limited adaptability of automated meter?reading technologies based on sensors and 3D vision, this study integrates computer vision and artificial intelligence technologies to develop a metering system that combines data acquisition, real?time monitoring, and data analysis. By improving the fast region?convolutional neural network（Fast R?CNN) algorithm through data augmentation and a lightweight feature extraction network, the system optimizes instrument positioning accuracy in complex environments. Additionally, the DeepLabv3+ model is enhanced by incorporating channel attention and spatial attention mechanisms, along with a hybrid loss function, to improve character segmentation efficiency. Experimental results demonstrate that the improved algorithm achieves an average positioning accuracy of 84% for instrument dial positioning and a mean Intersection over union of 78.6% for character segmentation in challenging industrial environments. Furthermore, the system reduces the time required for a single measurement by 85% compared to manual reading, confirming its high efficiency and strong adaptability. This research provides a scalable technical framework for intelligent monitoring of industrial equipment, offering the practical value for advancing digital and intelligent metering. |
|
| 中文关键词: 光学字符识别 设备现场计量 深度学习 仪表定位 字符分割 |
| 英文关键词:OCR equipment on⁃site metering deep learning instrument positioning character segmentation |
| 基金项目: |
| DOI：10.11823/j.issn.1674-5795.2025.03.10 |
| 引用本文：王晶星, 陈诗琳, 李一鸣, 王丽, 石伟, 刘芳.基于卷积神经网络的压力仪表OCR系统[J].计测技术,2025,45(3):111~122. |
| 关闭 |
|
|
|
