Metrology & Measurement Technology, Volume. 45, Issue 3, 111(2025)

OCR system for pressure instruments based on convolutional neural network

Jingxing WANG, Shilin CHEN, Yiming LI, Li WANG, Wei SHI, and Fang LIU
Author Affiliations
  • AVIC Changcheng Institute of Metrology & Measurement,Beijing100095,China
  • show less
    Figures & Tables(19)
    Flowchart of OCR technology
    The principle of object detection models based on CNN
    Schematic diagram of fast R⁃CNN
    Schematic diagram of MobileNetV3
    Comparison of algorithm effects in complex environments
    Schematic Diagram of SE Module
    Calculation process of the SE module
    Workflow diagram of the SAM module
    Comparison between Predicted Segmentation and Ground Truth Segmentation Effects
    Diagram of hardware selection testing device
    User interaction interface
    Effect of image pre⁃processing
    Object detection effect
    Segmentation effect of the improved model on the dial
    Pointer contour diagram
    • Table 1. Effect comparison of instrument location algorithms

      View table
      View in Article

      Table 1. Effect comparison of instrument location algorithms

      模型AP / %RecallFLOPs / GParams / M
      原始Fast R⁃CNN750.70180138
      YOLOv8800.8262.611.1
      EfficientDet770.7315.38.5
      改进后的Fast R⁃CNN840.809045
    • Table 2. Effect comparison of element segmentation algorithms

      View table
      View in Article

      Table 2. Effect comparison of element segmentation algorithms

      模型PA / %mIoU / %F1分数Params / M
      原始DeepLabv3+83.571.20.764.5
      引入注意力机制的DeepLabv3+86.274.50.795.2
      改进损失函数的DeepLabv3+87.175.30.804.5
      引入注意力机制和改进损失函数的DeepLab3+89.378.60.835.2
    • Table 3. Hardware selection testing device

      View table
      View in Article

      Table 3. Hardware selection testing device

      硬件设备关键参数
      工业相机镜头视野范围: 67 mm × 100 mm;特征分辨力: 15 μm × 15 μm;最小特征的像素点数: FpXFpY为3 pixels;像元尺寸: 2.4 μm × 2.4 μm;帧速率达5 fps;分辨力达2 000万像素。
      计算设备英特尔酷睿Ultra 5处理器125 H
      远心镜头

      像圈直径:11.0 mm;

      工作距离:181.8 mm;

      像侧分辨力:5 μm。

      LED环形光源光衰预测:35 000 h亮度仍然维持原来的75%;使用寿命:约50 000 h。
    • Table 4. System experimental results

      View table
      View in Article

      Table 4. System experimental results

      准确率召回率F1分数
      94.595.896.1
    Tools

    Get Citation

    Copy Citation Text

    Jingxing WANG, Shilin CHEN, Yiming LI, Li WANG, Wei SHI, Fang LIU. OCR system for pressure instruments based on convolutional neural network[J]. Metrology & Measurement Technology, 2025, 45(3): 111

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Metrology,Measurement and Calibration

    Received: Feb. 14, 2025

    Accepted: --

    Published Online: Sep. 3, 2025

    The Author Email:

    DOI:10.11823/j.issn.1674-5795.2025.03.10

    Topics