Pointer Meter Reading Method Based on YOLOv8 and Improved LinkNet

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Pointer meter reading can be inefficient and prone to errors.
• Automated reading systems require robust image processing techniques.

Purpose of the Study:

• To develop an accurate and efficient automated pointer meter reading method.
• To improve the performance of segmentation and angle prediction for meter readings.

Main Methods:

• Utilized YOLOv8 for meter dial detection.
• Implemented an improved LinkNet with partial convolution and attention mechanisms (CBAM, AFF).
• Employed a classification and regression approach for accurate rotation angle prediction.

Main Results:

• Achieved a mean absolute error of 0.20 for meter readings.
• Reached a processing frame rate of 15 frames per second.
• Demonstrated enhanced feature extraction and reduced model size.

Conclusions:

• The proposed LinkNet-based method significantly improves pointer meter reading accuracy and efficiency.
• The integration of partial convolution, attention modules, and refined angle prediction contributes to robust performance.