Research on water meter reading recognition based on deep learning.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Water Management Systems

Background:

• Traditional manual water meter reading is inefficient, costly, and prone to errors.
• Environmental factors like dial contamination hinder accurate manual readings.
• Lack of verifiability in manual readings poses auditing challenges.

Purpose of the Study:

• To develop and evaluate a deep learning method for automatic water meter reading.
• To improve the efficiency and accuracy of water meter data collection.
• To reduce labor costs and enhance the auditing process for water utilities.

Main Methods:

• Exploration of deep learning models including Faster R-CNN, SSD, and YOLOv3.
• Development of two datasets: original water meter images and cropped readings with bounding boxes.
• Comparative analysis of model performance on the prepared datasets.

Main Results:

• The YOLOv3 model demonstrated superior performance in the second proposed solution.
• Achieved a high accuracy rate of 90.61% for water meter reading recognition.
• The deep learning approach significantly outperforms manual methods.

Conclusions:

• Deep learning, specifically YOLOv3, offers a viable solution for automated water meter reading.
• This technology can substantially increase operational efficiency and reduce costs for water departments.
• Automated readings provide a verifiable and accurate data source for auditing.