| Abstract | Despite the effectiveness of smart gas meters, many older, non-smart meters are still in use, creating significant challenges and costs associated with upgrading to advanced systems. To address this, developing artificial intelligence-powered frameworks to monitor natural gas consumption accurately from these non-smart meters is essential. This study introduces a deep learning (DL)-based framework designed to automate gas consumption readings from traditional, non-smart meters. Utilizing real-time image processing and innovative data augmentation techniques, the proposed system significantly improves measurement precision from 1 cubic meter to 0.001 cubic meters. This approach provides an efficient bridge between traditional and modern energy monitoring systems without necessitating the costly replacement of existing meters. The framework's development and evaluation leverage the NRC-GAMMA dataset, a meticulously gathered and labelled collection of gas meter images. An extensive quality control process was conducted on the dataset, which included several rounds of annotation and verification, to guarantee high accuracy and reliability. The proposed DL model's robust performance across various environmental conditions is enabled by advanced data augmentation strategies and DL algorithms, making it a versatile solution for a broader range of automated energy meter readings, contributing significantly to the efficiency and accuracy of energy management systems. |
|---|
