Low-Power Edge AI System for Long-Term Autonomous Mechanical Water Meter Reading on Embedded Devices

¹ Faculty of Electronics Engineering 1 & EDA Lab, Posts and Telecommunications Institute of Technology, Vietnam

^* e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^*** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^** e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^† e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
^‡ e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Rapid urbanization and rising water consumption require more efficient water-use monitoring. Traditional mechanical water meters rely on manual readings, which are labor-intensive and error-prone. This paper proposes an Edge AI system for mechanical water meter reading using the ESP32-CAM platform. The system captures meter images, performs lightweight preprocessing, and executes on-device inference using a lightweight ResNet-based CNN. Experimental results show high performance, achieving a test accuracy of approximately 97.2%, with Precision, Recall, and F1-score values around 0.98 across 20 classes. The system is resource-efficient, using 207 KB RAM and 689.5 KB FLASH, with a total execution time of 194 ms per cycle. Fully local processing reduces latency, deployment cost, and cloud dependency, enabling real-time operation on a low-cost embedded platform. A low-power evaluation using a Nordic PPK2 and a LoRa transmission module confirms ultra-low energy consumption, with most of the system operating in deep-sleep mode. A daily transmission scenario further demonstrates stable energy use and supports long-term deployment.