<![CDATA[While accurate water management is crucial for public utilities, manual meter reading remains inefficient due to recording errors and high operational costs. This study proposes an automatic water meter reading detection system based on YOLOv11 with a variety of image augmentation techniques integrated into Telegram. The dataset was obtained through a combination of ESP32-CAM image captures and online sources totaling 1,207 images, followed by labeling on Roboflow and augmentation in the form of flipping, rotation, saturation, and noise. The YOLOv11 model was trained on Google Colab using an A100 GPU with 100 epochs. Performance evaluation was conducted using Precision, Recall, mAP50, and mAP50-95 metrics. The results showed that the application of augmentation significantly improved model performance, with Precision of 95.9%, Recall of 98.2%, and mAP50 of 97.4%. The combination of four augmentation techniques produced the highest mAP50-95 value of 0.575, indicating the model's robustness against variations in field conditions. The system is also capable of automatically sending detection result notifications via Telegram, enabling its implementation for real-time remote monitoring. Compared to previous studies with YOLOv4 and YOLOv5, this approach proved to be superior in terms of accuracy and efficiency. These findings indicate that the integration of YOLOv11 with image augmentation techniques and IoT support has the potential to be an optimal solution in the modernization of digital water meter reading.]]>
Copyrights © 2026
