IoT-Based Flash Flood Detection and Alert Using TensorFlow

It is important to have a real-time flash flood detection system to inform the public for them to take appropriate action. The current method of authorities using mainstream media such as newspaper, radio, TV, or public announcement is too slow to provide the local population ahead starts to prepare for coming flash flood. Several other early flood warning systems have been proposed but the system is already outdated and did not alert the user in real-time. Therefore, this paper proposes an IoT-based flash flood detection and alert using TensorFlow. The flash flood is detected by using machine learning technique and an alert will be sent to the user using Telegram. The detection did not rely on a conventional water sensor to detect floods, instead, it uses a video camera to monitor the water level. Moreover, the system was implemented in low-powered raspberry pi which can be deployed to many floods prone areas. Based on the test result, the system can differentiate between normal and flash flood water levels and alert users via Telegram Channel. The test results also show that using TensorFlow Lite with SSD-MobileNet-v2-Quantized model in IoT environment has the highest performance.