Tiny ML based early wildfire detection

Abstract

Wildfire is an enormous and increasing threat to the global ecosystems, human settlements, and climatic conditions. Traditional methods for wildfire detection such as satellite imaging and ground-based monitoring methods are not providing timely and spatially focused warnings, especially in remote forest regions. This research introduces a novel Tiny Machine Learning (TinyML) based Wireless Sensor Network (WSN) model to meet the critical demand for a strong, real-time, and energy-efficient early wildfire warning system. The proposed TinyML Wildfire Detection System (TWDS) integrates low-power sensors with on-device machine learning to detect environmental abnormalities typical of fire occurrences, such as temperature spikes, humidity declines, and particulate matter or gas detection, such as Total Volatile Organic Compound (TVOC) and Carbon Dioxide (CO₂). Decentralized processing is emphasized in the system design, where sensor nodes process information locally using a lightweight Deep Neural Network (DNN) trained on historical wildfire datasets. This method reduces latency and power consumption by reducing reliance on continuous data streaming to a central server. Hardware implementation of the sensor node uses an Arduino Nano 33 BLE Sense Lite, DHT22 temperature and humidity sensors, and SPG30 gas sensors. The model training was done by using TensorFlow and converting them using TensorFlow Lite and TensorFlow Lite Micro for deployment in TinyML devices. Evaluation metrics such as accuracy, precision, recall, F1-score, and Area Under Curve (AUC) establish the system to be reliable in detecting fire conditions and non-fire conditions. The results validate the system's performance in early wildfire detection with improved responsiveness and scalability, making it a viable candidate for deployment in resource-constrained forest environments. TWDS offers an advanced approach for wildfire prevention through the use of energy efficiency, low-latency, and scalability by combining WSN and TinyML technologies.