AI-Powered Forest Fire Detection and Early Risk Prediction System
CHANDANA K V, NIMITHA P, SHREYASHWINI G, PRIYADARSHINI R, Mrs. MANASA B S
CHANDANA K V CSE(DS) & SJBIT
NIMITHA P CSE(DS) & SJBIT
SHREYASHWINI G CSE(DS ) & SJBIT
PRIYADARSHINI R CSE(DS) & SJBIT
Mrs. MANASA B S CSE(DS) & SJBIT
Abstract - Forest fires pose a critical threat to biodiversity, environmental stability, and human safety, while their incidents are gradually on the rise owing to climatic changes, land-use changes, and rising temperatures. Most conventional monitoring techniques- satellite imaging, manual patrolling, threshold-based sensor systems-suffer from delayed detection, low spatial resolution, and high false alarm rates [3,4,16,26]. More recently, because of developments in AI, ML, DL, IoT, and remote sensing, significant strides have been made in wildfire prediction, early detection, and automated response [1,2,5,6,7,9,11,17,20,23,27,30]. Motivated by such developments, this work reports an integrated AI- powered forest fire detection and early risk prediction system, integrating CNN-based visual fire detection, random forest-based environmental risk classification, and IoT-driven sensor monitoring, as proposed in the project document.
The proposed system uses a transfer-learned CNN model to identify flames and smoke in real-time through webcam feeds, similar to the works on DL in [5], [6], [17], and [23], and optimized architectures like FireXNet and GoogLeNet-based detectors in [15] and [25]. Meanwhile, a Random Forest classifier will predict fire-risk levels of low, medium, and high from temperature, humidity, and air quality data captured using DHT11 and MQ-135 sensors. This will align with ML-based prediction studies found in [4], [8], [10], [18], and [19] and environmental analytics based on multisensor systems in [12], [14], [20], and [28]. Features extracted from both models will be combined into a decision-fusion mechanism that will trigger proactive alerts through LED and buzzer hardware, enabled by NodeMCU ESP8266, conforming to IoT-based early warning systems in [1], [9], [12], and [21] and secure forest monitoring frameworks in [22]. It ensures high accuracy and speed in response time, and robustness of operations, thus supporting the efficiency of the proposed multimodal AI-IoT integration for real-time wildfire monitoring. These features address the critical gaps in existing detection technologies and fall within various current global research trends in satellite-aided detection [3, 16, 26], UAV-supported surveillance [27, 29], and crowdsourced fire reporting [13].
Key Words: Forest fire detection; Early fire prediction; Convolutional neural network; Random forest; Internet of Things; DHT11; MQ135; NodeMCU ESP8266; Machine Learning; Deep Learning; Remote sensing; Wildfire monitoring; Fire risk classification; Sensor networks; Decision fusion.
