Federated Learning for Smart and Sustainable Forest Fire Detection in Green Internet of Things

Abstract

With the exponential rise in the adoption of the Internet of Things (IoT), sensors have become an essential part of smart systems, enabling real-time monitoring and control in applications such as energy management, security, and safety. Among these, early fire detection is a critical application to prevent devastating consequences. This paper introduces a novel Federated Learning (FL) framework designed for the rapid detection of forest fires within smart and sustainable environments using the Green Internet of Things (GIoT). The proposed framework integrates distributed learning across multiple edge devices to detect fire incidents without compromising data privacy. It leverages a modified DenseNet121 architecture enhanced with a soft attention mechanism, capable of accurately classifying fire and no-fire scenarios even under challenging weather conditions. The dataset was augmented to simulate fog and haze, ensuring model robustness in real-world environments. The experimental results demonstrate that the proposed system achieves outstanding performance with a training accuracy of 97.8% and a validation accuracy of 97.06%, confirming its effectiveness and scalability in edge-enabled fire detection systems.