A Comprehensive Survey on Intelligent Approaches for Early Forest Fire Detection: Deep Learning, IoT, UAVs, Remote Sensing and Explainable AI

Abstract

Forest fire is one of the most harmful environmental hazards that affects wildlife, human settlements, infrastructure and economic resources, affecting a vast area of the ecosystem. Early detection of fire and smoke is critical to limit response time and to minimize the growth of small fires into large disaster fires. Human observation, fixed sensing infrastructure, and satellite-sensed forest fire monitoring techniques may have drawbacks such as detection delays, limited spatial extent, environmental variability, communication limitations and false alarm rates. With recent developments in Artificial Intelligence (AI), Internet of Things (IoT), deep learning and unmanned aerial vehicles (UAVs), the prospects for intelligent and automated fire detection in forests are new and promising. In this survey, a thorough overview of the latest intelligent solutions for early forest-fire detection is provided, focusing on fire and smoke recognition in images, machine learning and deep learning models, object detection and segmentation, UAV-based surveillance, the IoT sensor network, satellite remote sensing, multimodal data fusion, and edge intelligence, as well as Transformer-based architectures and Explainable Artificial Intelligence (XAI). Existing approaches are clustered into a coherent taxonomy based on the sensing platform, data modality, computational intelligence, detection task and deployment environment. A comparative analysis is provided to consider the strengths and weaknesses, detection capabilities, computational requirements and practical usefulness of representative approaches. Furthermore, a number of forest-fire data sets and assessment indicators are analysed to assess the applicability of the current early-warning system evaluation tools to real-world situations.The survey also covers key challenges such as visibility of smoke in small and early stages, environmental interference, class imbalance, few works on cross domain generalization, detection of false positive, computational limitations on edge and UAV platforms, communication reliability, energy consumption, and lack of standardized real world evaluation protocols. Finally, emerging research directions related to multimodal sensor fusion, lightweight Transformer models, edge AI, autonomous UAV systems, federated learning, explainable fire detection and integrated satellite-UAV-IoT architectures are discussed. This survey aims to serve as a single source to enable the construction of reliable, low delay, scalable, energy-efficient and trustworthy forest-fire early-warning systems.