Unsupervised Classification and Detection of Forest Fire Images using Convolutional Autoencoder
V Tharun
Department of NWC
Faculty of Engineering and Technology, SRM Institute of Science and Technology, KTR campus, Chennai, India
S Navin Kumar
Department of NWC
Faculty of Engineering and Technology, SRM Institute of Science and Technology, KTR campus, Chennai, India
C Hemanth Varma
Department of NWC
Faculty of Engineering and Technology, SRM Institute of Science and Technology, KTR campus, Chennai, India
cc8658@srmist.edu.in
Dr P Visalakshi
Associate Professor
Department of NWC
Faculty of Engineering and Technology, SRM Institute of Science and Technology, KTR campus, Chennai, India
visalakp@srmist.edu.in
Abstract— Forests are essential ecosystems, providing habitats and resources for diverse species. Existing research has largely relied on supervised methods for classification and detection, but annotating large datasets remains a major challenge. Traditional forest fire detection methods, primarily reliant on sensors such as infrared and smoke detectors, often exhibit limited performance in accurately identifying fire events. In this study, we propose a novel approach leveraging the “ClusterMask Enhancer” to effectively segment and localize forest fires in aerial imagery. This method enables precise identification of fire-affected regions, empowering firefighters to strategize their response and mitigate fire spread more efficiently. This project addresses this gap by employing unsupervised autoencoder-based learning and segmentation, utilizing large sets of unlabeled data to minimize the need for extensive human labeling. The results show that the proposed model outperforms state-of-the-art models, showcasing its superior performance.
Key Words — Unsupervised Learning, Autoencoder, Data Annotation, Segmentation
DOI link – https://doi.org/10.69758/GIMRJ/2503I01S01V13P0009
Download