Publication

Abstract : Wildfires pose a significant threat to ecosystems and communities, requiring efficient detection and control measures. Aerial imagery-based identification using deep learning and machine learning ensures precise and timely interventions to mitigate damage and protect vulnerable areas. This study explores the efficacy of various models, including MobileNet, Vision Transformer (ViT), VGG16, ResNet50, Random Forest, SVM and XGBoost, for classifying aerial wildfire images. Vision Transformer outperforms with 97% accuracy and MobileNet remarkably competes with 0.93 accuracy compared to other deep learning models. Hybrid approach of deep learning and ML models were also experimented. These models ensure robust performance in identifying wildfires a cross diverse environmental conditions and forest types. The study offers an extensive overview of model architectures and their applicability to wildfire detection, creating the door for sophisticated automated systems in practical environment.