FOREST WILD FIRE DETECTION USING DEEP LEARNING TECHNIQUES CNN AND TRANSFER LEARNING

Abstract

Forest wildfires are one of the most destructive natural disasters, causing irreversible damage to biodiversity, property, and human lives. Rapid and accurate detection of wildfire incidents is essential to minimize their impact and assist emergency response systems. This paper presents a deep learning-based image classification framework for early detection of wildfires using the publicly available "The Wildfire Dataset" from Kaggle. The dataset comprises a balanced collection of forest scenes labeled as "fire" and "non-fire," making it suitable for training and evaluating classification models. A Convolutional Neural Network (CNN) architecture was designed and trained on this dataset to identify the presence of wildfires from static forest images. The proposed model achieved a final training accuracy of 99.73% and a validation accuracy of 92.60%, indicating high learning capability and strong generalization performance. These results highlight the model’s effectiveness in recognizing fire patterns from visual features, even in complex backgrounds. The study reinforces the feasibility of using deep learning for wildfire detection and paves the way for deploying intelligent surveillance systems in forest regions to aid in early intervention and disaster risk reduction.